rdmsm4x changelog 2026-09-24 19:30 — MCP/tooling + fleet-ops pass (grok0924)

This pass fixed ticket-mcp on all six hosts, fixed Tyrell SelfBootstrap paths (merged ca47653), fixed three failing hub launchd jobs, reconciled LANES.md, re-measured FLEET.md, corrected Tyrell build status, triaged the bus, made bus sends atomic (v1.5.2), and diagnosed the hub load.

Written 2026-09-24 19:30 EDT by grok@rdmsm4x/grok0924, working from rdmbair15m5; builds ran on rdmbair13m5.

Scope

• Rich's instruction: "resolve everything you see wrong".
• My lane: MCP/tooling and fleet-ops.
• Excluded: the ECSMemory/mem0 lane (ecsmem24), RTTy and its 523 gate (c13f617a), netdoc-0924 (devbackup, DOC-20260905-01), the LogTTY agy branch, and scanRoo native-camera.

Changes

1. ticket-mcp (ISSUE-20260924-25, RESOLVED)

• Server: ticket-mcp 3.2.0 adds initialize and ping, and no longer replies to notifications.
  • Committed on issues branch grok/ticket-mcp-initialize-ISSUE-20260924-25 (6bb4887), merged to hub main as 7b7ce16.
  • Not pushed; per the repo flow the issues repo is pushed by its own job.
• Reconciler: fleet_mcp_reconcile.zsh v1.7 (fleet/maintenance aeb0070):
  • distributes ~/bin/ticket-mcp only after it passes a handshake;
  • repairs broken Codex and agy registrations.
• Result: claude mcp list shows ticket-mcp, tyrell, replicantdb and mem0-shared Connected on all six hosts.

2. Tyrell SelfBootstrap paths (TASK-20260924-14, in-progress: needs a signed build)

• Branch grok/mem0-shared-path-ISSUE-20260924-24, commits 61bc9bc, 4aa9f1e and bc55248.
• configureAgentHarnesses now resolves each server to the first existing candidate, keeps or repairs registrations, and gates firecrawl on npx.
• Merged --no-ff to Tyrell main as ca47653 and pushed.

3. Hub launchd jobs

• certsync: ~/scripts/fleet_cert_sync.zsh v1.1.0.
  • An asleep host with a verified, unchanged cert is now WARN rc 0.
  • Exit codes are unchanged.
  • Backup: .bak-grok-20260924-1846. Re-run: 6/6 PASS, rc 0.
• nightlyaudit: ops/audit/nightly_audit.zsh v1.5 (fleet 7ee9667).
  • CHECK 1 retries once after a timeout.
  • CHECK 12 now parses only the ## Live lanes table and reports stale lanes.
• devsite: ~/dev/scripts/publish_site.zsh v1.2 (a0bac97).
  • Every stage uses timeout -k 30, and gen_site runs before the heavy stages.
  • The stale v1.1 run (pid 95904; its child date had been stuck in R for 33 min) was stopped at about 19:25 with launchctl kill SIGTERM on its own job.

4. LANES.md

• Reconciled against the live claims (issues a89d7c2), with Live, Dead and History sections.
• Backup under ~/dev/_handoff/grok-fleetops-20260924/backups/.

5. FLEET.md

• §1 host table re-measured live (fleet/maintenance bce376b; fleet 6a3976d).
• Propagated with sync_fleet_knowledge.zsh.

6. Tyrell build status

• All six hosts are on 0.2.0 (32).
• SESSION-STATE: prepended a new checkpoint correcting the stale "Build 31" line. STATUS.md: updated the soak row. Both are in bc55248.
• ~/dev/CLAUDE.md ticket counts refreshed. Backup taken; docs synced.

7. Bus

• Triage report: ~/dev/_handoff/grok-bus-triage-20260924/.
  • 1,538 high-priority messages with no ack: 33 undeliverable, 627 superseded automated reports, 878 that need their recipient.
• The 2 malformed 33-byte stubs were moved to ~/.agent-coordination/archive/bus-malformed-20260924/ on all six hosts.
  • The complete copy of 20260827-125805-1B2875CF survived on rdmpw3265m and has re-propagated.
• agent_msg.zsh v1.5.2 (atomic send):
  • Commits: fleet/maintenance 97abb76 (records the deployed v1.5.1) and e40c270.
  • Test: test_agent_msg.zsh v1.1.0 passes 52/52; v1.5.1 fails the new §11 case.
  • Installed on all six hosts, sha-verified (590eeefc…), with a v1.5.1 backup agent_msg.zsh.bak-v1.5.1-grok-20260924 beside each.
  • Installed on the spokes by hand, following the reconcile procedure, because configsync pid 36992 is wedged (state R for 52 min).

8. Hub load

• ISSUE-20260905-08: added two comments with measurements, recommendations and the ssh MaxSessions finding (rdmpw3275m's mux is at 10/10).
• Opened incident ISSUE-20260924-33: process-table enumeration (ps -ax, top) hangs from about 19:15, and load reached 400–565.

10. Other findings

• Filed ISSUE-20260924-31 (firecrawl/npx) and ISSUE-20260924-32 (host-dependent Tyrell test).

Verification

• claude mcp list on 6/6 hosts.
• ticket-mcp tests 31/31 plus the adversarial suite.
• Tyrell: MockOnboardingTests 10/10; 988/988 Swift Testing; XCTest 415 with 1 host-dependent failure (ISSUE-20260924-32).
• certsync rc 0.
• CHECK 12 on the new LANES.md flags only c13f617a.
• agent_msg tests 52/52; sha parity on 6/6 hosts.
• FLEET.md checked on jdmbair13m5.

Not changed

• No agent sessions were killed or restarted. No reboot. fileproviderd, ecsmem0d and tyrelld were not restarted.
• Other agents' mail was not acked, taken or moved.
• devbackup was left to netdoc-0924.
• The uncommitted ~/dev/lib/domains/changes/dns-migration-2026-09-24/ directory (the reason sharedtreeguard exits 1) belongs to its owner, so I reported it rather than touching it.
• Not touched: arista/, permissions.defaultMode, the UNATTENDED/BUDGET-HOLD markers, the firewall and tccutil.
• The Apple Notes entry for this changelog is pending a GUI-session agent (same as bus message 20260924-154321-2BE3CAE2, which is still untaken).