impl-validator gives an agent an independent second opinion on work that was just done. It receives a goal, a list of artifacts and a list of checks, reads each artifact, and rates every check as PASS, WARN or FAIL with a one-line note.
Another skill can spawn it as a subagent with a structured check block, or you can call it directly and describe what was just done. The report ends with an overall verdict and the specific issues to fix.
What You Get
- A structured report with a table of checks and a result for each
- An overall PASS, WARN or FAIL verdict
- A list of specific issues to fix, with file paths and lines where they apply
- Convention checks for skills, wiki pages, shell scripts and plist files
Common Use Cases
- Verify that a generated skill file has the required frontmatter
- Check a maintenance job’s state files after it runs
- Catch placeholder text left in generated output
- Get a second opinion on an implementation before showing it to a user
What Makes It Unique
It has a clear scope: it checks an implementation against its stated goal and leaves out style preferences, performance and the merits of the goal itself. Other skills in the same repository, such as daily-update and memory-bridge, call it as a subagent and apply its findings before reporting.
Under The Hood
The skill is instructions for the agent, with no scripts. In subagent mode it parses the check block it was given. It then reads the listed artifacts and returns a report as text. In user mode it asks one clarifying question if the goal is unclear and offers to fix issues after a FAIL. The skill text does not tell the agent to run commands, call web services or write files itself.