Skip to main content

pi-orchestrator: john H-task auto-pause root cause + #10063 reconcile

pi-orchestrator: john H-task auto-pause root cause + #10063 reconcile

Created: 2026-05-02
MC References: #10063 (phantom fix), #10517 (true fix)
Daemon: com.john.pi-orchestrator (currently STOPPED, reactivation pending CEO Step 3)

Symptom

John's H-priority tasks were being auto-paused without user action. The pi-orchestrator daemon would intercept high-priority john tasks and route them through queueForHuman instead of executing them, creating a silent work-stoppage pattern.

Investigation Finding — Phantom Fix in MC #10063

MC #10063 (2026-04-XX) claimed to fix the auto-pause behavior by adding configuration flags:

  • skip_interactive_owners: ["john", "alem"]
  • interactive_grace_seconds: 300

Problem: These config keys were specified in the task's acceptance criteria and marked COMPLETE, but were never actually written to ~/system/config/pi-orchestrator-config.json.

Anti-pattern identified: "Proveo PASS but code doesn't match documentation" — the validation passed based on spec intent rather than verifying actual configuration state.

True Root Cause

The mechanism actually auto-pausing john H-tasks was a dead fallback block in ~/system/kernel/pi-orchestrator.js:

// Original lines 3409-3421 (13 lines, now removed)
if (!selectedTask) {
  // Fallback: check for john tasks
  const johnTask = execSync(
    'node ~/system/tools/mc.js next-task --owner john',
    { encoding: 'utf8' }
  ).trim();
  
  if (johnTask) {
    queueForHuman(johnTask);
    return null;
  }
}

When task selection failed (empty queue or filter mismatch), this fallback would:

  1. Synchronously fetch the next john task via mc.js next-task --owner john
  2. Queue it for human review via queueForHuman()
  3. Return null, preventing execution

This created the observed auto-pause behavior regardless of the missing config flags.

Fix Applied — MC #10517

Date: 2026-05-02
Builder: Codecraft
Validator: Proveo

Changes:

  1. Configuration reconciliation — Added missing flags to ~/system/config/pi-orchestrator-config.json at lines 93-94: 
    "skip_interactive_owners": ["john", "alem"],
"interactive_grace_seconds": 300
  2. Dead fallback removal — Replaced the 13-line execSync fallback block in ~/system/kernel/pi-orchestrator.js (original lines 3409-3421) with a 4-line comment + null return: 
    // No fallback to john tasks — auto-pause removed per MC #10517.
// Configuration now controls interactive routing via skip_interactive_owners.
log('No task selected; returning null.');
return null;

Verification

Proveo validation: APPROVED 2026-05-02
Acceptance Criteria: 4/4 PASS

  • AC1: pi-orchestrator-config.json contains skip_interactive_owners: ["john", "alem"] ✅
  • AC2: pi-orchestrator-config.json contains interactive_grace_seconds: 300 ✅
  • AC3: Dead fallback block removed from pi-orchestrator.js (lines 3409-3421 replaced) ✅
  • AC4: No execSync call to mc.js next-task --owner john in the selection path ✅

Evidence:

  • Config diff: git diff ~/system/config/pi-orchestrator-config.json
  • Code diff: git diff ~/system/kernel/pi-orchestrator.js
  • No remaining queueForHuman calls in fallback path: grep -n "queueForHuman" ~/system/kernel/pi-orchestrator.js shows only intentional usage in interactive routing logic

Daemon State

Current state: com.john.pi-orchestrator is STOPPED (unloaded via launchctl unload).

Reactivation: Pending CEO Step 3 directive. DO NOT restart daemon until explicitly approved — this is part of a phased rollout to validate the fix does not introduce regression.

To check status:

launchctl list | grep pi-orchestrator
# Empty output = daemon not loaded

To restart (when authorized):

launchctl load ~/Library/LaunchAgents/com.john.pi-orchestrator.plist
tail -f ~/system/logs/pi-orchestrator.log

Cross-References

  • MC #10063: Original task claiming fix (phantom — config never written, Proveo validated spec not state)
  • MC #10517: True fix reconciling config + removing dead fallback (Proveo APPROVED 2026-05-02)
  • Related pattern: Feedback memo feedback_task_description_state_verify.md — agents must tool-verify state before writing it into MC descriptions or acceptance criteria

Lessons

  1. Proveo must verify actual state, not spec intent. A config flag in the task description ≠ the flag exists in the file.
  2. Dead code can be the true mechanism. The "fix" in #10063 was irrelevant because the real culprit was a fallback block that ran regardless of config.
  3. Daemon restart ≠ verification. Stopping the daemon masked the symptom but didn't prove the fix. Reactivation under observation is the true test.

Generated by Skillforge for MC #10517 documentation deliverable. HiveMind sync pending.

No comments to display

Back to top