Skip to content
Agentmart
Submit a tool

agents-md

Create repo-specific AGENTS.md guardrails from the codebase.

Skill by thunter009

IndexedMIT
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/thunter009-agents-md

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Create repo-specific AGENTS.md guardrails from the codebase.

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/agents-md

What's inside

Skill name
agents-md
Description
Create repo-specific AGENTS.md guardrails from the codebase.
License
MIT
Instructions
274 lines in SKILL.md · view on GitHub

Agent Skills spec issues

  • allowed-tools: must be a space-separated string

Files (1)

  • SKILL.md

SKILL.md (first lines)

# Bootstrap AGENTS.md

Generate a project-tailored AGENTS.md, the canonical instruction file for AI coding agents working in a repository.

## What AGENTS.md Is

AGENTS.md is a root-level markdown file that tells any AI coding agent (Claude Code, Codex CLI, Gemini CLI, Cursor, etc.) how to behave in this specific repo. It covers:

- **Safety invariants**: what agents must never do
- **Toolchain**: language runtimes, package managers, build tools
- **Architecture**: repo layout, key modules, data flow
- **Code discipline**: editing rules, file creation policy, backwards compat stance
- **Quality gates**: linters, tests, type checks agents must run
- **Session protocol**: how to land work cleanly so the next agent can pick up

## Process

### Phase 1: Discover the Repo

Run all discovery in parallel where possible. Gather:

1. **Root files**: `ls` the repo root. Look for README, package.json, pyproject.toml, Cargo.toml, go.mod, Makefile, Dockerfile, docker-compose, .env.example, etc.
2. **Existing agent config**: Check for CLAUDE.md, .cursorrules, .cursor/rules/, .claude/settings.json, .github/copilot-instructions.md, CONTRIBUTING.md, .editorconfig
3. **Git info**: default branch name, recent commit style (`git log --oneline -20`), any branch naming conventions
4. **Tech stack detection**:
   - JS/TS: package.json → runtime (node/bun/deno), package manager (npm/yarn/pnpm/bun), framework
   - Python: pyproject.toml/setup.py/requirements.txt → manager (uv/pip/poetry), framework
   - Rust: Cargo.toml → edition, workspace layout
   - Go: go.mod → module path
   - Other: detect from file extensions and build files
5. **Directory structure**: Use Glob (`**/*`) and `ls` to map top-level and second-level directories (skip node_modules, .git, __pycache__, .venv, etc.)
6. **Testing setup**: test runner, test directory, how to run tests
7. **Linting/formatting**: eslint, prettier, ruff, black, clippy, golangci-lint, etc.
8. **CI/CD**: .github/workflows/, .gitlab-ci.yml, Jenkinsfile, etc.
9. **Generated files**: any codegen, build outputs, compiled assets
10. **Issue tracking**: .beads/, Linear config, GitHub Issues conventions

**Check for existing AGENTS.md**: If one already exists, read it first. Ask the user whether to replace it entirely or merge discovered info into the existing structure. Preserve any custom rules or project-specific sections the user already wrote.

### Phase 2: Draft the AGENTS.md

Use the template below, adapting each section to what was discovered. **Omit sections that don't apply.** A Bash-only project doesn't need a "Generated Files" section if there aren't any.

**Key principles:**
- Be specific to THIS repo, not generic
- Include exact commands (not "run the linter" but `ruff check . --fix`)
- State the default branch by name
- List real directories from the repo layout
- Reference actual config files found

### Phase 3: Present and Refine

Show the draft to the user. Ask if anything should be added, removed, or adjusted. Common refinements:
- Adding project-specific forbidden patterns
- Adjusting the strictness of safety rules
- Adding tooling assumptions (what's installed on dev machines)
- Documenting deployment targets

Write the final AGENTS.md to the repo root.

## Template

Below is the canonical template. Adapt every section to the actual repo. Remove sections in `[brackets]` that don't apply. Replace all `{placeholders}`.

~~~markdown
# AGENTS.md: {Project Name}

> Guidelines for AI coding agents working in this codebase.

---

## Safety Rules

### Human Override

The user's instructions take precedence over everything in this file. If the user explicitly asks you to do something that contradicts a rule below, follow the user's instructions.

### No File Deletion

Never delete files or directories without explicit permission from the user in the current session. If something should be removed, ask first.

### No Destructive Git

The following commands are forbidden unless the user provides the exact command and explicit approval:

- `git reset --hard`
- `git clean -fd`
- `rm -rf`
- Any command that can delete or overwrite uncommitted work

Prefer safe alternatives: `git status`, `git diff`, `git stash`, copying to backups.

### No File Proliferation

Never create variant files (`main_v2.ts`, `component_backup.tsx`, `utils_improved.py`). Edit existing files in place. New files are only for genuinely new functionality.

---

## Toolchain

{Describe the exact runtimes, package managers, and build tools. Be prescriptive.}

[Example for JS/TS:]
- **Runtime:** Bun
- **Package manager:** Bun (`bun install`, `bun add`). Never use npm/yarn/pnpm.
- **Lockfile:** `bun.lock` only
- **Target:** Latest Node.js; no legacy compatibility needed

[Example for Python:]
- **Runtime:** Python 3.12+
- **Package manager:** uv (`uv pip install`, `uv sync`). Never use pip directly.
- **Lockfile:** `uv.lock`
- **Virtual env:** `.venv/` (managed by uv)

[Example for Rust:]
- **Edition:** 2024
- **Build:** `cargo build`, `cargo test`
- **Clippy:** All warnings are errors

From thunter009/skills, MIT.

More from thunter009/skills

Signals

Stars
0
Forks
0
Open issues
0
Last commit (this folder)
Aug 19, 2026 (2 months ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from thunter009

Cite this page

agents-md skill. Agent Mart. https://agentmart.dev/skills/thunter009-agents-md (last updated 2026-10-24)