How do I install the Claude Md Init skill?

Run npx skillmds@latest add skillkeel/claude-md-init 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.

Is Claude Md Init safe to use?

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

Which AI agents work with Claude Md Init?

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 Claude Md Init free to use?

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

Who published Claude Md Init?

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

Claude Md Init by skillkeel — Agent Skill

Name: Claude Md Init
Rating: 4.666666666666667 (3 reviews)
Author: Skillkeel

Claude Md Init

Use when a repo has no CLAUDE.md, or the user asks to create/refresh CLAUDE.md or project instructions for Claude Code. Generates a short, verified CLAUDE.md from the repo's real build/test/lint commands and layout, and never invents commands.

Skillkeel Updated Sep 22, 2026

File contents

claude-md-init

A good CLAUDE.md is short, true, and command-centric. Aim for 40 to 80 lines.

Procedure

Inventory (read, do not guess): ls, cat package.json | jq .scripts, Makefile targets, pyproject.toml / setup.cfg, Cargo.toml, go.mod, docker-compose*.yml, .github/workflows/*.yml (what CI actually runs), README.md headings, .editorconfig, lint configs.
Verify each command you intend to list by running its cheapest, read-only form (npm run / pnpm run to list scripts, make -n test, pytest --collect-only -q | tail -1). Do not run installs or anything that writes to the tree; if a check would modify files, dry-run or skip it and say so. Drop anything that fails.
Detect conventions from git: git log --format=%s -30 (commit style), branch naming, whether PRs are squashed.
Write CLAUDE.md with these sections, omitting empty ones:
- Project: one sentence, stack, runtime versions (from .nvmrc, .python-version, Dockerfile).
- Commands: install, dev, test (single test example), lint/format, build, migrate. One line each, copy-pasteable.
- Layout: 5 to 10 lines mapping top-level dirs to responsibilities.
- Conventions: commit style, branch naming, formatting tool, test placement, error-handling idiom if visible.
- Do not: files never to edit by hand (generated, lockfiles, migrations history), commands never to run without asking (deploy, db reset).
- Verification: the exact command(s) that must pass before claiming done.
Show the file, ask for corrections, then write it. If a CLAUDE.md already exists, propose a diff instead of overwriting.

Rules

Every command listed was executed or dry-run in this session. No exceptions.
No prose about "best practices"; only facts about this repo.
Do not include secrets, internal hostnames, or personal names.
Keep under 100 lines; longer instructions go in docs/ and are linked.

Eval

evals/claude-md-init/: fixture: Node monorepo with pnpm scripts + GitHub Actions running pnpm test:ci; expected: CLAUDE.md lists pnpm test:ci (not npm test), notes lockfile as do-not-edit, ≤ 100 lines.

---name: claude-md-initdescription: Use when a repo has no CLAUDE.md, or the user asks to create/refresh CLAUDE.md or project instructions for Claude Code. Generates a short, verified CLAUDE.md from the repo's real build/test/lint commands and layout, and never invents commands.---# claude-md-initA good CLAUDE.md is short, true, and command-centric. Aim for 40 to 80 lines.## Procedure1. Inventory (read, do not guess): `ls`, `cat package.json | jq .scripts`, `Makefile` targets, `pyproject.toml` / `setup.cfg`, `Cargo.toml`, `go.mod`, `docker-compose*.yml`, `.github/workflows/*.yml` (what CI actually runs), `README.md` headings, `.editorconfig`, lint configs.2. Verify each command you intend to list by running its cheapest, read-only form (`npm run` / `pnpm run` to list scripts, `make -n test`, `pytest --collect-only -q | tail -1`). Do not run installs or anything that writes to the tree; if a check would modify files, dry-run or skip it and say so. Drop anything that fails.3. Detect conventions from git: `git log --format=%s -30` (commit style), branch naming, whether PRs are squashed.4. Write `CLAUDE.md` with these sections, omitting empty ones: - **Project**: one sentence, stack, runtime versions (from `.nvmrc`, `.python-version`, `Dockerfile`). - **Commands**: install, dev, test (single test example), lint/format, build, migrate. One line each, copy-pasteable. - **Layout**: 5 to 10 lines mapping top-level dirs to responsibilities. - **Conventions**: commit style, branch naming, formatting tool, test placement, error-handling idiom if visible. - **Do not**: files never to edit by hand (generated, lockfiles, migrations history), commands never to run without asking (deploy, db reset). - **Verification**: the exact command(s) that must pass before claiming done.5. Show the file, ask for corrections, then write it. If a CLAUDE.md already exists, propose a diff instead of overwriting.## Rules- Every command listed was executed or dry-run in this session. No exceptions.- No prose about "best practices"; only facts about this repo.- Do not include secrets, internal hostnames, or personal names.- Keep under 100 lines; longer instructions go in `docs/` and are linked.## Eval`evals/claude-md-init/`: fixture: Node monorepo with `pnpm` scripts + GitHub Actions running `pnpm test:ci`; expected: CLAUDE.md lists `pnpm test:ci` (not `npm test`), notes lockfile as do-not-edit, ≤ 100 lines.

Frequently asked questions