docs/WAR-ROOM.md ships as scripts/warroom-loop.sh with a .warroom/ state dir — a bash loop implementing research → spec (SPEC.md AC checklist) → per-AC devagent task --auto-pr → evidence-judged completion, with guardrails inherited from the selfbuild loop. Comparable loops (pane-run, board-recovery) are typed, tested Go commands.
Gaps in the current form: bash-level state with no tests, WARMROOM_* env var misspelled throughout (doc + script), judge/AC-tick loop never validated against the herdr pane model, no structured ledger rows for war-room phase transitions.
Done when: devagent warroom exists in the Go CLI (research/spec/execute phases, resumable state, circuit breaker + starvation gate on par with selfbuild), ledger rows for phase transitions, the env spelling fixed, and scripts/warroom-loop.sh reduced to a thin caller or deleted.
docs/WAR-ROOM.md ships as
scripts/warroom-loop.shwith a.warroom/state dir — a bash loop implementing research → spec (SPEC.md AC checklist) → per-ACdevagent task --auto-pr→ evidence-judged completion, with guardrails inherited from the selfbuild loop. Comparable loops (pane-run, board-recovery) are typed, tested Go commands.Gaps in the current form: bash-level state with no tests,
WARMROOM_*env var misspelled throughout (doc + script), judge/AC-tick loop never validated against the herdr pane model, no structured ledger rows for war-room phase transitions.Done when:
devagent warroomexists in the Go CLI (research/spec/execute phases, resumable state, circuit breaker + starvation gate on par with selfbuild), ledger rows for phase transitions, the env spelling fixed, and scripts/warroom-loop.sh reduced to a thin caller or deleted.