Vigil: The OpenClaw Skill That Watches Your Agent While You Sleep

It was 2 AM. One of our agents had been running a background task since 9 PM. We checked in expecting a clean result. Instead: a vague "400 Invalid input" error, no context, no trace, nothing useful.

We spent 45 minutes digging. Plugin upgrade. Function signature changed. Memory system broke. Silent. No alerts.

We fixed it. Went to bed. And the next morning we started writing Vigil.

Agents Break Silently. That's the Real Problem.

If you run OpenClaw agents in production, you already know this. Errors don't announce themselves. Your agent doesn't page you. It either crashes quietly, falls back to something worse, or keeps running in a degraded state while you assume everything is fine.

We've lived through all three.

Our agent's model silently switched to a different one when the original got removed from the catalog without warning. We didn't notice for days. The outputs felt slightly off but nothing obvious. We only caught it during a routine config review.

We had 37 orphaned session files accumulating on disk. Old subagent tasks that completed but left files behind. Not a crisis. Until it was.

We had ghost messages leak into an active chat from a stale exec process. A task that finished hours earlier was somehow still echoing output into a live session. That one was deeply confusing at 11 PM.

The pattern was always the same: something broke silently, we noticed too late, we traced it manually, we fixed it, we thought "we should have a system for this."

So we built one.

What Vigil Actually Does

Vigil is an OpenClaw debugging skill. One SKILL.md file. Drop it in, and your agent gains a full OpenClaw error detection and self-healing layer.

Here's how it works in practice.

Every session, before your agent does anything, Vigil runs a silent preflight check. It verifies tools are available, the model is what you configured, memory is accessible, disk isn't critically full. If something is wrong, it tells you before you waste a session on a broken environment. Think of it like a cockpit check before takeoff.

If errors happen mid-session, Vigil doesn't just log them. It matches them against a library of 39 pre-diagnosed patterns. Each pattern has a known root cause and a recommended fix. When Vigil sees "400 Invalid input" from the memory API, it doesn't just surface the error. It tells you: "This matches Pattern #2. A plugin update likely changed a function signature. Check your memory plugin version."

We would have saved 45 minutes the first time we hit that bug.

For 19 of those 39 patterns, Vigil doesn't even need to tell you. It just fixes them. Stale sessions get cleaned. Disk warnings get handled. Common config drift gets corrected. You find out in the weekly digest, not at 2 AM.

The nightly sweep runs at 3 AM. It cleans orphaned session files, checks PM2 process health, and monitors disk usage. Every Sunday you get a digest of what happened during the week. Errors caught, patterns matched, things auto-healed.

And over time, Vigil learns. When your agent hits a new error it hasn't seen before, Vigil builds an error memory entry. The more your agents run, the smarter the pattern matching gets. It isn't static.

Real Examples From Our Production Setup

These aren't hypotheticals. These are things that actually happened to us while building The Agent Crew.

The silent model switch. We configured Felix to run on a specific model. That model got removed from the catalog without warning. Felix fell back silently. We ran on the wrong model for days. Vigil's preflight check now catches this on every session startup.

The orphaned session files. 37 of them. Subagent tasks that completed but left files behind. Vigil's nightly sweep handles this automatically. We don't think about it anymore.

The ghost exec. A stale exec from a dead subagent session leaked output into our active chat. Ghost messages from a task that had finished hours earlier. Vigil detects stale sessions and cleans them before they pollute live work.

The memory breakage. Plugin upgrade, silent function signature change, "400 Invalid input" with no useful context. 45 minutes of manual tracing. Vigil would have matched it to Pattern #2 in seconds and told us exactly where to look.

What's Inside

Vigil is a single SKILL.md file. No API keys. No external dependencies. It runs entirely on your local OpenClaw setup.

The skill includes:

• A 6-phase framework: PREVENT, DETECT, DIAGNOSE, HEAL, ESCALATE, LEARN
• 39 pre-diagnosed error patterns with regex signatures for fast matching
• Auto-healing logic for 19 patterns (the ones safe to fix without human review)
• preflight.sh: runs at session start, checks your environment before work begins
• nightly-sweep.sh: runs at 3 AM via cron, cleans stale sessions and monitors health
• A first-run onboarding flow that sets up cron automatically
• Weekly Sunday digest with a plain-English error summary

The 6-phase framework is the core idea. Vigil doesn't just react to errors. It tries to prevent them (PREVENT), catch them early (DETECT), explain them (DIAGNOSE), fix what it can (HEAL), escalate what it can't (ESCALATE), and remember everything for next time (LEARN).

Most debugging tools stop at detect. Vigil goes all the way through.

Who Should Get This

If you run OpenClaw agents in any serious capacity, Vigil is worth $9.

It's a good fit for:

• Solo builders who don't have a team watching their agents overnight
• Anyone running multiple agents (Felix, Teagan, custom builds) across different workspaces
• People who've debugged the same error more than once and are done doing it manually
• Teams moving from "testing agents" to "running agents in production"

Vigil is additive. It doesn't replace your agent or change how it works. It sits alongside any OpenClaw agent and adds the monitoring layer that OpenClaw doesn't ship with by default. If you're running a custom agent built from scratch, it works there too.

No API keys. No subscription. One file. $9 once.

Get Vigil

We built Vigil because we needed it ourselves. Then we packaged everything we learned about OpenClaw failure modes into one skill anyone can drop into their setup.

If your agents run while you're not watching, Vigil watches for you.

Get Vigil on Claw Mart for $9

One-time purchase. Instant download. Works with any OpenClaw agent.