Vigil

Your agent crashes at 2 AM. Nobody's watching. By morning, three tasks failed silently, a cron didn't fire, and your session is bloated with stale data.

Vigil watches while you sleep.

Ships with 39 pre-diagnosed error patterns and instant fixes. Not a debugging framework — a battle-tested library of real OpenClaw failures and the exact commands to fix them. Your agent matches errors in seconds instead of diagnosing from scratch.

What happens when an error hits:

DETECT → match against 39 known patterns → HEAL automatically
         ↓ (if unknown)
         DIAGNOSE → trace root cause → HEAL or ESCALATE
         ↓
         LEARN → save the fix so it's instant next time

What's inside:

vigil/
├── SKILL.md                  — 6-phase self-healing framework
├── references/
│   ├── known-errors.md       — 39 real-world error patterns + fixes
│   └── patterns.json         — regex-based instant matching index
└── scripts/
    ├── preflight.sh          — system health check (run anytime)
    └── nightly-sweep.sh      — automated cleanup + monitoring

The 39 patterns cover: silent model fallbacks, plugin breakage after upgrades, stale session leaks, SPA routing failures, missing environment variables, port conflicts, npm dependency breaks, expired API keys, rate limiting loops, disk filling from logs, cron delivery failures, WebSocket drops, database schema cache, git conflicts, embedding timeouts, Cloudflare blocks, PM2 zombie processes, Docker startup failures, Python venv confusion, CORS errors, SSH failures, memory leaks, timezone bugs, encoding errors, and more.

What Vigil does that free debugging skills don't:

• ⚔️ Auto-heals — retries, installs dependencies, fixes configs, kills zombie processes, clears stale data. No human needed for 19 of 39 patterns.
• 🌙 Nightly sweep — real bash script runs at 3 AM, cleans orphaned sessions, checks PM2 health, monitors disk trends, flags recurring errors.
• 📊 Weekly digest — aggregates the week's errors into one actionable summary.
• 🧠 Learns — every new error your agent solves gets saved. It never investigates the same problem twice.
• 🔍 Silent preflight — checks tools, model, memory, disk, and API connectivity every session. No output unless something's wrong.
• 🤖 Self-onboards — creates its own files, offers to set up crons, runs initial health check. Zero manual config.

Built from production, not theory. Every pattern in this skill came from running a 6-agent team on a live VPS — not a tutorial. The plugin signature fix? We hit that. The stale session leak? We hit that. The model silent fallback? We hit that. This is the playbook we wish we had on day one.

Works with any agent. Vigil is a skill, not a persona. Drop it into your skills directory alongside Felix, Teagan, or your custom agent. It adds protection without changing anything.

One-time purchase. No subscription. No API keys required. No external dependencies. Runs entirely inside your OpenClaw instance.

---

Read out Story here on how this Skill was created Built by The Agent Crew · v1.0.0