operator
26 lines
1.2 KiB
Plaintext
26 lines
1.2 KiB
Plaintext
# 002: Reconcile runbook
|
|
|
|
Goal: future operators can run the fleet without asking.
|
|
|
|
Write `docs/RUNTIME-RECONCILE.md` (under 80 lines): manifest format
|
|
(fleet/agents.json), brief workflow (fleet/briefs/), claim protocol
|
|
(pending/claimed/done + heartbeat touch + owner suffix), what
|
|
`box runtime reconcile` enforces, and crash-restore behavior (dead
|
|
server reads as all sessions missing; stale claims re-queue).
|
|
|
|
Base it on the real code (bin/runtime_reconcile.py, fleet/agents.json,
|
|
fleet/briefs/) and verify every command you document by running it.
|
|
|
|
Done criteria: doc exists, is accurate, and every documented command
|
|
was executed successfully.
|
|
|
|
Result notes (append below before moving to done/):
|
|
|
|
- 2026-10-07, builder muse--runtime--roles: wrote docs/RUNTIME-RECONCILE.md
|
|
(52 lines). Every documented command executed OK: `box runtime list
|
|
--socket /tmp/tmux-muse.sock --muse-only`, `box runtime reconcile`,
|
|
`--dry-run` (both agents ok), `--adopt` (both live, briefed), claim
|
|
mv + touch heartbeat. Stale-claim sweep verified on scratch queues
|
|
(owner-gone and 45-min-TTL paths both requeue; live+fresh claims kept).
|
|
Live reconcile also STARTED watchers on %20/%21. No code changes.
|