- Home
- Skills
- Prompt & Skill Engineering
- Agent Rules Writer: Verified AGENTS.md & CLAUDE.md
Works with the AI Agents you already use
Agent Rules Writer: Verified AGENTS.md & CLAUDE.md
They skip the DCO sign-off that blocks every PR, and they fill space with rules like "write clean code" that no one can check.
$9
Agent Rules Writer: Verified AGENTS.md & CLAUDE.md
Example session with this skill installed
Write an AGENTS.md for this repo and make Claude Code use it too. We already have a CLAUDE.md that someone wrote last year.
- Read your context and instructions
- Compiled the agent rules writer
Example session (illustrative).
Existing files found: CLAUDE.md (64 lines), .cursorrules (12 lines). No AGENTS.md. I audited the existing files first.
Audit of CLAUDE.md
| Severity | Line | Finding | Fix |
|---|---|---|---|
| high | 9 | npm test: the repo has pnpm-lock.yaml and packageManager: pnpm@9.12.0 | pnpm test |
| high | n/a | No sign-off rule, but dco.yml checks every commit and 40 of the last 40 commits are signed | Add git commit -s |
| high | 14 | npm run e2e: there is no e2e script in package.json | Removed (reported below) |
| medium | n/a | CI also runs pnpm typecheck and pnpm lint --max-warnings 0; the file mentions neither | Gate command added |
| medium | 3 | .cursorrules says to use yarn, which contradicts the lockfile | Fold into AGENTS.md (Cursor reads it) |
| low | 22-41 | Twenty lines of general style advice that no one can check | Cut |
Commands verified in this session
pnpm install --frozen-lockfile: exit 0, 21spnpm lint --max-warnings 0 && pnpm typecheck && pnpm test: exit 0, 74s (this mirrorsci.ymljobs lint, types, test)pnpm test -- src/billing: exit 0, 6s (fast targeted form)pnpm db:generate: exit 0 (regeneratessrc/db/generated/)pnpm test:integration: unverified, needs Postgres on :5432 (fromdocker-compose.yml)
Never hand-edit: src/db/generated/** (@generated header), vendor/, pnpm-lock.yaml, and CHANGELOG.md, which changesets generates.
Needs review: openapi.yaml and migrations/ (both owned in CODEOWNERS).
Proposed changes (diff shown; 96 lines in total):
- new
AGENTS.mdwith sections Build and verify, Hard rules, Commits and PRs, Layout, Do not CLAUDE.mdreplaced with@AGENTS.mdplus 3 Claude-specific lines.cursorrulesremoved; Cursor's current docs no longer cover it, and Cursor reads AGENTS.md directly
Broken command to fix separately: the README also documents npm run e2e, which does not exist.
Shall I write these changes?
Connects securely to your tools. The creator never sees your data.
About this skill
The problem. Most agent-instruction files are guesses. They say npm test when CI runs pnpm test -- --coverage. They skip the DCO sign-off that blocks every PR, and they fill space with rules like "write clean code" that no one can check. Agents believe what the file says, so one stale command can waste a whole session or produce a PR that fails CI.
What the agent does.
- Finds the facts. It reads package.json scripts and the lockfile, Makefile, justfile, Taskfile, Cargo.toml, pyproject.toml, tox and nox configs, pre-commit hooks, and CI workflows (GitHub Actions, GitLab, CircleCI, Azure, Jenkins).
- Works out the real gate. It writes down the exact local command that matches what CI requires, including flags, toolchain version and environment variables.
- Gets your conventions from the repo. It reads commit and PR rules from commitlint config, CONTRIBUTING, PR templates and
git log, including whether DCO sign-off is enforced. - Lists what needs care. Contract files go on a "needs review" list: CODEOWNERS, API schemas, migrations, RFC areas. Generated and vendored paths go on a "never hand-edit" list, each with the command that regenerates it. It also writes a concrete secrets rule.
- Runs every command it plans to write. A command that needs missing infrastructure is marked
# unverified. A command that is simply broken is reported to you and left out. - Keeps the output short. The file is imperative, under about 120 lines, and every rule can be checked.
- Sets up one source of truth. AGENTS.md holds the rules and CLAUDE.md imports it with
@AGENTS.md, so the copies don't drift apart. - Audit mode. It reviews an existing file and lists stale commands, contradictions with CI or other docs, vague rules, missing essentials and bloat, each with a severity and a suggested fix.
- Shows you a diff first. Nothing is written until you approve it.
Who it's for. Maintainers who accept agent PRs, teams adopting Claude Code, Codex, Cursor or Copilot, and anyone whose agents keep running the wrong commands.
Requirements. An agent that can read files and run shell commands. jq is optional (used to read package.json). Works with any language or build system. AgentHarness is not required.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Prompt & Skill Engineering.
Cannot verify commands requiring external credentials or Docker if not present. Skips deployment/publishing scripts. Verification is limited to commands that finish within a reasonable timeout.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 17 days ago
- Software
GitYour project's build and test tooling