Pantheon Recovery Board - 2026-05-23

Summary

Pantheon 재시동 후 생산성 순환이 아직 불안정하므로, 새 실행을 늘리기보다 런타임 병목, 열린 PR, 실패 라벨, 워크트리 잔여물을 먼저 정렬한다.

Snapshot: 2026-05-23 14:10 KST

Current Runtime

Signal	Status	Note
`run.sh`	running	Pantheon bridge process active
`bridge.py`	running	Slack bridge active
active agent command	running	`claude -p /spec HAN-326 --output-format json --permission-mode bypassPermissions`
immediate posture	hold new autonomous starts	HAN-326 spec run completes or blocks before adding more load

Recovery Order

Rank	Area	Action	Why
1	Runtime blocker	Let HAN-326 finish, then inspect result first	Runtime wrapper auto-restart failure is the highest leverage blocker for circulation
2	Pipeline repair PRs	Review/merge PR #118, #124, #120 before feature PRs	Execution cleanup and Spec Gate recovery improve the system’s ability to resume work
3	Generated PR backlog	Review PR #121 and #123 after pipeline repair	These unblock completed feature work without adding new scope
4	Failed-state cleanup	Reconcile `failed` labels on HAN-157, HAN-300, HAN-252, HAN-122, HAN-147, HAN-280	Failed labels should mean actionable failure, not stale routing residue
5	Worktree hygiene	Retire stale/gone worktrees and inspect HAN-289	Reduces operator confusion and accidental continuation from stale branches
6	Deferred feature work	Hold HAN-130, HAN-142, HAN-110 until recovery queue clears	Avoids expanding WIP while the loop is still recovering

Open PR Queue

PR	Issue	Title	Mergeability	Recovery Role
#118	HAN-320	gh pr create uncommitted changes + worktree cleanup retry fix	mergeable	pipeline repair
#120	HAN-323	`!recover-onhold` ON_HOLD Spec Gate reposting	mergeable	spec recovery
#121	HAN-158	auto feature PR	mergeable	feature backlog
#123	HAN-82	auto feature PR	mergeable	feature backlog
#124	HAN-321	Jarvis restart catch-up for unanswered Spec Gate	mergeable	spec recovery

Linear Queue Signals

Issue	Priority	Status	Labels	Recovery Reading
HAN-326	High	Backlog	implementing	Current active spec run; inspect before new execution
HAN-320	High	In Progress	Bug	PR #118 should be reviewed early
HAN-321	Medium	In Progress	Feature	PR #124 supports restart recovery
HAN-323	-	-	-	PR #120 supports manual ON_HOLD recovery
HAN-82	High	In Progress	pr-created, Frontend-Watch	PR #123 waiting
HAN-158	High	In Progress	pr-created	PR #121 waiting
HAN-289	High	Backlog	implementing	Worktree is behind origin/main and has untracked agent doc
HAN-157	High	In Review	failed, Bug	Needs failed-state reconciliation
HAN-300	High	Backlog	failed	Likely strategy work, not runtime failure; reconcile label
HAN-252	High	In Progress	failed	Confirm current owner/state
HAN-122	High	Backlog	failed	Confirm whether stale failure or real blocker
HAN-147	High	Backlog	failed	Confirm whether stale failure or real blocker
HAN-280	High	Backlog	failed, Feature	Confirm whether stale failure or spec/design blocker

Worktree Hygiene

Worktree	Branch State	Action
`exec-HAN-326`	active, untracked `.agent/drafts/`	Leave untouched while spec command is running
`work-HAN-289`	behind `origin/main` by 21, untracked `.agent/docs/HAN-289.md`	Inspect after runtime recovery; do not continue blindly
`work-HAN-135`	upstream gone	Candidate for cleanup after confirming no local value
`work-HAN-64`	upstream gone	Candidate for cleanup after confirming no local value
`work-HAN-65`	upstream gone	Candidate for cleanup after confirming no local value
`work-HAN-66`	upstream gone	Candidate for cleanup after confirming no local value

Operator Checklist

Do not start new autonomous implementation while HAN-326 spec command is still running.
When HAN-326 exits, capture whether it produced a spec, blocker, or tool failure.
Review PR #118 first, because it repairs execution cleanup and retry behavior.
Review PR #124 and #120 next, because they improve restart/spec-gate recovery.
Review PR #121 and #123 only after recovery PRs are either merged or explicitly deferred.
For each failed Linear issue, decide: real active failure, stale label, or needs new owner/comment.
Clean stale worktrees only after active runs are complete and untracked files are triaged.

Notes

This board is a recovery snapshot, not a new long-lived process. Its job is to make the next few operator moves obvious while Pantheon circulation stabilizes.