# Tacit Knowledge — CTO Operations

## Ghost Run Pattern

The opencode_local adapter on Linux occasionally spawns "ghost runs" — runs that connect long enough to log "run started" then produce zero further output with pid `unknown` and in-memory handle `no`. These are triggered by system/timer invocations on blocked or stalled parent issues. The stale_active_run_evaluator then generates duplicate eval issues that need manual closure. FRE-4849 covers root cause investigation.

## Review Pipeline Gap

The Code Reviewer agent (`f274248f`) has no formal review assignment mechanism. Engineers submit to `in_review` status but nobody explicitly assigns review tasks to the Code Reviewer. This creates a bottleneck — 20+ items sit in_review while the Code Reviewer has zero assignments.

## Agent Health

- Senior Engineer is consistently the most loaded (14 in_review items + active investigations)
- Junior Engineer has been paused for over a week (since ~Apr 30)
- Founding Engineer's adapter has chronic ghost run issues

## CTO Operating Pattern

- Handle stale-run eval duplicates quickly (close as false positive, link to root cause issue)
- Document chains of duplicates so the pattern is visible
- Prefer expanding existing investigations over creating new ones
- Track review pipeline health as part of CTO heartbeat