check-work

Verify completed work by spawning a subagent that reviews diffs, runs builds, and tests.

2|Updated Mar 29, 2015
One-click install
npx skills add https://github.com/ovisan/dotfiles --skill check-work-ovisan
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: check-work
Source: https://github.com/ovisan/dotfiles/tree/main/.grok/skills/check-work
Command: npx skills add https://github.com/ovisan/dotfiles --skill check-work-ovisan

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? After an AI agent completes a task, there is no guarantee the work is correct, complete, or free of regressions. This Skill adds an independent verification loop that catches broken builds, failing tests, missed requirements, and incorrect answers before you trust the result. ## Core Features & Use Cases - Subagent-Based Verification: Spawns a general-purpose verifier subagent that independently reviews the session trace, inspects the environment, and issues a PASS or FAIL verdict. - Two-Phase Review: Runs a trace review of every action taken plus a code review phase covering diffs, builds, tests, linters, security issues, and edge cases. - Iterative Fix Loop: On a FAIL verdict, fixes the identified issues and re-verifies up to three times until the work passes. - Use Case: After asking an agent to implement an authentication feature, invoke /check-work with the focus area "auth logic and JWT handling" to have a verifier confirm the implementation compiles, passes tests, and handles edge cases. ## Quick Start Ask the agent to check its work on the changes just made, optionally naming a focus area such as the auth logic.

Frequently Asked Questions about check-work

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I verify AI agent code changes automatically?

Invoke the check-work skill after the agent finishes its task. It spawns a verifier subagent that reviews the conversation trace, inspects the git diff, runs builds and tests, and returns a PASS or FAIL verdict with specific issues.

How to run a self-verification pass on a coding session?

Use /check-work, /check, /verify, or /self-verify after the work is done. You can optionally pass a focus area like "auth logic and JWT handling" so the verifier scopes its verdict to that part of the changes.

What does the verifier subagent check during code review?

The verifier checks correctness, adequacy, excess changes, and edge cases. It runs the project's build, test suite, and linters, reviews the diff for bugs and security issues, and confirms compliance with rules stated in the repo's AGENTS.md or Claude.md files.

Can check-work verify non-code tasks like deployments or research?

Yes. The trace review phase covers operational tasks, git and PR actions, research, and Q&A. The verifier independently confirms outcomes, such as checking that a submitted job was accepted or that an answer matches source material.

What happens when the verification verdict is FAIL?

On a FAIL verdict, the agent fixes the issues the verifier identified and runs the verification again. This fix-and-recheck loop repeats up to three times before stopping.

When should I not rely on check-work verification?

The verifier shares the same environment and tools as the original agent, so it cannot validate anything requiring external credentials or systems the session cannot access. It also depends on the repo documenting build and test commands.