Repository navigation
Document retrieval-first local-model choices and supersede foundation sources - #810
Draft
seathatflowsinourveins wants to merge 10 commits into
Draft
seathatflowsinourveins wants to merge 10 commits into
seathatflowsinourveins wants to merge 10 commits into
Conversation
seathatflowsinourveins
added
the
lane:foundation
seathatflowsinourveins
force-pushed
the
codex/ns2604-retrieval-first-local-models-20261006
branch
from
October 6, 2026 22:54
fe57e55 to
7750a37
Compare
5 of 7 tasks
Select Nemotron-3-Embed-8B-BF16 at d1f2f25730bbd775b99b29185134bc86653bf2d1 with 4096 dimensions; retain query/passage prefixes and watcher policy. Keep NativeStack2604 endpoint examples and lexical qmd cutover distinct from owner-applied deployment and acceptance. SOTA sources: nvidia/Nemotron-3-Embed-8B-BF16@d1f2f25730bbd775b99b29185134bc86653bf2d1:README.md:112,148,501; giancarloerra/SocratiCode@v1.15.0:src/services/embedding-config.ts:349,385 and src/constants.ts:72; akitaonrails/ai-memory@v2.5.2:docs/llm-providers.md:374-379; tobi/qmd@v2.8.3:README.md:1151; vllm-project/vllm@v0.31.0:vllm/engine/arg_utils.py:996 and vllm/config/pooler.py:50-63. Static JSON/TOML and command-argument parsing and git diff --check passed. No generators, tests, installs, model calls or host changes during the paper window; full integration validation is deferred to the parent after 2026-10-07T00:10:00Z.
Preserve historical settlements and project the dated current source ruling; update canonical source carriers without model execution. Sources: native-agent-stack@0d5e6506434fab598dee861c749a22e628beb75a:assemble_manifest.py:97-101; NVIDIA Nemotron8B@d1f2f257:README.md; tobi/qmd@v2.8.3:src/llm.ts. Generators and tests are held until the paper window ends; remote CI is the current check route.
Shared stack and evidence registry only, per docs/lanes.md:96-128. Receipt class is upstream provenance/source review; model runs and native validation remain deferred under the paper window. Main registry restored and owned source hashes re-registered.
seathatflowsinourveins
force-pushed
the
codex/ns2604-retrieval-first-local-models-20261006
branch
from
October 7, 2026 00:28
7750a37 to
aa17bb9
Compare
7 tasks done
seathatflowsinourveins
added a commit
that referenced
this pull request
Oct 7, 2026
### Scope Repair the native install-plan checks that lost prompts, ended before delivering results, inherited incompatible environments, or accepted incomplete output. Fresh Claude checks now use task-first native arguments, the shared foreground instruction file and session lock, bounded execution, and independent completion/artifact gates. The plan also carries the gate-1 environment, messaging, conformance, observability and canonical gateway repairs. - Base: `3d7d4a4b2640c8583d8855670111ef5d023d3e03` (final landing cue after #758; main's registry and stack retained). - Review head: `979291ded5f39ddf8cdcba4ffb29df4656b407d8`; final table commit `c6f7ca4`, registry commit `979291d`, exactly two commits after reviewed `8484ddbc`. - Lane: `lane:foundation`; draft until the command center reads this exact head. - Owned paths: install-plan scripts/configuration/checker/tests; client-config drift fixes; source decisions/receipts and maintained generated outputs. - Local-model finding 2 is a separately preserved source checkpoint for #810. It is deferred from this critical head under START-PATH/FULL-RESOLUTION; #723 does not wait for that PR. - AgentsView's operational plan uses 0.44.0, its verified launcher and known-alias migration. #807 carries the shared repository pin. Frozen 0.43.0 observations remain historical; CODE-MODE attribution remains incomplete. ## SOTA sources - GO-PUSH-723 final microfix follows native-stack@8484ddbcf138cc96c853fb2720fe016daee6b035:tests/test_adoption_docs_consistency.py:815-829 (GFM0.29-gfm §4.10 contiguous-table contract) and scripts/host_receipts.py:register_file. It removes one empty line without changing any row, then updates only that file's sha256 and byte count in the last commit. - J723/J723b/J723c delta: Node's [WHATWG URL hostname](https://nodejs.org/api/url.html#urlhostname) and [port](https://nodejs.org/api/url.html#urlport); systemd@v259 [microsecond show formatting](https://github.com/systemd/systemd/blob/v259/src/systemctl/systemctl-show.c#L1120); uutils/coreutils@0.10.0 [date](https://github.com/uutils/coreutils/blob/0.10.0/src/uu/date/src/date.rs#L596) and [stat](https://github.com/uutils/coreutils/blob/0.10.0/src/uu/stat/src/stat.rs#L1117); CPython3.13 [recursive paths](https://docs.python.org/3.13/library/pathlib.html#pathlib.Path.rglob). The existing handbook receipt guard remains exact; prior output pairs and observed validation history are retained. Both research clone tag SHAs are enforced. SOURCES.md gives the remaining repository-native contract locators. - [Claude native headless CLI](https://code.claude.com/docs/en/headless), [system-prompt flags](https://code.claude.com/docs/en/cli-reference#system-prompt-flags) and [environment controls](https://code.claude.com/docs/en/env-vars), checked against installed 2.1.289. The retained byte proofs and all per-row pins are in `evidence/artifacts/new-wsl-install-plan-20261002/SOURCES.md`. - [openai/codex@a956835d:review.rs:105](https://github.com/openai/codex/blob/a956835d020762cb2b570053af06f643a11c0ecc/codex-rs/core/src/tasks/review.rs#L105): bind the actual native reviewer child, model and effort; native exec_command/write_stdin completion is distinct from an MCP transport timeout. - [aannoo/hcom@2c5f343b:tests/support/mod.rs:949](https://github.com/aannoo/hcom/blob/2c5f343b2f9ec4bf2acf49c0431860e7c2ae578b/tests/support/mod.rs#L949): first-marker identity capture. Landed #771 supplies the relaxed operational adapter and native rules check; the duplicate A50 historical-only policy was dropped. - [util-linux@v2.41.3:flock.1.adoc:80](https://github.com/util-linux/util-linux/blob/v2.41.3/sys-utils/flock.1.adoc#L80): lock acquisition bounds; installed timeout is uutils coreutils0.10.0, with native timeout source cited in SOURCES. Queue is at most3600s and Claude execution at most3300s, separately. - [DeerFlow@v2.1.0 config](https://github.com/bytedance/deer-flow/blob/345f08be00c8a9495079b732a39b46aa9af1584e/config.example.yaml): native headless/keyless provider and model metadata. Canonical Sol-max suffix precedence follows [the #744 decision](https://github.com/seathatflowsinourveins/native-agent-stack/blob/0fb32ee583589f1a0809b20d18dff40ce2f65a1c/docs/decisions/2026-10-05-omniroute-gateway-composition.md#L37); `supports_reasoning_effort=false` is a derived configuration choice, not a statement in #744 or wire-effort proof. - [MCP conformance@c321dd3 runner](https://github.com/modelcontextprotocol/conformance/tree/c321dd3/examples/servers/typescript): unchanged upstream servers/CLI inside the cited private user/PID/network namespace, with owned group cleanup. External URL acceptance stays needs_owner78 until an in-namespace adapter is supplied. - [Playwright system dependencies](https://playwright.dev/docs/browsers#install-system-dependencies), Inspect AI0.3.273, OpenHands SDK1.50.1 and SRT0.0.78 pinned sources are recorded per row in SOURCES. The worker retains production's Unix-socket block and the upstream subprocess backend; no test-only relaxation. - [AgentsView v0.44.0](https://github.com/kenn-io/agentsview/releases/tag/v0.44.0), commit413a87f7bfbd67b2815b1119ac51abc1efbeeaba, official Linux archive SHA037ea7a46d52e06b20363b4aa7cd7f28e32f31d8215803d6e9a0c96bac5818e3. Native B2 results and failed earlier attempts remain separate from frozen0.43 evidence. - Native repository builders: `assemble_manifest.py:876-896`, `scripts/build_new_wsl_handbook.py`, `host_receipts.register_file`, `component_matrix.py` and `new_host_grand_list.py`; `docs/lanes.md:96-128` governs the last generated/registry commit. Decision corrections are dated addenda; original observed values are preserved. ### Required verdict changes | Item | Resulting source | | --- | --- | | P1 hcom posture | Rebased on #771; dropped15f24fe1 and its generated companion. Relaxed adapter remains operational. Independent naming fix retained. | | P1 canonical route | Exactly cx/codex gpt-6.1-sol-max accepted with all provider, suffix, status, endpoint and join negatives retained. | | P2 timeouts/background | Foreground1560000ms; native background-disable env; flock wait outside execution timeout;48 turns/fixture/schema retained. | | P2 skill config | Native installed default config; concrete triggers and foreground wait; custody proposal outside the operational skill. | | P2 conformance URL | Unsupplied target stays needs_user78; external URL without namespace startup adapter is needs_owner78. No host-network fallback. | | P2 gateway custody | Unknown operative assets still fail closed; regular example files retained. Explicit owner backup/staging/reload/restart/read-back/rollback runbook in the dated addendum. | | P2 citations/evidence | FAST suffix is high; #744 only supplies suffix precedence; removal condition is issue8939/PR8940 and its regression test; cleanup is helper-derived local integration. | | Rebase/hot files | Main's generated inputs/registry taken; maintained outputs regenerated; all shared generated/registry edits last. | ### Evidence-class table | Claim | Class | Evidence | | --- | --- | --- | | Source agreement and artifact integrity | local_integration | Plan/client-map/handbook checks and maintained generators; final validator result below | | Completion, custody, endpoint and alias controls | synthetic / local_integration | Four touched fixture modules,379 tests,3 existing skips | | Earlier native attempts | Their original classes | Existing returned streams/receipts preserved; later success does not replace a failure | | Model routes and metadata | source_review / local_integration replay | Exact aliases and joined records; where spelling changed and delivered wire effort remain unproved | The new sanitized source receipt is `evidence/artifacts/new-wsl-install-plan-20261002/fixwave-20261006-verdict-rebase-receipt.json`. It does not declare a new native-ready slot or upstream-suite pass. ### Local commands run Final GO-PUSH-723 change: exactly one empty line removed at harness-defaults.md:122, plus that file's registry sha256/bytes. Two commits, same base3d7d4a4b; no rebase or other source change. `nice -n 19 ionice -c3 python3 -m unittest tests.test_adoption_docs_consistency` passed39tests/1existing skip in17.985s; final validator69components/10408hashes/4profiles/224receipts0; diff0. A structural comparison proves the registry differs in exactly one file entry. The command center's ACK stands under the explicit range-diff/green-CI condition; neither CI nor landing is claimed by this lane. Final J723d delta: conformance installs native EXIT/INT/TERM cleanup before the path-length assertion and removes only its frozen owned TMPDIR after process teardown. The early-failure control preserves unrelated cache bytes and the evidence directory; existing success/failure/cancellation controls now also require the allocation gone. Nine focused lifecycle cases passed. After the final landing-cue rebase onto #758, all seven full touched modules passed **589 tests / 3 existing skips in 163.180s**. Final plan0 remains84rows/192commands/113acceptance; F9readonly0 with the same two warnings; native builder/check0; diff0; validator0:69components/10408hashes/4profiles/224receipts. Current handbook pair is c8ecbb5d/d7cc2ccb, with all predecessor output bindings/observations retained. Native [Bash trap documentation](https://www.gnu.org/software/bash/manual/html_node/Bourne-Shell-Builtins.html#index-trap) and the existing identity-checked cleanup seam govern the bounded fix. No other functional delta was added. Completeness critic found no material blocker, source review only. CI/CodeQL and destination apply remain owner gates; the PR stays a draft under the granted landing hold. Earlier published J723/J723b/J723c delta: all seven full touched modules passed **581 tests / 3 existing skips in 166.987s**, under the CI-pinned Python3.13 environment, explicit private-venv PATH and umask022, at nice19/ionice3. The run includes all four J723c failing modules and the earlier client-config/gateway/conformance modules. Earlier delta attempts remain distinct:99tests/2fixture failures, then focused12pass;468tests/1unsettled synthetic login-path failure, whose unchanged isolated retry passed. No assertion or skip was relaxed. Focused CI4 controls17pass. Then-current validator0:69components/10401hashes/4profiles/224receipts. These remain local integration, not destination acceptance. The earlier #809-only rebase had proved unchanged foundation bytes. The final #758 landing-cue rebase adds its time-zone/client-config sources, so the full seven-module suite was rerun, with main's source/registry/stack and regenerated publication bindings retained. Review-thread read found no chatgpt-codex-connector threads; the CodeQL thread remains for its scan result. ```bash nice -n 19 ionice -c3 python3 -B -m unittest -v tests.test_wsl_new_distro_recipe tests.test_agentsview_qualification tests.test_new_wsl_definitive_defaults tests.test_new_wsl_handbook tests.test_new_wsl_client_config tests.test_new_wsl_gateway_composition tests.test_new_wsl_mcp_conformance_lifecycle nice -n 19 ionice -c3 python3 scripts/validate.py ``` The owner-pending Claude route now produces **needs_owner/78 before loading Promptfoo's configuration**, rather than a failed provider evaluation; malformed supplied routes still fail. Parsed URL host and port must match the plan-owned127.0.0.1:21128 exactly. Both gateway health stages require the keyed microsecond UTC active-start timestamp to be strictly later than both rendered unit files, preserving existing identity and completion gates. CI enumerates nested config files, keeps an exact Inspector script set, uses native Git/Python/env without a forced RTK runtime dependency, and classifies old AgentsView aliases as predecessor fixtures. Source completeness critic found no material blocker; its review ran no tests or host calls. J723b's native-messaging smoke observation from the dropped A50 draft is retained privately, byte-identical, mode0600, SHA256 **a14356efc2ba3e6c7d44b2d9818538a824b9429fc5d9eaff929ee56e38efee83**. That dropped draft observation is superseded by #771; no public A50 amendment, retired posture or enforcement test is restored. Earlier source-head validation follows: ```text python3 -B evidence/artifacts/new-wsl-install-plan-20261002/check_plan.py rc0:84 rows,192 commands,113 acceptance entries; earlier kind-label mismatch retained. python3 tools/adoption/new_wsl_client_config.py --check rc0; two pre-existing Inspector ownership warnings, addressed separately by C11; no apply. python3 scripts/build_new_wsl_handbook.py --check rc0. python3 -B -m unittest -v tests.test_new_wsl_client_config tests.test_new_wsl_definitive_defaults tests.test_new_wsl_gateway_composition tests.test_new_wsl_mcp_conformance_lifecycle First attempt rc1:379tests/21errors/3skips; undeclared fixture PyYAML import. Maintained standard-library fixture repair: targeted21cases rc0. Full retry rc0:379tests,3skips,120.182s; no production/helper/dependency or assertion weakening. git diff --check rc0. nice -n19 python3 scripts/validate.py rc0:69 components,10385 hashed files,4 profiles,215 receipts. The earlier historical-inventory non-live limitation failure is retained; payload and manifest were clarified together. ``` ### Decision record Current instructions and the complete gateway custody runbook are in the dated addendum to `docs/decisions/2026-10-06-native-plan-gate1-repairs.md`. Its original decision text remains unchanged. The earlier acceptance decision also receives an appended FAST-effort correction. Original historical observations are never rewritten. ### After landing: co-op apply and read-back The co-op is the sole plan writer. Apply the approved main commit at its booked quiet point; retain the previous gateway prefix and per-asset backup packet. First follow the addendum's gateway custody/reload pre-gates. No lane applies a host change or widens AF_UNIX. ```bash plan=evidence/artifacts/new-wsl-install-plan-20261002 bash "$plan/install.sh" --only agent-messaging bash "$plan/accept.sh" --only agent-messaging --stage post_install </dev/null bash "$plan/install.sh" --only inspect-ai bash "$plan/install.sh" --only trajectory-analysis bash "$plan/accept.sh" --only trajectory-analysis --stage post_install </dev/null # Gateway: execute the owner-custody addendum first; preserve rollback prefix/packet. bash "$plan/install.sh" --only gpt-gateway bash "$plan/accept.sh" --only gpt-gateway --stage post_install </dev/null bash "$plan/accept.sh" --only gpt-gateway --stage service_health </dev/null bash "$plan/accept.sh" --only gpt-gateway --stage after_sign_in </dev/null bash "$plan/install.sh" --only research-harnesses bash "$plan/install.sh" --only skill-vetting bash "$plan/install.sh" --only mcp-protocol-conformance bash "$plan/install.sh" --only session-analytics ``` The hcom apply uses the relaxed mapped adapter and native smoke; it has no custom quiet/deny posture. Gateway owner custody, daemon-reload, unit restart and packet-based rollback are fully specified in the addendum. Existing environment credentials are left by pointer and are never copied into the packet. Archive-version rollback for AgentsView additionally requires its consistent pre-migration archive; preserve the existing backup, whose restoration has not been qualified. For each changed row, retain the install/read-back rc, then one smoke in each client that uses it and subsequent organic counters. Use the canonical shared flock for each exact fresh Claude invocation. Existing repaired-stage commands and returned-artifact gates remain in accept.sh; targetless conformance78 is not a protocol pass. No new measurement campaign or A/B gates this apply. ### Checklist - [x] Sources pinned; failed attempts and unknown usage preserved. - [x] No credential values or private client configuration published. - [x] Peer-owned settlement assertions preserved; no #775 hunks. - [x] Scoped379-test fixture retry passed; no new skip. - [x] Final registered-tree validate passed. - [ ] Exact-head GPT/CC read; the PR remains a draft. - [ ] Co-op host apply/read-backs and native smoke/counters after landing.
seathatflowsinourveins
added a commit
that referenced
this pull request
Oct 7, 2026
Inherited PYTHONOPTIMIZE=1 removed identity, vector and document assertions from the initial finding2 source. Reproduced 15 false passes with the unchanged controls. Invoke these standard-library validators with native python3 -I and explicitly inherit optimization in the independent positive/negative fixtures. No assertion, checker, source choice or client/provider setting is relaxed. Validation: focused8 pass; authentic coupled129 tests/3 existing skips pass; scripts/validate.py passes after temporary maintained hash registration. Retain original74bff464 and its failed reproduction. #810 carries both authored source commits unchanged and registers main's evidence last. SOTA: https://docs.python.org/3.13/using/cmdline.html#cmdoption-I SOTA: installed CPython3.13.16 python3 --help (-I implies -E/-P/-s; -O removes assert) SOTA: native-agent-stack@74bff464:tests/test_new_wsl_definitive_defaults.py:2119 (same independent controls)
5 tasks done
seathatflowsinourveins
added a commit
that referenced
this pull request
Oct 7, 2026
### Scope Route catalog lookup through QMD's lexical queries and bounded document retrieval; use SocratiCode for semantic catalog search at the main checkout's projectPath. Keep reranking available and prohibit qmd embed/pull. Update current maintenance recipes and native instruction carriers through the repository renderers. Existing Serena consumers read its manual before navigation, and jCodeMunch consumers obtain the current guide before a typed route with the actual caller model and execution off. Four existing Claude roles gain only Serena's manual tool. Reviewer/builder keep their no-menu boundary, and a focused read of a known path and line remains valid. This landing repair appends the preregistration's Amendment4 to bind the deliberately changed role bodies and compacts lane-authored RTK explanations within the unchanged compact and startup budgets. The entire composed pre-RTK prefix, upstream awareness, owner's blocks, exact SKILL line, seven exceptions, models and effort are preserved. Prior amendment rows, observations, seals and RUNBOOK bytes remain unchanged. The canonical code-graph rename is prepared as a separate patch and is excluded from this head under command-center ruling081016Z. Main's existing names remain in both templates, the map, fixtures and carriers. The protected lane token requires the owner's word before the rename can land. Until that follow-up, this host's Claude code-graph access is through the vendor's hook channel only; the stale MCP carrier id is a recorded limitation and is not claimed fixed. - Base commit: `c44993b379da25fae923dbbe70188978af5104aa` (verified native source; #803 already landed). - Lane: `lane:foundation`; draft. - Head: `9985e468fa4997d22630714b2fa6a76666e3cda9`. Native bytes: compact 8076 <8192, rendered 9142, three-file startup 20101 <=20103; full Claude block 4087 <=4100. These are byte gates, not token-use measurements. - Owned paths: token-lane carriers and their paired handbook text, minimal manual grants and mirrors, Codex template/rendered copies, native loader/routing/sequence tests, current recipes and generated handbook, append-only decisions/preregistration amendment, follow-up receipts and checksum/registry projections. - ROLE-CARRIER-GAPS and ROLE-GRANTS are separate future work and stay out of this change. Landing path: explicit command-center cue070358Z and fallback ruling081016Z permit one pushed rebase/content repair onto then-current main while #802/#813/#830 land. The co-op performs the new-head delta read with its CI result; the command center ACKs; 5f lands. A further pushed rebase requires the separate dated landing-conflict cue. Host configuration, cutover and fresh-session/working-day acceptance belong to their owners. ## SOTA sources - [QMD v2.8.3 lexical query](https://github.com/tobi/qmd/blob/v2.8.3/src/mcp/server.ts#L294), [typed searches/rerank](https://github.com/tobi/qmd/blob/v2.8.3/src/mcp/server.ts#L321) and [document retrieval](https://github.com/tobi/qmd/blob/v2.8.3/src/mcp/server.ts#L412): native lexical subqueries, retrieval and optional rerank. Rerank defaults true (:334-335). [README maintenance syntax:1033-1037](https://github.com/tobi/qmd/blob/v2.8.3/README.md#L1033-L1037) supplies bounded commands. - [SocratiCode v1.15.0 codebase_search](https://github.com/giancarloerra/SocratiCode/blob/v1.15.0/src/index.ts#L138): semantic search at an absolute projectPath. Owner ruling224125Z selects the routing and preserves rerank-on acceptance; this PR performs no host cutover. - [Serena c6fbd1c5 manual order](https://github.com/oraios/serena/blob/c6fbd1c5932df2494ffa0020af5a9fbe80b82143/src/serena/resources/config/prompt_templates/system_prompt.yml#L5-L6), [manual tool:28-40](https://github.com/oraios/serena/blob/c6fbd1c5932df2494ffa0020af5a9fbe80b82143/src/serena/tools/workflow_tools.py#L28-L40) and [project/session binding:44-49](https://github.com/oraios/serena/blob/c6fbd1c5932df2494ffa0020af5a9fbe80b82143/src/serena/tools/config_tools.py#L44-L49): manual first, active cwd project and returned session id for switches. - [jCodeMunch 1.108.330, 28803366, typed route:446-463](https://github.com/jgravelle/jcodemunch-mcp/blob/288033668f0425ab0547f420dd647c14bd186c7f/src/jcodemunch_mcp/server.py#L446-L463), [guide:4743-4752](https://github.com/jgravelle/jcodemunch-mcp/blob/288033668f0425ab0547f420dd647c14bd186c7f/src/jcodemunch_mcp/server.py#L4743-L4752) and [policy:101-128](https://github.com/jgravelle/jcodemunch-mcp/blob/288033668f0425ab0547f420dd647c14bd186c7f/src/jcodemunch_mcp/cli/policy.py#L101-L128): guide-first and typed task/model route. These files are unchanged at target1.108.331/d94049d0. The executable schema corrects the guide's query example; no new adaptive-tier setting or upstream rebuild. - `native-agent-stack@c44993b:tests/test_token_e2e_preregistration.py:637-663,905-909`: later dated role-pin amendment and sealed RUNBOOK contracts. Amendment4 preserves every earlier row/seal and extends the current-row selector. The command center records merge chronology at landing; this is a structural repair and authorizes no run. - [DeusData/codebase-memory-mcp v0.11.0 release](https://github.com/DeusData/codebase-memory-mcp/releases/tag/v0.11.0), the repository's pinned component, and the command center's native-client read-back identify the canonical server name. Its proposed client-key/wire-id/owner-token correction is kept in the separate patch, preserving every payload and strict fixture. Ruling081016Z explicitly keeps main's names in this publication while the owner's instruction-token decision remains open. - RTK explanation provenance: `native-agent-stack@2d849ba` (#726) and `@50b9579` (#389), `adoption/templates/codex.AGENTS.template.md` after its exceptions marker. The eight explanations shorten while all facts and protected bytes remain. Command-center064832Z accepted their class and scoped wording. - Native helpers at `native-agent-stack@c44993b`: `tools/adoption/managed_block.py:169` (`codex_block`), `tools/adoption/codex_roles.py:287` (`f4_block`), `scripts/host_receipts.py:710` (`register_file`), `docs/lanes.md:94-128`. Current generated inventory/handbook and checksum bindings use those repository tools; historical receipts and the frozen catalog passage are preserved. ### Evidence-class table | Claim | Evidence class | Command / receipt | | --- | --- | --- | | Documented native routing/manual/guide interfaces | source_review | Pinned upstream files above | | Carrier rendering, protected bytes, new body pins and tool ids | local_integration | Existing native helpers, hashes and required modules | | Permission/sequence/byte-mutant/frozen-row controls | synthetic | Existing tests with assertions preserved | | Deployed use and token savings | Pending owner evidence | No model, host apply or working-day measurement in this repair | ### Local commands run ```text $ CI=true nice -n 19 ionice -c3 python3 -B -m unittest tests.test_token_lanes_session_start tests.test_token_lanes_subagent_start tests.test_scaffold_repo tests.test_new_wsl_handbook tests.test_adoption_docs_consistency tests.test_codex_agents tests.test_codex_roles tests.test_codex_worker_lane tests.test_token_e2e_preregistration tests.test_new_wsl_client_config.RecordTests tests.test_install_claude_profile tests.test_managed_block tests.test_task_model_routing tests.test_new_wsl_client_config.MapTests tests.test_new_wsl_client_config.AgentGapTests tests.test_new_wsl_client_config.RenderTests tests.test_new_wsl_client_config.ApplyTests.test_the_first_run_writes_what_the_wired_pieces_name_and_nothing_else exit 0; Ran 595 tests in 109.929s; OK (skipped=14) $ CI=true nice -n 19 ionice -c3 python3 -B -m unittest tests.test_install_claude_profile tests.test_adoption_docs_consistency tests.test_new_wsl_client_config.RecordTests exit 0; Ran 150 tests in 16.306s; OK (skipped=2) $ CI=true nice -n 19 ionice -c3 node examples/claude-native/workflows/test-envelope.mjs exit 0; SUMMARY passed=254 failed=0 total=254 $ CI=true nice -n 19 ionice -c3 python3 -B tools/adoption/new_wsl_client_config.py --check --markdown exit 0 $ CI=true nice -n 19 ionice -c3 python3 -B scripts/build_new_wsl_handbook.py --check exit 0 ``` The passing 595-test run is retained on unchanged source/assertion inputs from the pre-window checkpoint; the 150-test run checks the concrete relocation-fixture and workflows-README changes in the new main base. Their counts overlap and are not added. The fresh Node run passes 254/254. Strict checksums in both Codex directories and the Claude hook directory, plus `git diff --check`, exit 0. All three Amendment-4 role digests were re-derived in both copies and are unchanged from the previous published head. ```text $ CI=true nice -n 19 ionice -c3 python3 -B scripts/validate.py exit 0; 69 components, 10,482 hashed files, 4 profiles, 224 receipts, status passed ``` The earlier landing composition used source `630b6ece8485a7710ea51321682d5a12e364e25c` and hot commit `5703ef1061b982078038425fada888fd35153868`; its native checks and source-critic result are retained. The authorized locator follow-up changes only `catalogs/foundation/upstream-surface-dispositions.json` and that file's registry row: the `otel.environment` citation now points to the command at `examples/codex-native/README.md:214`. Source commit `2b36cfee` is followed by registry-last commit `9985e468fa4997d22630714b2fa6a76666e3cda9`, with no further rebase. Claude/Codex instruction bytes, role hashes, historical records, namespace and byte budgets remain unchanged. The registry starts from exact main `c44993b379da25fae923dbbe70188978af5104aa`, registers the same 67 changed-file rows through the native producer, and preserves main's receipt and convergence arrays. The two-file correction passed the 114-test citation module, the two exact failing methods and a second native validator run. The worktree is clean; the shifted-citation scan found no other current target to repair. Historical before/after references remain intact. ```text $ CI=true nice -n 19 ionice -c3 python3 -B -m unittest tests.test_upstream_surface_watch exit 0; 114 tests, 6 skipped $ CI=true nice -n 19 ionice -c3 python3 -B -m unittest tests.test_upstream_surface_watch.DispositionCitationTests.test_every_cited_line_names_the_key tests.test_upstream_surface_watch.DispositionCitationTests.test_a_dotted_key_has_its_parent_near_the_citation exit 0; 2 tests $ CI=true nice -n 19 ionice -c3 python3 -B scripts/validate.py exit 0; integrity and scope passed ``` The hosted run at5703 returned two failures in11,495 tests because the documentation locator moved. A reviewed locator still gates when its repository test fails. This follow-up closes that exact contract; it adds no assertion waiver or namespace change. Read-only overlap metadata checked all 47 open PRs, including full file pagination where needed. Source overlaps: #645, #706, #709, #754, #769, #770, #775, #776, #795, #810, #821, #826, #829. Shared registry/checksum paths use the hot-file protocol; no peer branch is changed. Earlier failed CI at3ba ran11,381tests and failed six role-body-pin subcases; its complete failed log is retained privately. The failed 560-test and earlier 595-test preparation runs also remain retained. This repair uses a dated amendment, preserves the older source evidence and strict current-body comparison, and performs no full-suite or provider acceptance run. Validator results are integrity and scope evidence only. ### Decision record `docs/decisions/2026-10-06-qmd-lexical-catalog-instructions.md` and `docs/decisions/2026-10-07-serena-jcodemunch-native-navigation-wiring.md`, with appended clarifications and the new landing follow-up, explain the behavior and limits. The preregistration README gains Amendment4 only; its RUNBOOK and earlier sealed records stay byte-identical. No prior result is recast as a new run. ### Host evidence Repository-only change. No live instruction file, MCP registration, hook trust, client setting, gateway setting/key or model session changed. Source/render checks and synthetic fixtures remain distinct from the configuration owner's read-back and organic-use acceptance. ### Checklist - [x] No GitHub Actions or paid service change. - [x] No credentials, raw conversations or live client configuration committed. - [x] Models, effort, owner's blocks and historical observations preserved. - [x] Peer-owned worktrees and source preserved. - [x] Native validation passed; changed checksums/registry committed last at the final head.
seathatflowsinourveins
added a commit
that referenced
this pull request
Oct 7, 2026
### Scope Enforce the owner's 42-day model-currency rule in the existing offline validator, add daily review proposals, and fill the ruled partial inventory. This combined revision closes J826's three P2 findings, keeps J826b's simple background tasks on Haiku's own line, prepares independent model carriers and proposes MinerU's persistent local-source unit. - Base: `3d7d4a4b2640c8583d8855670111ef5d023d3e03`; no rebase for this revision. - Head: `de7a7f2b6b6d7eac4241701b5c06a63b3265cbe0`. - Lane: `lane:foundation`; draft. - Owned paths: model catalog/schema, existing validator/freshness proposer, currency and MinerU unit templates, focused tests, independent model carriers, current guides, dated decision amendments and new evidence. Shared stack model rows, current catalog citations and registry go in the last commit. - **Named landing path:** co-op GPT delta read → command center ACK → 5f's queue **after #810 and before #645**. Keep draft until that read and ACK; rebase once on the landing cue. The inventory declares **21 artifact/model-consumer rows and seven package snapshots**, remains `pending`, and retains **eight unresolved groups**. Passing validation checks declared rows; it does not prove complete inventory or deployed selections. **FILL item 7 is a prerequisite, not a completed fold.** Independent Nemotron endpoint/prefix carriers are prepared, but fixwave-defects owns the six-field memory-owner amendment to the living install plan/generated installer. It rides #810 after finding-2, per A26/A28/A27b. This lane neither recreates the absent generator nor hand-edits `install.sh`. The ai-memory example requires 2.5.2 prefix support; its stack component remains 2.4.1 under the model-only scope. This PR does not claim all folds 7/8/9 complete. Bounded item-7 state at landed #723 (`41b65ac6bf574afa247a2542e70340a3cadf38bb`): [install-plan.json:1590-1597](https://github.com/seathatflowsinourveins/native-agent-stack/blob/41b65ac6bf574afa247a2542e70340a3cadf38bb/evidence/artifacts/new-wsl-install-plan-20261002/install-plan.json#L1590-L1597) selects ai-memory 2.5.2, but its [memory command at :1599](https://github.com/seathatflowsinourveins/native-agent-stack/blob/41b65ac6bf574afa247a2542e70340a3cadf38bb/evidence/artifacts/new-wsl-install-plan-20261002/install-plan.json#L1599), [notes at :1644](https://github.com/seathatflowsinourveins/native-agent-stack/blob/41b65ac6bf574afa247a2542e70340a3cadf38bb/evidence/artifacts/new-wsl-install-plan-20261002/install-plan.json#L1644) and [generated install.sh:465](https://github.com/seathatflowsinourveins/native-agent-stack/blob/41b65ac6bf574afa247a2542e70340a3cadf38bb/evidence/artifacts/new-wsl-install-plan-20261002/install.sh#L465) still select local MiniLM. The [stack component at :96](https://github.com/seathatflowsinourveins/native-agent-stack/blob/41b65ac6bf574afa247a2542e70340a3cadf38bb/manifests/stack.json#L96) remains 2.4.1. Fixwave's prerequisite patch owns all six settings: `embedding_provider=openai-compat`, `embedding_base_url=http://127.0.0.1:28231/v1`, `embedding_model=nvidia/Nemotron-3-Embed-8B-BF16`, `embedding_dim=4096`, `embedding_query_prefix="query: "`, and `embedding_document_prefix="passage: "`. This source review does not establish the host's effective configuration. **MinerU's unit is source-only.** It sets `MINERU_MODEL_SOURCE=local`, supervises the shipped foreground DocLib module and carries `OnFailure=stack-alert@%n.service`. The embedded CPU guard is a **gap dated 2026-10-07**: no documented embedded ingress was found in the inspected installed source/tag. The maintained external wrapper offers a separate supported ingress and lifecycle for the CC to consider. The CC resolves the CPU precondition before activation, verifies managed mode/tier, applies the unit and performs the fresh-shell smoke. No host setting is applied here. ### CI repair at this head The previous head `5a6842f1` failed the hosted 11,544-test suite (nine failures, one error). This revision repairs all four dependency groups: retain the complete historical 1B model/evidence record with retired-on-2604 scope and its current locator; restore the portable Claude endpoint defaults while retaining 8B/4096; disclose the post-release MCP change on three setup pages; and move the SDK disposition citation to its actual assertion line. The setup-page changes regenerate both handbook outputs with the repository's existing builder. The original October 1 receipt stays byte-identical. A new dated [handbook projection](https://github.com/seathatflowsinourveins/native-agent-stack/blob/de7a7f2b6b6d7eac4241701b5c06a63b3265cbe0/evidence/artifacts/model-currency-enforcement-20261007/handbook-projection.json) binds the current bytes; the publication test changes only its receipt locator and keeps every exact assertion. No parity, release-marker, pin-source or citation oracle is relaxed. All 23 targeted failure/negative-control tests and 95 handbook/stack/pin tests pass (exit 0). Validation passes before and after the final shared-metadata/registry commit (exit 0, 10,404 file hashes). The [CI repair receipt](https://github.com/seathatflowsinourveins/native-agent-stack/blob/de7a7f2b6b6d7eac4241701b5c06a63b3265cbe0/evidence/artifacts/model-currency-enforcement-20261007/ci826-repair.json) retains the hosted failure and local failed attempts. The new head's hosted CI and synthetic-merge execution are separate, still unverified observations. One same-branch push; no rebase, host apply or #790 change. ## SOTA sources - CI dependency sources: `native-agent-stack@5a6842f`: `adoption/hosts/example.json`, `tools/adoption/install_claude_profile.py:350-378`, `tests/test_install_claude_profile.py:1509-1530`, `tests/test_adoption_docs_consistency.py:499-538`, `tests/test_new_wsl_handbook.py:1672-1689`, and `tests/test_upstream_surface_watch.py:2054-2060`; original 1B fields at `native-agent-stack@3d7d4a4:manifests/stack.json:2098-2109`. The new dated decision amendment records the exact source contracts and preserves historical acceptance. - `huggingface/huggingface_hub@v2.1.1` / `bd4a76030582d118e28dc9ad6c7a5911ea76176b`: [native models-info CLI](https://github.com/huggingface/huggingface_hub/blob/bd4a76030582d118e28dc9ad6c7a5911ea76176b/src/huggingface_hub/cli/models.py#L223-L237), [moving main revision](https://github.com/huggingface/huggingface_hub/blob/bd4a76030582d118e28dc9ad6c7a5911ea76176b/src/huggingface_hub/constants.py#L60), [metadata semantics](https://github.com/huggingface/huggingface_hub/blob/bd4a76030582d118e28dc9ad6c7a5911ea76176b/src/huggingface_hub/hf_api.py#L799-L819), [implicit-token early return](https://github.com/huggingface/huggingface_hub/blob/bd4a76030582d118e28dc9ad6c7a5911ea76176b/src/huggingface_hub/utils/_headers.py#L147-L151). Installed native 2.1.1 is reused. P2a keeps exact Hub origins independent of package age. - [GitHub latest-release API](https://docs.github.com/en/rest/releases/releases#get-the-latest-release) and [PyPI JSON API](https://docs.pypi.org/api/json/#get-a-project), read 2026-10-07, supply primary metadata; changes become review requests. - [Python 3.13 URL parsing](https://docs.python.org/3.13/library/urllib.parse.html#urllib.parse.urlsplit) documents lazy port errors and control/space stripping. The check rejects raw whitespace/C0/C1 controls before parsing and evaluates `parts.port` (P2c). [JSON Schema object semantics](https://json-schema.org/understanding-json-schema/reference/object#additionalproperties) supplies `additionalProperties: false` on complete object definitions, mirrored by validator allowed-key checks (P2b); partial conditional schemas are not incorrectly closed. - [Filled primary provenance](https://github.com/seathatflowsinourveins/native-agent-stack/blob/de7a7f2b6b6d7eac4241701b5c06a63b3265cbe0/evidence/artifacts/model-currency-enforcement-20261007/fill-source-review.json) preserves packet SHA-256 `44679fd1d26dfec36b3b57cdbde2df4290d1b068118562cb8811bf3eb7cd6941`, per-row vendor/artifact/package sources and corrections. Unknown identities remain pending. Haiku's public date is from [Anthropic's October 15, 2025 announcement](https://www.anthropic.com/news/claude-haiku-4-5); the exact ID is checked against [the current Haiku specification](https://platform.claude.com/docs/en/models/haiku-4-5/overview), read 2026-10-07. Its snapshot suffix is not its public launch date. A Haiku 5.x ID overturns this dated exception. - `akitaonrails/ai-memory@v2.5.2` / `7580b74d0fb9d14a6d949dc92f5ea8bb7feb3c83`: [prefix configuration](https://github.com/akitaonrails/ai-memory/blob/7580b74d0fb9d14a6d949dc92f5ea8bb7feb3c83/crates/ai-memory-cli/src/config.rs#L454-L502) and [factory wiring](https://github.com/akitaonrails/ai-memory/blob/7580b74d0fb9d14a6d949dc92f5ea8bb7feb3c83/crates/ai-memory-llm/src/factory.rs#L262-L269). Other independent carrier sources and exact pins remain in the [fill amendment](https://github.com/seathatflowsinourveins/native-agent-stack/blob/de7a7f2b6b6d7eac4241701b5c06a63b3265cbe0/docs/decisions/2026-10-07-model-currency-enforcement.md#amendment-2026-10-07-ruled-partial-fill-and-current-carriers). Installer/component alignment stays owner-routed. - `opendatalab/MinerU@mineru-4.0.10-released` / `c221cc41bc911ad0df3eaeda97acf6a1bfe9bf93`: [documented server command](https://github.com/opendatalab/MinerU/blob/c221cc41bc911ad0df3eaeda97acf6a1bfe9bf93/README.md#L462-L484), [native foreground-module launch](https://github.com/opendatalab/MinerU/blob/c221cc41bc911ad0df3eaeda97acf6a1bfe9bf93/mineru/cli/commands/server.py#L242-L248), [module entry](https://github.com/opendatalab/MinerU/blob/c221cc41bc911ad0df3eaeda97acf6a1bfe9bf93/mineru/doclib/app.py#L443-L449), [environment override](https://github.com/opendatalab/MinerU/blob/c221cc41bc911ad0df3eaeda97acf6a1bfe9bf93/mineru/config.py#L156-L176), and [managed-parser inheritance](https://github.com/opendatalab/MinerU/blob/c221cc41bc911ad0df3eaeda97acf6a1bfe9bf93/mineru/doclib/background/parse_server_health.py#L205-L221). [Embedded kwargs](https://github.com/opendatalab/MinerU/blob/c221cc41bc911ad0df3eaeda97acf6a1bfe9bf93/mineru/model/vlm/runtime.py#L175-L180) differ from the [external wrapper's environment ingress](https://github.com/opendatalab/MinerU/blob/c221cc41bc911ad0df3eaeda97acf6a1bfe9bf93/mineru/kit/vlm_server/llama_cpp_server.py#L263-L273); installed `mineru_llama_cpp/engine.py:46-52,107-108` defaults to 99 GPU layers. This is repository systemd integration, not an upstream systemd recipe. - `systemd/systemd@v259.5` / `b3d8fc43e9cb531d958c17ef2cd93b374bc14e8a`: [currency command-failure prefix](https://github.com/systemd/systemd/blob/b3d8fc43e9cb531d958c17ef2cd93b374bc14e8a/man/systemd.service.xml#L1443). MinerU follows [systemd@v260 service semantics](https://github.com/systemd/systemd/blob/v260/man/systemd.service.xml#L158), [failure/start limits](https://github.com/systemd/systemd/blob/v260/man/systemd.unit.xml) and [control-group stop semantics](https://github.com/systemd/systemd/blob/v260/man/systemd.kill.xml#L56), checked against the installed 259 manual. Alerting covers DocLib/unit failure, not a parser failure that leaves DocLib alive. - `native-agent-stack@3d7d4a4`: `scripts/freshness_propose.py:861,942`, `scripts/validate.py:144,484`, `adoption/templates/systemd/stack-currency.service:16-26`, `docs/acceptance-evidence-policy.md:24-38`, `docs/lanes.md:94-115`. The owner-defined 42/7-day policy and package join are the deterministic gap. No upstream is rebuilt and no validator executable is added. ### Evidence-class table | Claim | Evidence class | Command / receipt | | --- | --- | --- | | UTC clock, 42/43-day and 7/8-day boundaries, package join, J826 P2 positive/negative controls | `synthetic`, `local_integration` | 113 combined policy/proposal/existing-validator tests, exit 0; [new J826 checks](https://github.com/seathatflowsinourveins/native-agent-stack/blob/de7a7f2b6b6d7eac4241701b5c06a63b3265cbe0/evidence/artifacts/model-currency-enforcement-20261007/j826-checks.json). | | Primary filled dates/revisions, retained exceptions and unresolved roles | `source_review` | `fill-source-review.json` records the earlier 20-row partial fill; `j826-checks.json` adds the source-reviewed Haiku ruling and current 21-row state. Seven packages and eight unresolved groups remain; no inference or live-selection claim. | | Legacy currency proposer/timer compatibility | `local_integration` | Earlier 91 checks, exit 0, in `checks.json`. | | Handbook/stack/pin carriers and SDK controls | `synthetic`, `local_integration` | Earlier 95 handbook/stack/pin checks and 27 SDK tests, exit 0, in `fill-checks.json`; the recorded 516-test carrier check also exited 0. | | MinerU source/failure/lifecycle template | `synthetic`, `local_integration` | Eight focused tests and unit syntax verification, exit 0, in `fill-checks.json`; no host application. | | Repaired architecture, portable parity, release disclosures and citation controls | `synthetic`, `local_integration` | 23 targeted tests plus 95 handbook/stack/pin tests, exit 0; `ci826-repair.json` and dated `handbook-projection.json`. | | Publication consistency after the registry-last commit | `structural_validation` | `exit 0; 69 components, four profiles, 223 receipts, 10,404 file hashes; structural integrity/scope only`. | ### Local commands run Checks use owned durable scratch and `nice -n 19 ionice -c3`. ```text python3 -B -m unittest tests.test_model_currency tests.test_model_currency_propose tests.test_validate exit 0: 113 tests after J826 fixes. python3 -B -m unittest tests.test_new_wsl_handbook tests.test_stack_lifecycle tests.test_foundation_stack_pins exit 0: earlier 95 tests. uv run --no-project --with claude-agent-sdk==0.2.162 python3 -B -m unittest discover -s examples/claude-runtime-sdk/tests exit 0: 27 tests. uv run --no-project --with claude-agent-sdk==0.2.162 python3 -B examples/claude-runtime-sdk/tests/check_controls.py --disarm-route exit 1: deliberate negative; unchanged normal control exit 0. python3 -B -m unittest tests.test_mineru_user_unit exit 0: 8 tests. systemd-analyze --user verify adoption/templates/systemd/mineru.service exit 0: syntax only; no unit loaded or started. nice -n 19 ionice -c3 python3 -B scripts/validate.py exit 0; 69 components, four profiles, 223 receipts, 10,404 file hashes; structural integrity/scope only ``` The **two-day advance refresh is included**, beginning on landscape-check day five. The collector appends private review JSON/Markdown rather than publishing a GitHub PR, renewing dates or switching models. Hub SHA differences are revision candidates, not proof of new weights. Unsupported vendor-page adapters request source review. The initial pending/empty offline dry run returned `complete_inventory`, zero source checks and no writes; it remains a dated scaffold observation. Retained failed attempts remain in the original receipts: the initial legacy check's old owned-file hash failure, the fixture's missing overturn field and the scratch-classification correction. Historical observations and decided text are preserved; new checks and dated amendments extend them. ### Decision record [Dated decision and amendments](https://github.com/seathatflowsinourveins/native-agent-stack/blob/de7a7f2b6b6d7eac4241701b5c06a63b3265cbe0/docs/decisions/2026-10-07-model-currency-enforcement.md) records alternatives, sources, overturn triggers, partial fill and J826/J826b. [Current guide](https://github.com/seathatflowsinourveins/native-agent-stack/blob/de7a7f2b6b6d7eac4241701b5c06a63b3265cbe0/docs/model-currency.md) carries the pending item-7 prerequisite. [MinerU recipe](https://github.com/seathatflowsinourveins/native-agent-stack/blob/de7a7f2b6b6d7eac4241701b5c06a63b3265cbe0/recipes/mineru-local-service.md) carries CC activation and the dated embedded-guard gap. Receipts: [initial checks](https://github.com/seathatflowsinourveins/native-agent-stack/blob/de7a7f2b6b6d7eac4241701b5c06a63b3265cbe0/evidence/artifacts/model-currency-enforcement-20261007/checks.json), [fill checks](https://github.com/seathatflowsinourveins/native-agent-stack/blob/de7a7f2b6b6d7eac4241701b5c06a63b3265cbe0/evidence/artifacts/model-currency-enforcement-20261007/fill-checks.json), and [new J826 checks](https://github.com/seathatflowsinourveins/native-agent-stack/blob/de7a7f2b6b6d7eac4241701b5c06a63b3265cbe0/evidence/artifacts/model-currency-enforcement-20261007/j826-checks.json). Raw logs stay private. ### Host evidence No new host receipt/status is promoted. No weights, inference, package bump, live model switch, timer/service installation or client-configuration change was run for this revision. CC MinerU application and fresh-shell acceptance remain pending; A23's one-page parse remains a separate dated observation. Complete inventory, unsupported adapters, native timer execution and whole-unit timeout recovery remain unqualified. Item 7's owner-routed prerequisite is not counted as a completed fold. ### Checklist - [x] Draft, exactly `lane:foundation`, named read/ACK/queue path. - [x] No new validator executable, rebuilt upstream, workflow or paid surface. - [x] No credentials inspected, logged or committed; Hub implicit auth disabled. - [x] Peer-owned paths/heads and queued branches preserved. - [x] Historical results/decided text preserved; dated amendments and new receipts used. - [x] Targeted controls pass; partial inventory and pending owner/CC apply gates explicit. - [x] Final stack/registry commit is last; validation passed before and after it (exit 0).
seathatflowsinourveins
added a commit
that referenced
this pull request
Oct 7, 2026
### Scope Restore the handbook test's live publication binding so a later profile or handbook regeneration updates its own maintained receipt, instead of editing #826's dated landing record. - Base commit: `6d252c9c102e575bc9229bf6bb2149409655d5f3`; head: `4d2ce2a7719adca918025ffde295ef1dfa30bf02`. - Lane: `lane:foundation`; draft. - Exactly four paths: `tests/test_new_wsl_handbook.py` (one receipt-path line), `evidence/artifacts/new-wsl-handbook-20261001/receipt.json` (current computed binding plus one appended regeneration), its one-line `README.md`, and registry last. - The generator, profile, both generated outputs, hash/inventory assertions, existing receipt observations/history and #826's dated record stay byte-identical. - **Named landing path:** co-op cross-family exact-head read, hosted validate, command center CI read and cue; **right after #820, ahead of #775**. ONE replay at that cue if #820 changes main; full test phase in that landing turn. Hold this draft head for the reads. ## SOTA sources - `native-agent-stack@6d252c9:tests/test_new_wsl_handbook.py:1672-1691`: the publication contract checks actual current generator/profile/output hashes and inventory, with byte-change negative controls. Its comment says hashes follow regeneration. Only its receipt locator changes. - Same pin, `evidence/artifacts/new-wsl-handbook-20261001/receipt.json:268,656,700`: the receipt is a current hash mirror and prior entries expressly refresh mutable producer/output bindings while preserving observations. This existing repository practice fills the publication-binding gap; no new manifest, generator or validator is introduced. - Same pin, `docs/acceptance-evidence-policy.md:59-65` and `docs/landscape-domain-notes.md:33,38`: retain historical inputs/results/failures/pins. All prior receipt fields, regeneration prefix and `previous_outputs` are retained; #826's dated record remains historical. - Same pin, `docs/lanes.md:94-115`: main-copy/own-registration protocol; `scripts/host_receipts.py:710` updates the three own hashes and inserts the README, in the registry-only final commit. ### Evidence-class table | Claim | Evidence class | Command / receipt | | --- | --- | --- | | Existing strict publication contract passes at the maintained binding | `local_integration`, `synthetic` | 87 handbook tests, exit 0; existing byte-change negative controls unchanged. | | Actual generated bytes are current | `local_integration` | Native builder `--check`, exit 0; MD a1a0cb17… and JSON a42c8915… as recorded in the appended regeneration. | | Dated records/history/other test assertions preserved | `source_review`, `local_integration` | Exact base-file/JSON comparison, exit 0; independent bounded source critic ACK. | | Publication integrity after registry-last commit | `local_integration` | `validate.py` before/after, exit 0; 69 components, four profiles, 224 receipts, 10,508 hashes. | ### Local commands run All checks use the owned external TMPDIR, `nice -n 19 ionice -c3` and closed stdin. ```text python3 -B scripts/build_new_wsl_handbook.py --check exit 0, status passed; current outputs unchanged. python3 -B -m unittest tests.test_new_wsl_handbook exit 0: Ran 87 tests in 37.368s; OK. Exact unchanged-field, regeneration-prefix, dated-record and one-test-line comparison exit 0; all preservation controls true. python3 -B scripts/validate.py exit 0 before and after final registry commit; 10508 file hashes. git diff --check exit 0. ``` The scoped cached open-PR check found 44 open PRs and these overlapping neighbours: #645, #754, #769, #770, #810 on the maintained receipt; #776 and #810 on the test. The CC assigned this narrow critical-path repair before later regeneration PRs. No peer head or file is edited by this lane. ### Decision record The CC's J-HANDBOOK-BINDING ruling supplies the bounded disposition. The one-line receipt README identifies the maintained current-state carrier. No old decision or dated observation is rewritten, and no extra decision file is added to this small unit. ### Host evidence No host/client/installation or acceptance-status changes. These checks establish the repository publication contract, not upstream model or native host acceptance. ### Landing read fields The intentional main-test edit is `tests/test_new_wsl_handbook.py:1677`, disposed of in advance by the CC for this exact receipt rebind. Name it in `main-tests=` at the landing read. Declare every actual post-read/replay change in `adapted=`; the full-suite phase is deferred to the landing turn. Nothing is re-pointed to a new dated path. ### Checklist - [x] Existing current-state receipt reused; earlier observations and dated record preserved. - [x] Only one test line changes; assertions, producer and profile unchanged. - [x] Own entries registered last; required local checks pass. - [x] No workflow, dependency, paid surface or credential change. - [x] Draft and one foundation label; named read/ACK/queue path.
seathatflowsinourveins
added a commit
that referenced
this pull request
Oct 9, 2026
### Scope - What this PR changes, in one or two sentences: the secret-path guard's PreToolUse hook declares Claude Code 2.1.295's `onFailure: "block"` in both repository declarations, so a guard that can't start, times out or exits with an unexpected code blocks the Bash call instead of letting it through. A new test pins that the guard hooks carry the key and no other hook does. - Base commit: `9f7b38b23d15e47ceb6f68c6651b86088f621797` - Lane: `lane:foundation` - Owned paths touched: `.claude/settings.json`, `adoption/templates/claude.settings.template.json`, `adoption/new-wsl/client-config-map.json` (the guard entry's note), `tests/test_install_claude_profile.py`, `docs/decisions/2026-10-08-guard-hook-fails-closed.md`, `manifests/evidence.json` (hashes re-registered with `scripts/host_receipts.py` `register_file`) ### SOTA sources - [anthropics/claude-code `CHANGELOG.md` at `602df92bf481ed904533e95c09f740f40aab5aed`](https://github.com/anthropics/claude-code/blob/602df92bf481ed904533e95c09f740f40aab5aed/CHANGELOG.md#L5), entry `2.1.295`, line 5: "Added `onFailure: "block"` for command and HTTP hooks: a hook that can't start, times out, or exits with an unexpected code blocks the action instead of letting it through". The file's SHA256 is `c0b1f9aa313a8021d863ec0053c047b0acc792d74e04af9c7741d3ecd23c3d1a`, the same bytes as `main` when read on 2026-10-08. - The installed Claude Code 2.1.295 client (binary SHA256 `4503bfe11a6c7fcc1e0b39b5e0d347c04248f750b03b0977b3ad6b531fe6f358`). Its hook schema declares `onFailure` on command and HTTP hooks. Its runner skips the key on `async` and `asyncRewake` command hooks and ignores it on `Stop`, `SubagentStop`, `TaskCompleted` and `TeammateIdle`. - [Claude Code hooks reference](https://code.claude.com/docs/en/hooks.md), read 2026-10-08: "Other exit codes" (only exit 2 blocks without the key; exit 1 is a non-blocking error) and "Disable or remove hooks" (`--settings '{"disableAllHooks": true}'` for one run; file edits are normally picked up by the file watcher). The page does not document `onFailure` yet. ### Changed existing test expectations None. One test method is added: `tests/test_install_claude_profile.py` `SecretGuardProfileTests.test_only_the_secret_guard_hook_fails_closed`. It walks every command hook in `.claude/settings.json` and in every JSON file under `adoption/`. Every hook that runs `secret_path_guard.py` must carry `onFailure: "block"`, no other hook may carry the key, and both known declarations must be found. No existing assertion changed. ### Evidence-class table | Claim | Evidence class | Command / receipt | | --- | --- | --- | | With the key on the hook entry, a PreToolUse Bash hook that exits 3, is not found (127) or times out blocks the call. Without the key, the call runs. On the matcher group the key does nothing, and a hook that exits 0 is unaffected. | `native_proven` | Ten `claude -p --model claude-haiku-5-5 --strict-mcp-config` runs on this host, 2026-10-08 22:13Z to 22:48Z: cases A to J in the decision record, with transcript hashes | | The key also works from a `--settings` file; a user settings file has the same hook shape | `native_proven` / `source_review` | Cases I and J; the user scope was not run, because a probe must not edit the live user settings | | Client 2.1.294 loads the file, ignores the key and runs the call | `native_proven` | Case B run with the 2.1.294 binary | | Only the guard hooks carry the key | `local_integration` | New test method; four mutations each fail it (either guard without the key, `rtk` with it, effort guard with it), and the files were restored byte for byte | | A new-WSL render writes the key on the guard hook only | `local_integration` | `new_wsl_client_config.py --render --host example --out <scratch>`: rc 0; `settings.json` has the key on the guard and on none of the other 15 command hooks | | `apply_claude_settings.py` does not carry the key to a host that already runs the guard | `local_integration` | `merge_settings` run against an existing-host base and an empty base (see open points) | ### Local commands run ``` $ timeout 600 python3 -m unittest tests.test_install_claude_profile # rc 0, Ran 110, OK (skipped=1) $ timeout 600 python3 -m unittest tests.test_adoption_docs_consistency # rc 0, Ran 39, OK (skipped=1) $ timeout 600 python3 -m unittest tests.test_new_wsl_client_config # rc 0, Ran 228, OK $ timeout 600 python3 -m unittest tests.test_apply_claude_settings # rc 0, Ran 53, OK $ timeout 600 python3 -m unittest tests.test_workflow_policy # rc 0, Ran 29, OK $ timeout 600 python3 -m unittest tests.test_upstream_surface_watch # rc 0, Ran 114, OK (skipped=6) $ timeout 600 python3 -m unittest tests.test_credential_status # rc 0, Ran 48, OK $ timeout 600 python3 -m unittest tests.test_task_model_routing # rc 0, Ran 10, OK $ timeout 900 python3 scripts/validate.py # rc 0, {"hashed_files": 10901, "status": "passed"} $ python3 scripts/evidence_manifest.py --check # rc 0 $ python3 -B tools/adoption/new_wsl_client_config.py --render --host example --out <scratch> # rc 0 ``` The first six modules were run again at the committed head, with the same results. ### Decision record [`docs/decisions/2026-10-08-guard-hook-fails-closed.md`](docs/decisions/2026-10-08-guard-hook-fails-closed.md). It holds the sources, the evidence table and the risk: a broken guard now blocks every Bash call, and the guard's median run time was 74 to 87 ms against its 10 s timeout. It also notes the SchemaStore schema (no `onFailure`, `additionalProperties: false`) and the inverse: delete the key, or for one run pass `--settings '{"disableAllHooks": true}'`. ### Open points - **Existing hosts.** `tools/adoption/apply_claude_settings.py` merges hooks by command and keeps the existing hook object. On a host that already runs the guard, applying this template therefore leaves the live hook without the key. A new host and a new-WSL build do get it. On an existing host the live change is one key added by hand to the guard's PreToolUse `Bash` hook object. Changing the merge contract is a separate change. - **Minimum client.** The key needs Claude Code 2.1.295. Older clients ignore it and keep failing open. - **One SocratiCode registration (the brief's second change) is not in this PR, by coordinator ruling.** On this host Claude Code skips the plugin's server because the user-scope entry has the same command line, so each Claude session connects one SocratiCode server. The standing decision (`docs/decisions/2026-10-07-clean-upstream-install-finalizes-a-candidate.md:150-152`) pairs the plugin with that entry. The repository lags that host state: - `adoption/mcp/claude-user.json` still declares the older command form. - The map's `claude/mcp/server/socraticode` row cites a 1.15.0 path. - No template enables the plugin or sets `SOCRATICODE_SPEC`. Drafts #776, #810 and #834 touch the same file. This is an open item for a single owner, to be assigned by the coordinator. ### Host evidence Not applicable: no file under `evidence/hosts/` changes. ### Checklist - [x] New/changed GitHub Actions are pinned to a full commit SHA with a version comment (no floating tags). (No workflow change.) - [x] New/changed workflows declare top-level `permissions: {}` and grant each job only what it needs (`contents: read`, or an explicitly justified addition). (No workflow change.) - [x] No secrets are printed, logged or committed; no new required secret was added without a documented owner. - [x] No new paid hosting, subscription or billing surface was introduced. - [x] Peer-owned untracked files and worktrees were preserved (not deleted, moved or overwritten). 🤖 Generated with [Claude Code](https://claude.com/claude-code)
This branch has not been deployed
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.
Scope
Record NativeStack2604's final retrieval-first local-model choices and correct the foundation sources that still recommend the retired picks. This draft carries the regenerated source projections and actual local check results. The install-plan coupling remains a retained failure pending the plan owner’s atomic correction; this head is not ready for queue.
SOTA sources
Evidence-class table
Current source choices have measurement=null. Historical returned measurements and their receipts stay under historical_settlement, with their original family verdicts. Current family fields expressly record the source-only choice. No new model verdict or local check is upstream acceptance, and no readiness aggregate is derived.
Local commands run
Generation and checks
The direct LOCAL-MODEL-NOW direction lifted the initial execution hold. Commands use nice19 ionice-c3, one heavy step at a time. The assembler, render_tables.py --write, build_new_wsl_handbook.py --write/--check and build_ecosystem.py --write returned0. vLLM-only freshness/source identity was rechecked against its official0.31.0 release and tag; the complete old0.30 row remains separately recorded. The exact row diff awaits convergence-practice review.
Passing repaired scopes: handbook87tests; install-profile106tests; adoption-docs39tests(one skip); client configuration documentary count; source-projection4tests; architecture pin-source and render-config tests. Full definitive-defaults87tests still returns1 because the unchanged plan contradicts the three new defaults in7 checks. Full client-config205tests rerun returns1 with exactly two plan-owner failures; the documentary-count correction passes. Final validate and tables/handbook/ecosystem current-output checks all return0. This is a generated documentary draft with retained integration failures, not a green queue request.
First failures and repaired results remain separate in evidence/receipts/ns2604-local-models-local-checks-20261006.json with actual returned summaries and output hashes. Logs are durable private observations. None is an upstream model test, host installation, organic use or a readiness count. The plan owner authors the atomic correction; no owner assertion is skipped, weakened or silently waived.
Documentary repair P2-1 and historical point5
The new source-review receipt now contains all16 /syn/delete/i dispositions, each bound to the verified rulingSHA cc8dec6a8981966fdc6dfcb2929b6722843e30bd526c3b981e9289cdedd8059a. Each keeps the executor, preparation scope, original gate, ordering, exclusions and separately dated observed state. Three CC manifests record deletion (8B, Gemma and laterMiniLM);12states remain unverified, Docling held. EarlierMiniLM HOLD and its23:47integration rerun remain alongside the23:52:52owner deletion report; separate Gate3/rollback-closure provenance is not inferred. This PR performs no deletion or model run.
The original09-27 record is restored byte-for-byte through its57,789-byte prefix, SHA c3e2d0a439a8c0a0e74d9f607e69ddde65ffbbc1cdcf4da6b0048f6af5076564. Both original headings and all holds/tables stay unchanged; the dated supersession amendment is appended at the end.
Rebased onto0c4d3286: main registry restored, main stack plus only owned field/model deltas preserved, main generator-owned history unioned with our owned appends.37shifted pin-source bindings are refreshed through the existing tests/test_architecture_pin_source.py:49 resolver, not a new utility or invented source. The exact approved vLLM snapshot row remains unchanged.
Post-rebase95handbook/pin/source-projection tests returned0(37.738s), all tables/handbook/ecosystem current-output checks0 and final full validate0(69components/4profiles/226receipts/10387files). Initial post-rebase ecosystem attempt returned1 before owned receipts were re-registered; after registration it returns0. All actual attempts remain durable/publicly summarized. Same-family source critic reports0critical gaps in the immutable documentary repair; that is not independent native acceptance.
The7model/plan incompatibilities and2related client-plan failures remain open under A28. Owner-authored FINDING2-FOR-810 is carried unchanged only after#723lands; no plan/test edits, skipped gates or green queue request.
Later adjudication provenance is now a NEW linked record: evidence/receipts/ns2604-minilm-cleanup-adjudication-20261007.json and an appended Amendment(2026-10-07) in the new decision. The earlier16-item receipt, HOLD, rerun and deletion observations remain unchanged. The CC's explicit Gate3PASS/tier-1rollbackretirement is item235312Z:12/SHA921574d6..., with an actual ledger kind=status (not the co-op's initial 'send' description). Ruling/deletion rows are recorded23:53:12Z; the manifest's completion remains23:52:52Z. No earlierdecisiontime or newnativequalification is inferred. FinalNO-REWRITESCOPE applies: immutableobservations/DECIDEDtext, generator-ownedcurrentcontent allowed, newdatedsource/addendum andregistrylast throughvalidate.
Decision record and application boundaries
docs/decisions/2026-10-06-retrieval-first-local-models.md records all ten roles, the bundled PocketSphinx named exception, prior 1B/MiniLM/generation supersession, and model/consumer/date/re-drive limits.
The generic renderer has no automatic existing-index namespace binding. adoption/update.md retains the Native2604 owner deployment/read-back contract for ns2604_lm20261006_8b_. Template fixes do not apply either client's settings. Claude's endpoint migration, both-client qmd cutover, index/model deletions, Granite activation and MinerU verification remain owner-gated. The rc6 A26 hold is unchanged.
Function ownership and integration order (A27)
This PR lands first through the CC read/ACK and 5f queue; #723 and #776 then rebase onto it. Current sources do not waive any CI/acceptance gate.
Root owns Manifest.as_decided and the original/current source distinctions in:
Fixwave-defects/#723 owns all install-plan assertions, including unchanged InterimPlanChecks and:
Those three classes are byte-equal to the base in this PR. Client-config plan assertions also remain owner-held: ManifestRuleTests.test_the_rule_selects_the_rows_the_plan_installs_and_the_two_on_demand_rows and PlanConfigurationTests.test_plan_checker_counts_direct_config_installs_and_refuses_an_uncopied_file. #723 applies the retired Ollama/Swift/qmd-embedding plan corrections and new pinned-model rows; this PR edits none of its plan files. Coupled manifest/plan consistency must be checked after the stated integration order, with actual failures retained.
Trading handoff — finding 6 (NO trading-path edits)
At 0d5e650, all following paths are trading custody under docs/lanes.md:25:
Convergence-practice's sweep-catalog PR is untouched. Host state and dated raw receipts are preserved.
Checklist