⏳ This skill is pending AI review.
Scores will appear once the review pipeline completes.
public-claims
Write and fact-check public copy so that every figure, credential, promise and command has a receipt - a source with a locator, a command that re-derives it, or the author's own words - and remove any claim without one instead of softening it. Use when drafting or reviewing a README, release notes, a launch post or thread, slides, or a website page; and whenever a draft states a number, a ranking, a credential, a promised result or a command.
Choose how to use this skill
You do not need every option. Choose the path your AI client supports. The stable page stays the same; versioned files are immutable.
1. Native installer
This listing has no registered native installer command. Use the complete package or source fallback below, depending on what your client supports.
Do not guess an installer command or replace an existing version without reviewing the diff.
2. Complete package recommended
Download the ZIP when available. It includes SKILL.md plus the references, security notes and version metadata.
No complete ProSkills package is published for this listing yet.3. Prompt-only
Copy the prompt above when the agent can read the stable page or when you want to adopt the workflow without installing a skill.
Need only the instruction file?
Download SKILL.md only if your client requires a single file. The complete ZIP is safer for a full installation because it preserves the references and release context.
No path installs or executes anything by itself. Your agent still needs access to the project files. Before updating, compare the installed version and review the diff.
// RATINGS
// README
Done is a claim
English · Türkçe
18 rules and 3 skills for coding agents, each one paid for by a real failure.
Your agent says "done". Usually it is. Sometimes the tests never ran, the exit code belonged to tail, or the fix reached one call site out of four. These rules make an agent show its evidence before it says the word.
We wrote them while running Claude Code and Codex agents on our own repository. Between 14 and 29 September 2026 a server re-ran the acceptance command of 3,489 agent tasks: 2,282 passed on that first independent re-run, and 736 tasks had passed their own test while turning the full suite red on main. Every rule below comes with the incident behind it in INCIDENTS.md, and every figure with its source in CLAIMS.md.
Use it
Copy AGENTS.md into the root of your repository. Codex and other agents that read AGENTS.md pick it up as it is.
For Claude Code, also copy CLAUDE.md. It holds one line, @AGENTS.md, which imports the rules.
Already have an AGENTS.md or CLAUDE.md? Paste the rules under your own.
The rules
Before you start
- Write the acceptance command before the work, and never edit it to pass.
- Check that every path you were given exists.
Before you say "done" 3. Re-run the acceptance yourself, on the current revision, and quote its output. 4. Run the whole suite, not only your test. 5. Record the command's own exit code, not the pipe's. 6. "No tests ran" is a failure. 7. Unknown is not pass. 8. Green means every check.
When you report 9. Read the artefact, not the summary. 10. A file that exists is not a file that works. 11. Look at it yourself. 12. Verify every citation. 13. State the base, and re-check it before you publish.
When you write tests and fixes 14. A test that repeats a constant proves nothing. 15. Fix the class, not the instance. 16. Run a detector on a known-good case first. 17. Register cleanup before the work it cleans up.
Never 18. Never run a destructive command to answer a question.
Skills
Three skills for the moments the rules are hardest to follow:
| Skill | Use it when |
|---|---|
| reading-measurements | before reporting any figure: an exit code, a test count, a green, a list size |
| whose-red | tests fail and someone has to decide: the change, or main? |
| public-claims | writing a README, release notes or a launch post with numbers in it |
For Claude Code, copy a skill's folder into .claude/skills/ in your repository, or into ~/.claude/skills/ for every project:
git clone https://github.com/Grit-77/done-is-a-claim
cp -r done-is-a-claim/skills/* ~/.claude/skills/
Each skill is one Markdown file, so any other agent can be pointed at it directly.
Rules are not a gate
A rule in a prompt is advice: the agent can still skip it. We built that gate as a separate tool, RADAR, which refuses to close a task until its acceptance has passed again on the current revision. These rules are the part that fits in one file.
Contributing
Have a rule that cost you a real failure? Open an issue with the incident: what the agent claimed, what was true, and the number or output that showed the difference. Rules without an incident are not added.
Licence
Apache-2.0. Made by Grit, Ankara. Also from us: cinematic-site, a skill set for websites that ship checked.
// HOW IT'S BUILT
KEY FILES