How do I install the Verify skill?

Run npx skillmds@latest add zivtech/verify in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

What does the Verify skill do?

How to verify changes in this prompt-only repo: reproduce committed evidence-harness recipes from scratch, don't re-run CI. Covers the evals/results/* harness pattern. It is listed under AI & ML on SkillMD.

Is Verify safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Verify?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Verify free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Verify?

zivtech (@zivtech) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Verify by zivtech — Agent Skill

Name: Verify
Author: zivtech

File contents

Verifying changes in accessibility-skills

This is a prompt-only repo — most diffs are docs/skill text with no runtime surface (SKIP those). The exception is committed evidence harnesses under evals/results/<tool>/harness/: each carries a README whose central claim is a reproduction recipe with expected output. Verify those by executing the recipe as a fresh consumer, never by re-running in the directory where the artifacts were authored.

Recipe that works:

mktemp -d under the session scratchpad; follow the harness README's commands verbatim, copying any story/test artifacts from the committed repo paths (drift between scratchpad and archive is exactly what this catches).
Compare the run's resolved dependency versions against the README's as-run table — report drift even when results hold.
Probe beyond the happy path: extract any code snippet quoted in .claude/skills/*/SKILL.md verbatim (sed the fenced block) and run it standalone — doc snippets rot independently of archives. Reproduce any "measured" behavioral claims (e.g., the virtual-screen-reader missing-stop log-leak) in the fresh environment.

Gotchas: the user-level npm guard min-release-age=7 blocks packages younger than 7 days — init tools fail with a message naming the compliant pin; use that pin, never bypass the guard. Playwright Chromium is cached machine-wide (~/Library/Caches/ms-playwright), so browser-mode Vitest runs need no download. Skill mirrors (.claude/skills/ and .agents/skills/) must stay identical — diff the edited sections as part of any verification touching them.

Frequently asked questions

Verify