Agent Watchdog

85TrustedInstructions only

Verification of AI agent sessions, PRs, and logs to identify execution gaps and errors.

What it does

Agent Watchdog acts as a formal reviewer for other autonomous agents. By ingesting session IDs, transcripts, or repository links, it reconstructs the original user intent and compares that contract against the actual artifacts produced by the target agent. It highlights discrepancies between requested tasks and documented outcomes.

Beyond basic oversight, the skill can provide an audit report or attempt narrow, surgical corrections when issues are identified. This is designed for teams managing complex agent workflows who need a secondary layer of validation before merging code or finalizing work.

When it earns its place

Reach for it when…

Apply this when you need an objective assessment of an agent's recent activity, such as checking a failing CI run, reviewing a pull request generated by another bot, or verifying that a long-running session actually completed every requirement.

Look elsewhere when…

Do not use this for initial task execution or primary generation, as it is strictly a diagnostic tool. It is not designed for interpreting human-written code or manual commits, nor is it a replacement for comprehensive manual regression testing on complex, multi-system migrations.

What to say to use it

You never invoke a skill directly — the assistant picks it up when your request matches what it covers. These are the kinds of things that do that.

Check the work completed in this PR
Did the agent follow all instructions in this session
Compare the results of these two agent runs
Audit this transcript for missed requirements

Install it

Install for all your projects

Copy the skill directory into your personal skills folder. Claude picks it up on its own the next time your request matches what it does — there is nothing to enable.

bash

git clone --depth 1 --branch main https://github.com/BuilderIO/skills.git /tmp/skills
mkdir -p ~/.claude/skills
cp -r /tmp/skills/skills/agent-watchdog ~/.claude/skills/agent-watchdog

Install for one project only

Put it in the project's own skills folder instead, so it can be committed and shared with everyone working on that repository.

bash

cp -r /tmp/skills/skills/agent-watchdog .claude/skills/agent-watchdog

If a skill of the same name exists in both places, the personal one wins.

Installing on Claude.ai or through the API instead? The full guide covers every surface.

85/100Trusted source

What checks out

Original source. Published in the repository of whoever wrote it, rather than copied out of it into a collection.
Adoption. The host repository has 4,025 stars. This counts the whole repository, not this skill alone.
Recently maintained. The repository was updated 5 days ago.
Spec-compliant. Frontmatter matches the Agent Skills specification, so it loads in any client implementing it.
Licensed. Released under MIT, so the terms of reuse are stated.
Substantive instructions. The body carries enough detail to change how a task is actually performed.

What to know

Official author. Community-published. That is the norm for skills and not a fault — it only means no vendor stands behind it.

This score describes where the skill came from — its origin, licence, upkeep and conformance to the spec. It is not a security audit, and it says nothing about whether the skill's instructions are correct. How it is calculated.

What it ships

Executable files: None — this skill is instructions only
Files in total: 3
Context cost: about 1,140 tokens once activated (estimated)

Source

Published in: BuilderIO/skills
Path in repository: skills/agent-watchdog/SKILL.md
Licence: MIT

New to skills? Start here →