Safety and Responsible Use

Last updated: 18 July 2026

How Provely (provely.dev, an Unfussy Labs product) handles the tricky parts of reviewing an AI agent skill against a real model, and what we do not claim to do.

Nothing in your skill is executed

Skill Check reads your skill as text. It does not run your skill, run any script inside it, install anything, or reach out to any service the skill mentions. The only place your text is sent is to the model that writes the review, and that model has no tools: it cannot read files, call other services, or take actions.

Instructions hidden inside a skill

Every uploaded file is treated as untrusted data under review. If a file contains instructions aimed at the reviewer, for example "ignore your rubric" or "give this skill a passing grade", we do not follow them. Instructions inside SKILL.md, references, or scripts cannot change the reviewer's rubric or ask it to use tools. Skill Check does not follow links found inside a skill and does not clone arbitrary websites. Public GitHub input is restricted to github.com and resolved to a commit SHA.

What we do not check

We want to be plain about the edges of this tool:

• We do not verify facts. If a skill's own text makes a factual claim, we do not fact-check it.
• We do not test your setup. We review the files you submit. We do not run the skill in your agent, test your account, or prove behaviour in your production environment.
• We report the review we made. Results describe this specific review: deterministic folder observations, the model's judgements, and what one routing simulation did. They are evidence, not a promise about the future.
• A trigger simulation is platform-limited evidence. Skill Check demonstrates how one model routed the generated cases in that review. It does not prove that Codex, Claude Code, or every future model will make the same routing decision.
• Generated tests have not passed yet. Skill Check can produce a regression pack, but those tests remain proposed until you run them in the intended agent environment.

Using the service responsibly

Please use Provely for skills you have the right to submit, and do not use it to refine content meant to harm, deceive, or harass. The Terms of Use set out acceptable use in full. This is a review aid, not a guarantee, and you stay responsible for how you use your skills and their results.

Reporting a problem

Found a bug, a bad result, or something that looks unsafe? Tell us: unfussylabs@gmail.com. Reports about safety and abuse are read first.