Skip to content

Instantly share code, notes, and snippets.

@belisarius222
Last active March 26, 2026 18:19
Show Gist options
  • Select an option

  • Save belisarius222/d5b081c5544e7e517a97c74e600f1a36 to your computer and use it in GitHub Desktop.

Select an option

Save belisarius222/d5b081c5544e7e517a97c74e600f1a36 to your computer and use it in GitHub Desktop.
VOID Maintenance Cron Prompt #pagedrop
<!DOCTYPE html>
<html><head><meta charset="utf-8"><title>VOID Maintenance Cron Prompt</title>
<style>
body { font-family: system-ui, sans-serif; max-width: 800px; margin: 40px auto; padding: 0 20px; background: #0d1117; color: #c9d1d9; }
h1 { color: #58a6ff; border-bottom: 1px solid #30363d; padding-bottom: 8px; }
h2 { color: #79c0ff; margin-top: 24px; }
pre { background: #161b22; border: 1px solid #30363d; border-radius: 6px; padding: 16px; overflow-x: auto; font-size: 13px; line-height: 1.5; }
.key { background: #1f6feb22; border-left: 3px solid #1f6feb; padding: 8px 12px; margin: 8px 0; border-radius: 0 6px 6px 0; }
.warn { background: #da363422; border-left: 3px solid #da3634; padding: 8px 12px; margin: 8px 0; border-radius: 0 6px 6px 0; }
.new { background: #23863622; border-left: 3px solid #238636; padding: 8px 12px; margin: 8px 0; border-radius: 0 6px 6px 0; }
</style></head><body>
<h1>VOID Maintenance Cron Prompt</h1>
<p>Runs every 10 minutes. Session-only. Auto-expires after 7 days.</p>
<h2>Prompt Summary</h2>
<pre>
VOID system maintenance — keep the ENTIRE recursive loop moving, including experiments.
1. SYSTEM HEALTH
- Check task queue (pending/completed/failed)
- Runner alive? If dead, restart.
- Conductor alive? If dead, restart.
2. IF IDLE (0 pending, 0 claimed) → system stuck.
Investigate and fix IMMEDIATELY:
- Runner logs, failed tasks, missing model config task kinds
- Source pipeline broken → trigger tick, check errors
- Approval blocking → auto-approve
- Code changes needed → make them, don't wait
3. IF VOID PIPELINE COMPLETE (0 pending + directions forecasted + no experiments running):
This is the EXPERIMENT PHASE. Do NOT report idle. Instead:
- Rank directions by expected_value
- Select top 6 for experiments
- Write/reuse experiment scripts
- Deploy to GPU machines (H100, C45, C47)
- Launch experiments in parallel
- Monitor, collect results, promote winner
4. AUTO-APPROVE stuck tasks
5. EXPERIMENT MACHINE HEALTH (every check)
SSH with 3 retries. DM Ted if unreachable.
- H100: ubuntu@16.56.2.253
- C45: ubuntu@198.13.252.91
- C47: ubuntu@198.13.252.92
6. DISK SPACE (every 3rd check)
</pre>
<h2>Key Design Decisions</h2>
<div class="key"><strong>Immediate investigation on idle:</strong> If 0 pending + 0 claimed, debug and fix now.</div>
<div class="new"><strong>Experiment phase is part of the loop:</strong> When VOID finishes forecasting, automatically select top experiments and run them on GPU machines. Never ask permission. Never report "idle, ready for experiments."</div>
<div class="key"><strong>Self-healing:</strong> Add missing model config task kinds, auto-approve, restart processes, make code changes.</div>
<div class="warn"><strong>Telegram alerts for:</strong> Experiment machine unreachable after 3 retries, or system completely stuck.</div>
<h2>Lessons Learned</h2>
<ol>
<li><strong>v1:</strong> Let system sit idle 2 hours reporting "stable" while source→concept pipeline was broken.</li>
<li><strong>v2:</strong> Fixed pipeline bugs but still asked "want me to run experiments?" after VOID finished forecasting. Running experiments IS the next step — do it automatically.</li>
</ol>
</body></html>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment