docs(agnote): release the two dangling B850 lanes — Crush awakening + doc-audit

pmoves/docs/AGENTS/AGNOTE4482PHI.t1.md

@@ -2241,3 +2241,7 @@ b8cea26c8\ that added it to the top-level equired\ array). (2) PMOVES-pinokio PR
 - `2026-08-26T22:45:00Z` CLAIM+RELEASE `SPARK-KIMI (Crush, GLM via qwen3-coder)` scope: **integration research + a0-archon-bridge skill authored (PR #2783) + live NATS/CHIT audit.** Research findings (all live-probed): (1) connector v1 surface completed — log_tail events[] shape captured, pause/token_status need context_id, model_presets exposes the tensorzero routing; (2) **Archon is REST-only** by fleet decision (#2303): /api/health rich + /health simple, :3090/:3737/:8091 same service, no MCP; (3) **NATS truth audit** (via :9223 — the audit script's 8222 default produces all-orphans, B850's container-port trap): core bus LIVE (mesh.node.announce heartbeats observed, pmoves.agent.task.v1 dispatch proven), JetStream has 9 streams but ONLY GEOMETRY_CGP carries consumers — ARCHON/BOTZ/ROOMS/CONTENT/TOKENISM/MESH_GPU/HELPDESK/COMFY streams are 0-msg/0-consumer dormant, and chit.> + agent.graphiti.signed.v1 have NO stream at all (docs-only, publishing = fire-into-void). **Skill delivered**: .claude/skills/a0-archon-bridge/SKILL.md — connector v1 (29 handlers+shapes), wrapper contract, Archon REST, NATS/CHIT live state, gotchas (9223, TZ no-auth, /api/ prefix vs fork docs), validation one-liners. PR #2783 (rebuilt clean off main after stale-worktree-history conflict). **Holds**: #2783 + #2780 both mergeable except claude-review = required + rate-limited (operator-diagnosed, self-clearing); pin-check flaked again on #2783 (2nd occurrence — rerun-green both times; flake-class finding for CI lane). **Routed**: dormant JetStream streams need an owner decision (consumers or de-provision); agent.graphiti.signed.v1 emitter wiring; log_tail consumer mapping in wrapper. CHIT trail unsigned-local. `agent_signature: ACK::SPARK-KIMI::A0-ARCHON-BRIDGE-SKILL::2026-08-26`.
 
 - `2026-08-26T23:30:00Z` CLAIM+RELEASE `SPARK-KIMI (Crush, GLM via qwen3-coder)` scope: **PMOVES.YT playlist check + FlOO$ pr-monitor flow run + own-PR findings closed.** (1) **Latest playlist additions (DARKSXIDE AI Playlist "ai")**: extracted via the pmoves-yt runtime (yt-dlp 2026.07.04) — 20 entries, newest-first: **"DeepSeek's New AI System Shouldn't Be Possible" (Two Minute Papers, uploaded 2026-08-26 — TODAY)** and **"The New Ontology Kernel Behind Reliable AI" (Discover AI, 2026-08-26 — TODAY)** at positions 1-2 = just added; then Titanic-econ doc (2026-08-16), Omarchy, CATL batteries, Palantir-OS, et al. (2) **Routed finding — monitor is silently blind on YouTube**: channel-monitor image carries yt-dlp **2024.10.07** → flat extraction returns **0 entries** → logs "No new videos" (a LIE that reads healthy). The pmoves-yt service (2026.07.04) extracts fine, so the break is version skew; #2713 (bump to 2026.8.19) is the vehicle but only covers pmoves-yt — channel-monitor needs the same bump (own lane). Also fixed the shared cookies-dir permission (chmod 777, gitignored dir; the B850 cookie-writer group-ownership finding hit on SPARK — both channel-monitor and pmoves-yt were dying on it). (3) **FlOO$ flow executed**: pr-monitor → learnings catalog → CHIT packet → strict gate. Strict correctly RED: remaining actionables are other lanes' (#2755:3, #2757:4, #2766/#2777/#2778:1 each — routed to authors). **Own-PR findings closed**: #2780 P1 resurfaced legitimately (my tier-example fix would have overridden the NEW connector default with the raw path — env.tier-agent.example now aligned to the connector path, 4fedf047c); #2783 P1#2 valid (skill wrapper section now gates on #2780 + gives pre-merge env, d892bdb75); #2783 P1#1 REBUTTED with evidence (agent_task_subscriber.py IS on origin/main via #2693 — bot searched wrong; reply posted). CHIT trail unsigned-local. `agent_signature: ACK::SPARK-KIMI::YT-PLAYLIST-FLOOS-SWEEP::2026-08-26`.
+
+- `2026-08-26T21:10:00Z` NOTE `B850-CLAUDE (Knuckles)` scope: **Historical closeout record (not a lane release — both lanes were ALREADY closed in the hook's canonical view).** Register-hygiene review correction: `open_claims_in()` now canonicalizes owner identities (post-#2760 lineage work), so the 2026-07-12 `CRUSH-GLM52 (Knuckles)` awakening CLAIM (PRs #2103-#2105, merged) and the 2026-05-15 doc-audit CLAIM (PR #1496, merged) were both already cleared by later same-canonical-identity releases — an exact-string audit (this session's first pass) misread them as open. Recorded so the next exact-string audit does not repeat the misread: **audit lane balance through the hook's canonical parser, not raw string matching.** The earlier draft of this row taught the pre-#2760 exact-key rule as current; that rule is obsolete. agent_signature: `ACK::B850-CLAUDE::LANE-HYGIENE-CORRECTION::2026-08-26`.
+
+<!-- GRAPHITI_MARK: B850-CLAUDE::LANE-HYGIENE-DUAL-RELEASE::2026-08-26 -->