verify-done
Runs the comprehensive quality gate before declaring work done: discovers the checks CI actually runs, executes them in order, then reviews git status and the session diff. Use when the user says 'verify done' or '/verify-done', or before pushing any branch.
Install
npx skills add https://github.com/domengabrovsek/agent-config/tree/main/skills/verify-done
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install domengabrovsek-claude@llmmart
git clone https://github.com/domengabrovsek/agent-config.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole domengabrovsek/claude collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
Comprehensive quality gate before declaring work done.
When package.json declares a verify script, the repo owns the gate. Skip steps 1 and 2:
- Run
npm run verifyonce in the foreground with a 600000ms timeout. It takes minutes. Run it alone: no pipe,;or||. - Do not read workflow YAML or run individual checks.
verifymirrors CI and prints what it skips. - On success with a clean tree, the
post-verify-stamp.shhook writes HEAD to<git-dir>/verify-passed. The PR-open hook reads that stamp. A background run never stamps. - Continue at step 3.
Otherwise:
- Discover CI steps: read
.github/workflows/*.yml(or.gitlab-ci.yml,Jenkinsfile, etc.) andpackage.jsonscripts to find the actual checks CI runs. Only run what CI actually runs - do not guess or add extra steps. - Run each CI step in order: execute the discovered commands (lint, typecheck, test, build, etc.) in the same order as CI. Stop at the first failure.
- Git status: show uncommitted changes and untracked files.
- Diff review: summarize what changed in this session (files, lines added/removed).
- Spec check: if a spec in
.claude/state/specs/hasbranch: <current branch>, spawn theSpec Verifierand include its verdict. Any automated criterion not PASS at HEAD fails the gate.
If all CI steps pass and git status is clean: output "READY - all quality gates passed." If any step fails: list failures and stop. Do NOT declare work done. If no CI configuration is found, say so and ask the user what checks to run.
Files (claude)
-
SKILL.md 1.8 KB
--- name: verify-done description: "Runs the comprehensive quality gate before declaring work done: the repo's `npm run verify` when package.json declares it, otherwise the checks CI runs, then reviews git status and the session diff. Use when the user says 'verify done' or '/verify-done', or before pushing any branch." --- Comprehensive quality gate before declaring work done. **When package.json declares a `verify` script**, the repo owns the gate. Skip steps 1 and 2: - Run `npm run verify` once in the foreground with a 600000ms timeout. It takes minutes. Run it alone: no pipe, `;` or `||`. - Do not read workflow YAML or run individual checks. `verify` mirrors CI and prints what it skips. - On success with a clean tree, the `post-verify-stamp.sh` hook writes HEAD to `<git-dir>/verify-passed`. The PR-open hook reads that stamp. A background run never stamps. - Continue at step 3. **Otherwise:** 1. **Discover CI steps**: read `.github/workflows/*.yml` (or `.gitlab-ci.yml`, `Jenkinsfile`, etc.) and `package.json` scripts to find the actual checks CI runs. Only run what CI actually runs - do not guess or add extra steps. 2. **Run each CI step in order**: execute the discovered commands (lint, typecheck, test, build, etc.) in the same order as CI. Stop at the first failure. 3. **Git status**: show uncommitted changes and untracked files. 4. **Diff review**: summarize what changed in this session (files, lines added/removed). 5. **Spec check**: if a spec in `.claude/state/specs/` has `branch: <current branch>`, spawn the `Spec Verifier` and include its verdict. Any automated criterion not PASS at HEAD fails the gate. If all CI steps pass and git status is clean: output "READY - all quality gates passed." If any step fails: list failures and stop. Do NOT declare work done. If no CI configuration is found, say so and ask the user what checks to run.
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.