I've completed a comprehensive root cause analysis and created a detailed implementation plan for issue #441. Here's the summary:

The ruflo_with_timeout function in scripts/lib/ruflo-adapter.sh:404 runs background processes without setsid, then uses pkill -P to kill them on timeout. The problem: pkill -P only kills direct children, not grandchildren. When the ruflo binary spawns Node processes that fork/exec further, those grandchildren persist as orphaned processes (~385 accumulating).

1. Current Implementation (line 404):

   ( "$@" ) >"$_rft_tmp" & # No setsid — no process group isolation

2. Inadequate Timeout Kill (line 416):

   pkill -TERM -P "$bg_pid" # Only kills direct children, not grandchildren

3. Missing Fallback: No SIGKILL escalation if TERM doesn't work.

Use process groups (setsid) + two-phase kill:

• Create new session group with setsid (line 404)
• Capture process group ID (new lines after 405)
• Phase 1: Kill entire group with kill -TERM -- -$bg_pgid
• Phase 2: If still alive after 1s, use kill -KILL -- -$bg_pgid

Why this works: Process groups kill all descendants at once, not just direct children.

See TodoList for tracking. Tasks flow as:

• Tasks 1-4: Code changes (sequential)
• Tasks 5-7: Testing (parallel after code)
• Tasks 8-10: Validation (parallel after tests)

✅ pgrep -fc "ruflo memory" ≤ 2 after 10 iterations
✅ All tests pass (npm test)
✅ Memory leak eliminated (385 procs → 0)
✅ Code documented with comments explaining setsid + process group kill

1. scripts/lib/ruflo-adapter.sh — Update ruflo_with_timeout (lines 404-428)
2. scripts/test-ruflo-leak.sh — Create new test script

Ready to proceed to build stage.