docs: technical memory architecture + landing page clarity by UfukNode · Pull Request #30 · NousResearch/hermes-agent

UfukNode · 2026-02-25T23:24:00Z

No description provided.

Adds a Knowledge tab alongside the existing Memory tab under /memory. The knowledge browser renders markdown wiki pages from ~/.hermes/knowledge/ with support for: - YAML frontmatter parsing (title, type, domain, status, tags) - Directory-based navigation with folder tree sidebar - [[wikilink]] resolution as clickable in-app navigation - Backlinks ('Pages that link here') - Full-text search across all wiki pages - Tag filtering - Metadata display (type, domain, status, created/updated) - Graph view dialog showing page connections - 'Ask agent about this' button linking to chat - Graceful empty state when no knowledge directory exists Server side: - src/server/knowledge-browser.ts — wiki scanner, frontmatter parser, wikilink extractor, graph builder - 4 API routes: /api/knowledge/list, /read, /search, /graph Frontend: - src/routes/memory.tsx — tabbed layout (Memory | Knowledge) - src/screens/memory/knowledge-browser-screen.tsx — full browser UI Knowledge tab works independently of gateway capabilities (reads local filesystem directly). No new dependencies added. Closes NousResearch#30

) When the WebUI's model picker sets the active provider to "venice" (or "crof", "bankr", "cometapi"), the agent's resolve_provider_client() looked up the slug in PROVIDER_REGISTRY, failed to find it, and raised: "Provider 'venice' is set in config.yaml but no API key was found. Set the VENICE_API_KEY environment variable..." VENICE_API_KEY WAS set in env — the failure was because the agent had no idea which env var or base URL to use for "venice" as a slug. PROVIDER_REGISTRY entries for the other OpenAI-compatible aggregators (Z.AI, MiniMax, Mistral, etc.) already existed; Venice/CrofAI/Bankr/CometAPI just hadn't been added in the upstream code. Adds four api_key-auth entries mirroring the existing OpenAI-compat ones: - venice → https://api.venice.ai/api/v1 + VENICE_API_KEY - crof → https://crof.ai/v1 + CROF_API_KEY - bankr → https://gateway.bankr.bot/v1 + BANKR_API_KEY - cometapi → https://api.cometapi.com/v1 + COMETAPI_API_KEY Companion to the WebUI side (PRs #26/#27/#29/#30/#32/#33) which made these providers discoverable in the model picker with their live /v1/models catalogs. Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>

…(was lost in chain shift)

…on + dispatcher-only lookup Second self-review found one BROKEN issue and one WEAK adversarial concern. Both addressed. ## BROKEN #7 — same-tenant cross-umbrella collision Two goal_mode umbrellas sharing a tenant (the common marvel-swarm-* case) would both look up integrative reviews by the same title prefix + tenant. Umbrella B's archive could find umbrella A's review and either falsely unblock (if A was approved) or falsely block (if A was in flight). Fix: embed `umbrella_id` in the review title: ``Integrative architectural review (v6.7 NousResearch#30) for <umbrella_id>`` Lookup uses a per-umbrella LIKE pattern via two new helpers: `_v6_7_integrative_title_for(umbrella_id, round)` and `_v6_7_integrative_title_pattern(umbrella_id)`. Tenant is no longer load-bearing for isolation — the title is unique per umbrella. ## WEAK #8a — manual fake-review bypass mitigated A worker with kanban write access could create a task with the right title, mark it done with `result = "verdict: approve"`, and unblock the umbrella. Now the lookup query also filters `created_by = 'dispatcher'`, so hand-crafted fake reviews are ignored. This isn't a complete defense (a worker who can write arbitrary `created_by` values could still spoof) but defense in depth that catches the common case. ## Tests 3 new tests added (30 total, up from 27): - TestCrossUmbrellaIsolation.test_two_umbrellas_same_tenant_have_separate_reviews - TestCrossUmbrellaIsolation.test_approving_one_umbrella_does_not_unblock_another - TestFakeReviewMitigation.test_lookup_requires_created_by_dispatcher Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…ousResearch#30) archive_task gates JARVIS goal-mode umbrella archives behind a final T'Challa integrative architectural review. When all per-block reviews are terminal and ≥1 build child + ≥1 review child exist, the dispatcher spawns a PEER task (not child — that caused a deadlock in the first attempt) assigned to tchalla with the 4-item review scope: end-to-end request trace, page render trace, adversarial enumeration, error-path audit. Strict line-anchored verdict parser (^verdict: approve|reject). Reject verdict triggers re-spawn with :rN suffix the next archive call. Per-umbrella title isolation prevents same-tenant cross-umbrella collision. created_by='dispatcher' filter mitigates fake-review bypass. Closes hermes-jarvis#30. Context: hermes-jarvis#61. 30 tests after three self-review passes. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…h#79) (#19) * feat(kanban): v6.8 Part 4 — umbrella keep_running until reviewers spawn (NousResearch#79) The 2026-06-10 v6.7 validation chain caught this orchestration bug: JARVIS umbrella spawned Pepper + Friday from the umbrella, then called kanban_complete on itself before any reviewer was queued. The umbrella showed as done with build-only descendants; Kaipo had to manually spawn Tony / Tchalla / Vision to continue the chain. Closes hermes-jarvis#79. ## What this PR adds New gate ``verify_umbrella_review_coverage`` that fires when a goal_mode umbrella calls kanban_complete. The gate walks the umbrella's transitive descendants (reusing _v6_7_walk_descendants from NousResearch#73 so chained shapes count). If no review-role descendant exists anywhere in the subtree, the gate rejects. Two violation messages: 1. ``no descendants at all`` — pathological goal-mode umbrella that never decomposed. The orchestrator misfired. 2. ``build-role descendants but NO review-role descendants`` — the exact 2026-06-10 case. Message points the operator to spawn tony/tchalla/vision now via kanban_create --parent <build-id>. Opt-out: ``metadata={"x_no_review_needed": "<≥20-char reason>"}`` for legitimate cases (e.g. pure status-ack umbrellas). Standard v6.7/v6.8 opt-out conventions — emits completion_opt_out_used audit event with verbatim reason; bool/short strings rejected. ## Relationship to v6.7 NousResearch#30 - NousResearch#30 (integrative review at archive): fires at ARCHIVE time, spawns Tchalla after a chain settles. - NousResearch#79 (this PR): fires at COMPLETE time, forces JARVIS to spawn reviewers BEFORE marking itself done. Together they catch the empty-chain failure mode at both ends of the lifecycle. ## Tests 12 new tests: TestUmbrellaReviewCoverage (7) — non_goal skips, review descendant passes, only-build rejects with helpful message, no-descendants rejects with different message, tchalla/vision also satisfy, opt-out bypasses. TestUmbrellaReviewCoverageIntegration (5) — end-to-end via complete_task: build-only blocks, adding chained tony unblocks (closes NousResearch#73 + NousResearch#79 together), x_no_review_needed passes, short opt-out rejected, non-goal_mode tasks unaffected. 135 in test_kanban_completion_gates.py pass. 231/231 across full v6.7+v6.8 + adjacent regression set, zero failures. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(kanban): self-review fixes for v6.8 Part 4 umbrella gate (NousResearch#79) Three findings from the independent code review on PR #19: 1. Scope tightening: verify_umbrella_review_coverage now requires umbrella_assignee in ORCHESTRATION_ROLES (jarvis/pepper/banner). A goal-mode worker like Friday is out of scope — only orchestrators own umbrella-decomposition discipline. Was over-applying. 2. Empty-assignee message correctness: has_non_review_descendant is now bool(descendants), not the unset has_non_review flag. Prevents the "no descendants at all" message lying when descendants exist but all have empty assignees. Also added break after first review match for early termination. 3. Opt-out rename: x_no_review_needed → x_umbrella_no_review across gate messages, kanban_db opt-out validation, accepted_opt_outs audit entry, and tests. Avoids future collision with x_no_reviewer_fields (reviewer-field opt-out). Plus 8 new unit tests covering the scope tightening (friday/tony skipped, empty-assignee skipped, pepper/banner in scope, case- insensitive matching) and edge cases the reviewer flagged (blocked-status review counts, deep transitive review, multi-children multi-reviews). Module docstring updated: was "Three gates ship today (Tranche 1 of v6.7)" — now reflects the v6.7+v6.8 gate roster (5 gates, closing 7 issues). All 144 gate tests pass. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Jarvis <jarvis@Kaipos-Mac-mini.local> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…h#33 adopted; journal + v1.3.4 ATTEMPTED: - Ran the pre-registered NousResearch#32 gate (objective outcome harness, n=34 = 20 micro + 14 agentic, K=3, all-deepseek, --max-rounds 1, --strict-preflight; 170 cells / 52 min; raw ~/.hermes/outcome_eval_32.json) and applied the ADOPT rule mechanically. Journaled the verdict both ways across README §6, design-decisions (§First-order candidate source NousResearch#32 + §Discrimination preflight NousResearch#33), evsi-validation-findings (pre-registered rule quoted verbatim, incl. the efficiency ceiling), and the nbq-improve backlog. Bumped nbq 1.3.3 -> 1.3.4 (negative result, no default flip). WHY: - NousResearch#32 tested the NousResearch#28 successor hypothesis: the residual P4 gap over the naive zeroshot arm is generation altitude, not the value judge. First cost-aware gate: efficiency (wall/tokens/calls) measured alongside Δresult, with a pre-registered 10%-added-wall ceiling that vetoes even a result win. RESULT: - NousResearch#32: NO ADOPT. Per-arm pass vs baseline 0.460 -- nbq +0.083 (9W/4L, un 74%, 25.0s/20221tok/38.3c); nbq-firstorder +0.132 (7W/3L, un 78%, 29.2s/23944tok/45.4c); zeroshot +0.274 (15W/1L, p=0.0005, un 31%, 5.9s/154tok/1c). Paired nbq-firstorder vs nbq: Δ+0.049 > 0 but 6W/6L/22-tie (broad-win guard fails); unanswerable 77% (>50%); lens-payoff regression (log-clean 0.67->0.33); +16.8% wall (>10% ceiling). All four adopt conditions fail. Altitude has signal (mean beat plain nbq) but did NOT close the gap -- zeroshot still dominates at ~1/5 wall, ~1/150 tokens; the residual gap is ANSWERABILITY. firstorder stays built, off-by-default. NousResearch#30 answerability weighting re-opened (its condition, unanswerable >50%, is now met). - NousResearch#33: PASS/adopted as opt-in instrument -- discrimination_preflight fast 8/8, deepseek 8/8. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…retro probe ATTEMPTED: First real use of the nbq-improve protocol (iteration two) — a conditional full lap. Built a zero-model-call retro probe (evals/probe_answerability.py + 5 tests) to gate whether to build NousResearch#30 answerability weighting, testing NousResearch#30's premise against the existing objective corpus (outcome_eval_32.json: per-question EVSI meta.q_values, answerability qa[].revealed, outcome frac). WHY: NousResearch#32 pinned the residual P4 gap on answerability (unanswerable 77% > 50%), re-opening NousResearch#30. Rather than build first, probe the premise for free: do kept high-EVSI *unanswerable* questions cause objective failure? The cheapest falsifying test uses data we already have — zero new model calls. RESULT: PARK (premise NOT supported, n=34). top1_unans x fail r=+0.052, SE=0.177, does not clear SE (no association); any_unans x fail r=-0.112, base rate 0.971 => DEGENERATE (near-universal unanswerability => no answerable-question contrast to steer toward) + wrong-direction; tertiary n_unans x frac r=-0.219 (weak whiff only). Delta-result: none (no NousResearch#30 build; conditional gate honored in commit order). Delta-cost: zero (probe is offline). Verdict per the pre-registered rule (quoted verbatim in evsi-validation-findings.md): PARK. NousResearch#30 re-open now requires a higher-contrast corpus (candidate 2/3) plus a non-self-rated mechanism (batched strict-simulator answer/refuse probe, designed + parked in prereg-iteration-two.md item B, unbuilt). Methodology banked: cost is multi-dimensional — verdict-rubric.md now requires a per-axis ceiling (wall, tokens, calls), any bust vetoing a result win. Suite 206 -> 211 green. next-best-questions 1.3.4 -> 1.3.5; nbq-improve 1.0.0 -> 1.0.1. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…nt != state ATTEMPTED: Built an opt-in nbq-reach-investigate arm in outcome_eval.py (a fixture-aware mock_investigator resolving strict-unanswerable questions from observable state, leakage-guarded) + 6 tests + a --arm flag on probe_answerability.py. Gated on the agentic bank (n=14) to test candidate 3: does resolving reach questions via an investigator lift objective pass and create the answerability<->pass contrast NousResearch#30 needs? WHY: Iteration two parked NousResearch#30 because unanswerability was near-universal with no answerable-question contrast. Candidate 3 was the route to build that contrast — route reach questions to a (mocked) investigator that turns unknowns into observables. RESULT: NO ADOPT. 0 investigator resolutions across all 42 rows — the mock never fired (it works on observable questions; the build smoke resolved one). nbq's high-EVSI questions are about INTENT (which reading, crash-vs-fallback, detail level), and an investigator observes STATE, not intent. Unanswerable ROSE (78.6%->81.0%). The +0.100 arm-mean gap (baseline 0.351 / nbq 0.301 / reach 0.401) is unpaired sampling variance: 0/14 tasks shared questions across arms. NousResearch#30 did not un-park (all 14 tasks frac<1.0 => fail has zero variance => probe r=0.000). Delta-cost: 0 incremental (0 resolutions). THE FINDING: intent != state — the valuable clarifications are unobservable by any hop, answerable only by the user, so the answerability/reachability lever is a dead-end; value is in the unobservable intent questions. Route forward = candidate 2 (nbq->relentless, real user answers intent). Methodology banked to verdict-rubric.md: paired-design validity (test the answering, hold questions fixed). Suite 211->217 green. next-best-questions 1.3.5->1.3.6; nbq-improve 1.0.1->1.0.2. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

… learning even without code) ATTEMPTED: Codify into verdict-rubric.md's commit-message contract that git history is the experiment log: every lap commit cites the key git references its learning chains from, and the thesis+lesson are committed EVEN WHEN the experimental code is not kept. WHY: Per jim — commit history is the durable log. A negative result or a discarded experiment must not vanish because its code didn't ship; the learning and its provenance (prior SHAs) must persist in git even when the code does not. RESULT: verdict-rubric.md commit-message contract extended (code-optional, learning-required; cite prior-lap SHAs). nbq-improve 1.0.2->1.0.3. This commit is the first to cite the chain it builds on: lap one nbq 1.3.3->1.3.4 (baseline 9e15281) lap two canonical 2305998, hub 8165e32 (NousResearch#30 answerability PARKED; cost-is-multidimensional) lap three canonical 1d0291b, hub a52ac40 (reach->investigate NO ADOPT; intent != state) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…h#30)

…sResearch#30) The artifact tool whitelisted a fixed set of render dialects; HTML wasn't one, so the agent couldn't maintain an HTML-document artifact (a custom dashboard or styled report the structured kinds can't express). The gateway store already handles unknown kinds correctly — merge_content defaults to replace, which is exactly right for a whole HTML doc — so this only unblocks the tool surface: - Add "html" to VALID_KINDS and both schema enums. - Describe the kind: content is raw HTML (not JSON), no per-kind merge, so always write the complete document (get-before-write still applies). The native client renders it in its existing WKWebView-backed artifact view. Test: extend test_agent_tool_surface with an html set/get round-trip. Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

NousResearch#30) * feat(memory): memgw provider — Memory Gateway hybrid recall as default New plugins/memory/memgw/ MemoryProvider that connects Hermes to the self-hosted Memory Gateway over Streamable-HTTP MCP: - memgw_recall / memgw_retain / memgw_reflect tools - background prefetch (recall|reflect), non-blocking sync_turn - on_delegation -> Experience capture, on_session_end summary - circuit breaker + threaded event loop (mirrors mem0/hindsight patterns) - cloud (Bearer) or local (keyless) mode - set as Hermes default memory.provider (degrades to built-in if unavailable) - 13 tests passing, ruff clean; provider discovered + loads via plugin system Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * fix(memory/memgw): address Codex PR NousResearch#30 review findings - is_available() now checks for the mcp package before returning True, preventing ModuleNotFoundError on activation when credentials are set but the extra was never installed (P2) - pin pip dep to mcp>=1.26.0,<2 in plugin.yaml, matching the audited version in pyproject.toml (P2) - clear DEFAULT_CONFIG memory.provider so fresh installs don't default to an external provider the user hasn't configured (P1) - add on_session_switch() to clear _prefetch_result on session rotation so stale prefetch from a prior session can't bleed into a new one (P2) - pass user_id in all gateway calls (recall/write/reflect) when running in a multi-user gateway context, enabling server-side user scoping (P1) - remove the blocking join on the previous sync thread from sync_turn() so the turn path is never delayed by a slow/hung prior sync (P2) - track delegation threads in _delegation_threads list and join them in shutdown() so on_delegation writes aren't silently killed on exit (P2) - raise RuntimeError in _unwrap() when isError=True so circuit-breaker records the failure and tool calls surface the error to the model (P2) - guard _ensure_loop() with _loop_lock to prevent concurrent first-calls from racing and leaking a loop thread (P2) * fix(memgw): address Codex review — scoping, session/writer safety, MCP errors Addresses the 6 live findings on PR NousResearch#30: - multi-user scoping: recall/write/reflect payloads now include user_id (_user_scope) so gateway sessions don't share one namespace - MCP client: _unwrap raises on CallToolResult.isError so tool-level failures don't reset the circuit breaker as successes - is_available checks the optional mcp dependency is installed - prefetch: monotonic generation guard so a slow older worker can't overwrite a newer result; on_session_switch clears + invalidates in-flight prefetch - sync writers tracked in a list; shutdown joins all (not just the last) so an in-flight retain isn't dropped - delegation threads tracked + reaped - doc: README explains the deliberate bundled-plugin exception (personal default for this fork; degrades to built-in when mcp/key absent) - +4 tests (scoping, session-switch invalidation, stale-prefetch guard) 17 memgw tests pass, ruff clean Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

… — Codex PR NousResearch#30 review (NousResearch#31) Two chatgpt-codex-connector[bot] review comments on PR NousResearch#30 that still applied to current main (the other 13 were already addressed in the merged PR): 1. client.py NousResearch#13 — Cancel timed-out MCP calls: _run_sync now cancels the concurrent.futures.Future on timeout before re-raising, so a stalled MCP endpoint doesn't leave a pending HTTP session running on the shared background loop after the caller has given up / opened the breaker. 2. __init__.py NousResearch#15 — Parse the host before allowing keyless local mode: is_available() now urlparse's the URL and requires an exact loopback host (localhost/127.0.0.1/::1) instead of a substring match, so a URL like 'https://localhost@example.com/mcp' (host=example.com) is no longer trusted as local keyless mode. Adds regression tests: test_memgw_client_timeout.py (cancel-on-timeout fails against pre-fix client) + TestKeylessLocalModeHostParsing (3/4 fail pre-fix). Deferred (already fixed in merged PR NousResearch#30): MCP dep import check, MCP dep pin, loop lock, isError handling, stale-prefetch generation, sync/delegation thread tracking + shutdown join, session-switch prefetch invalidation, no join on turn path, default provider unset. Deferred (complex, needs interface change): NousResearch#3/NousResearch#4 per-turn user_id refresh for shared gateway sessions — sync_turn/prefetch only receive session_id, not user_id, so threading identity through requires a MemoryProvider interface change; NousResearch#1 in-tree placement is an architectural call (AGENTS.md says existing in-tree providers stay).

…odex P1) (NousResearch#33) * fix(memgw): pass user_id per-call to prevent cross-user memory scoping in shared gateway sessions Closes the remaining open P1 Codex finding from the PR NousResearch#30 review chain. In shared gateway sessions (thread_sessions_per_user=False), multiple users share a cached AIAgent instance. _user_id was stored once at initialize() time and reused by _user_scope() for all subsequent sync_turn/queue_prefetch calls, routing User B's memories into User A's gateway namespace. Fix: propagate user_id as an optional keyword argument through the full call chain: run_agent._sync_external_memory_for_turn -> MemoryManager.sync_all / queue_prefetch_all -> MemoryProvider.sync_turn / queue_prefetch (base interface updated) -> MemGatewayProvider._user_scope(user_id) -- per-call override wins, falls back to self._user_id for non-gateway single-user sessions _user_scope() now accepts an explicit uid that takes priority over the cached self._user_id, so every background write and prefetch is scoped to the user who actually triggered the turn, not the user who first initialized the session. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NFLEjSSx3Wr2j4D3v4HNN4 * fix(ty): add user_id kwarg to queue_prefetch/sync_turn overrides The base class MemoryProvider now declares user_id as a keyword-only default arg on both queue_prefetch and sync_turn. Seven providers and one test stub that override these methods were not updated in PR NousResearch#33, causing ty to report invalid-method-override (17 new diagnostics). Add user_id: str = "" to each override; providers that do not use multi-user scoping can safely ignore it. Also bound pytest-timeout in [dependency-groups] per supply-chain policy: >=2.4.0,<3. * chore: regenerate uv.lock after pytest-timeout upper-bound pin The previous commit added <3 upper bound to pytest-timeout per repo supply-chain policy, but did not regenerate uv.lock. Running `uv lock` updates the lockfile to reflect the new constraint. * fix(ty): add user_id keyword arg to retaindb overrides The base MemoryProvider.sync_turn and .queue_prefetch gained user_id: str = "" as a keyword-only default, but these overrides were missed in the previous commit. Fixes 5 remaining invalid-method-override ty warnings. * fix(ty): add user_id keyword arg to mem0 overrides The base MemoryProvider.sync_turn and .queue_prefetch gained user_id: str = "" as a keyword-only default, but these overrides were missed in the previous commit. Fixes 5 remaining invalid-method-override ty warnings. * fix(ty): add user_id keyword arg to agent overrides The base MemoryProvider.sync_turn and .queue_prefetch gained user_id: str = "" as a keyword-only default, but these overrides were missed in the previous commit. Fixes 5 remaining invalid-method-override ty warnings. --------- Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

… Sprint NousResearch#30 evidence: none verified: - zentao story list --product=11 → 4 stories (NousResearch#117-120) - zentao task list --execution=30 → 13 tasks (NousResearch#45-57) - http://pm.test.com/zentao/story-browse-117.html Changed: - .zentao/mapping.json: winpeek#6 → peeka数字机器人NousResearch#11, SCRUM NousResearch#29, Sprint NousResearch#30 - tasks/mim-chat-v1-*: updated zentao IDs NousResearch#32-44 → NousResearch#45-57 - tasks/README.md + CLAUDE-module.md: zentao URLs/commands Co-Authored-By: Claude <noreply@anthropic.com>

P3-M1 合入列车阶段 4： - fork PR #4 base 已由 GitHub API 从 feature/p3-m1-provider 改为 main （main 现含阶段 2 rebrand merge e3121a2 + 阶段 3 provider merge 7f9d1c9） - HA @ pinned SHA 由 PR NousResearch#30 旧 HEAD ad44bd11 改为阶段 1 真实 merge SHA a3ef38298e6a42a88ec93d75af88b4027f7febbe（HA 默认分支 claude/hermes-desktop-multi-ai-phone-aiw5mr 新 HEAD）仅改 pin / provenance 相关位置（Codex 硬约束）： - .github/workflows/p3-m1-mcp-seven-step.yml: env HERMES_AI_PIN_SHA + actions/checkout ref（两处 SHA） - tests/native_mcp/test_seven_step_via_registry.py: module docstring Prereq 段 SHA 说明（1 处） - docs/p3-m1-mcp-seven-step-fork.md: Scope 段（PR NousResearch#30 已 merged @ merge SHA） + Codex E 表 Step 1 说明（`importlib.metadata` 断言 + `git rev-parse HEAD` 硬闸，删除旧的 `pip show grep` 描述）+ 关联 PR 堆叠段（阶段 2/3/1 merge SHA 全部记录）七步测试逻辑零改动。生产代码零改动。uv.lock/pyproject.toml 零改动。本地: 1 passed in 5.06s。Secret 扫净。等专用 push+PR CI + umbrella CI + 原三审计 Agent 只读复审 base 变化和 SHA pin 变化。

P3-M1 合入列车阶段 4。fork 自己生产路径 config.yaml → tools/mcp_tool.py::discover_mcp_tools() → 两 stdio server → tools/registry.py::registry.dispatch("mcp__<server>__<tool>", args) 完成真实七步链路，含 tenant/thread 权威源锁定、strip_sources C4、send_reply 幂等、严格 shutdown 与后台 mcp-event-loop 线程硬门。 HA @ pinned merge SHA a3ef38298e6a42a88ec93d75af88b4027f7febbe（阶段 1 HA PR NousResearch#30 merge commit）via actions/checkout + HERMES_AI_TOKEN （fine-grained PAT，仅 HA Contents: Read-only）+ persist-credentials: false + Verify HEAD == HERMES_AI_PIN_SHA 硬闸。依赖全部由 fork uv.lock 锁定：mcp==1.28.1 / httpx==0.28.1 / websockets==15.0.1 只断言不重装。 CI (HEAD 1c23846)： - 专用 push run 31611539996 GREEN - 专用 PR run 31611537510 GREEN（tests=1 passed=1 skipped=0 failures=0 errors=0） - umbrella CI 31611538177 attempt 2 GREEN（含 12/12 Python slices + Review label gate + All required checks pass 全 success） 3 原审计 Agent 对 base+pin 变化只读复审全部 PASS · blockers []。 ci-reviewed 标签由 Codex 授权添加，触发 Review label gate GREEN。

docs: technical memory architecture + landing page clarity

b406c03

UfukNode closed this Feb 25, 2026

UfukNode reopened this Feb 25, 2026

UfukNode closed this Feb 25, 2026

SHL0MS mentioned this pull request Mar 30, 2026

[Bug] Thread safety: unprotected shared mutable state between UI and agent threads #4072

Closed

kshitijk4poor mentioned this pull request Apr 15, 2026

feat: add TouchDesigner integration skill (twozero MCP) #10081

Closed

ashneil12 mentioned this pull request May 18, 2026

fix(auth): register Venice/Crof/Bankr/CometAPI in PROVIDER_REGISTRY ashneil12/vanilla-hermes-agent#16

Merged

3 tasks

chungty mentioned this pull request Jun 2, 2026

[Z2O-1694] fix(mcp): resolve ${context:} headers caller-side, not on the MCP loop VerdigrisAI/hermes-agent#2

Merged

dmascord added a commit to dmascord/hermes-agent that referenced this pull request Jun 6, 2026

Fix chain: add groq at NousResearch#13, gemini-lite at NousResearch#30 …

897c6e2

…(was lost in chain shift)

This was referenced Jun 9, 2026

v6.7 Tranche 1: kanban_complete verification gates (#28, #62, #64) #43104

Closed

v6.7 Tranche 1: kanban_complete verification gates (#28, #62, #64) 1Team-Engineering/hermes-agent#11

Merged

This was referenced Jun 11, 2026

v6.8 Part 1: integrative-review gate walks transitive descendants (#73) 1Team-Engineering/hermes-agent#16

Merged

v6.8 Part 4: umbrella keep_running until reviewers spawn (#79) 1Team-Engineering/hermes-agent#19

Merged

Soju06 added a commit to Soju06/hermes-agent that referenced this pull request Jul 16, 2026

docs(patches): register session-db-read-path-split (patch NousResearc…

b18f240

…h#30)

Soju06 added a commit to Soju06/hermes-agent that referenced this pull request Jul 16, 2026

docs(patches): update NousResearch#30 commit hashes after stack merge

f8dcb6e

bri12afdsarker96-lgtm mentioned this pull request Aug 12, 2026

P3-M1 MCP 七步验收 · 走 fork 原生 registry.dispatch（stacked on Provider PR #3） bri12afdsarker96-lgtm/hermes-agent-mercury#4

Merged

Navigation Menu

Uh oh!

docs: technical memory architecture + landing page clarity - #30

docs: technical memory architecture + landing page clarity#30
UfukNode wants to merge 1 commit into
NousResearch:mainfrom
UfukNode:docs/technical-memory-architecture

UfukNode commented Feb 25, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

UfukNode commented Feb 25, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants