Worktree cleanup leaves stale heartbeat worktrees active indefinitely

[chubes4]

Summary

Workspace cleanup is reporting hundreds of active_no_signal worktrees as if they are still in progress. On this machine the report showed 408 active/no-signal rows, which is not a credible real active workload count.

Evidence

• Command: studio wp datamachine-code workspace worktree active-no-signal-report --limit=25 --offset=0 --until-budget=60s --format=json
• Observed summary: total_active_no_signal: 408
• Sample rows had stale lifecycle metadata, e.g. liveness_reason=heartbeat_stale, last_seen_at days old, no session ids, and often no PR/task metadata.
• Many rows are dirty or have unpushed commits, so deletion must remain guarded, but they should not remain classified as active/in-progress indefinitely.

Expected behavior

Stale heartbeat worktrees should transition out of generic active lifecycle or into a distinct review/remediation bucket after a bounded TTL. Cleanup/hygiene should surface them as abandoned/stale-review candidates, not count them as active in-progress work forever.

Actual behavior

Rows with lifecycle_state=active and stale heartbeats remain active_no_signal indefinitely. This makes cleanup output misleading and hides a lifecycle accounting bug behind safety blockers.

Impact

• Cleanup reports imply hundreds of active local tasks.
• Disk cleanup cannot make progress without manual interpretation.
• Operators may be tempted to force-remove dirty worktrees because the lifecycle state is not trustworthy.

Suggested fix

Add a lifecycle reconciliation path for stale heartbeat active worktrees:

• Detect lifecycle_state=active plus liveness_reason=heartbeat_stale beyond a TTL.
• Move them to a non-active review state such as abandoned_review, stale_active, or an explicit cleanup triage bucket.
• Preserve dirty/unpushed safety gates for deletion.
• Update cleanup summaries so stale-active rows are counted separately from truly live active work.

AI assistance

• AI assistance: Yes
• Tool(s): openai/gpt-5.5 via OpenCode
• Used for: Investigation summary and issue drafting.