agentleFS
Sign inStart free
DiUSDiUS/agent-toolkit/AGENTS.md

agent-toolkit

AGENTS.md2 starsChanged 4 months agoMITView on GitHub

Working context and rules for any AI agent (or human) contributing to this repository: the agent-toolkit repo itself, not the components it ships. CLAUDE.md imports this file and adds Claude-Code-specific notes. Keep it lean: it loads into an agent's context every session. agent-toolkit is DiUS's shared collection of skills, commands, agents and hooks for AI coding agents. It ships two ways: as portable components installable with npx skills, and as a single Claude Code plugin bundling skills, commands and agents.

  • Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.

    Of the AGENTS.md files we read, the 31,774 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.

  • In this repository4 other files it gives its agents.

The file

We split this AGENTS.md into the rules it gives an agent

2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Add hooks/MCP as a hard requirement, or otherwise break host-agnostic behaviour.

    line 82
  2. Very specific:

    Commit with a failing scripts/validate.js.

    line 86

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
AGENTS.md

AGENTS.md

Working context and rules for any AI agent (or human) contributing to this repository: the agent-toolkit repo itself, not the components it ships. CLAUDE.md imports this file and adds Claude-Code-specific notes. Keep it lean: it loads into an agent's context every session.

What this repo is

agent-toolkit is DiUS's shared collection of skills, commands, agents and hooks for AI coding agents. It ships two ways: as portable components installable with npx skills, and as a single Claude Code plugin bundling skills, commands and agents.

For the catalogue of what's currently in the toolkit, see README.md. For how any one component works, read that component's own entry point (a skill's SKILL.md, a command or agent's frontmatter). This file is about contributing to the repo.

Repo map

skills/<name>/                 Self-contained skills; everything a skill needs lives inside it.
commands/                      Slash commands — single Markdown files with frontmatter.
agents/                        Claude Code subagents a skill can delegate work to.
hooks/                         Optional, opt-in Claude Code hooks — not wired into the plugin.
.claude-plugin/                Claude plugin + marketplace manifests.
docs/getting-started.md        Per-tool install notes.
scripts/validate.js            Verification gate (run before committing).
.github/workflows/validate.yml CI that runs the gate on push/PR.

Principles

  • Host-agnostic first. Components should run on any capable agent. Do not add hard dependencies on hooks, MCP servers, or a specific runtime. Optional integrations must degrade gracefully: used when present, skipped cleanly when absent. Claude-specific wiring stays confined to .claude-plugin/, commands/, agents/ and hooks/.
  • Skills are self-contained. Everything a skill's entry point references (playbooks, references, templates) lives inside its own skills/<name>/ directory, which is the unit npx skills and the plugin install. Never point a skill file at something outside its own directory.
  • Lean docs. Docs (including this file) earn their length by making a contributor productive. Prefer prose and small tables; push depth into the specific file, not the index.

How to add each component type

  • Skill → new skills/<name>/SKILL.md (or equivalent entry point) with name + description frontmatter, self-contained. The ./skills path in .claude-plugin/plugin.json already covers new subdirectories; add a row to the skills table in README.md.
  • Command → new commands/<name>.md with name + description frontmatter. It ships automatically via the commands path in .claude-plugin/plugin.json. See commands/README.md.
  • Agent → new agents/<name>.md with name + description frontmatter and a tools list scoped to the minimum it needs (read-only where the agent only inspects code). Add it to the agents array in .claude-plugin/plugin.json to ship it in the plugin.
  • Hook → new hooks/<name>/ with a hooks.json and its script(s), following hooks/README.md. Do not wire it into .claude-plugin/plugin.json; hooks stay opt-in, and consumers copy them into their own project.

See CONTRIBUTING.md for the full mechanics.

Verification gate (required)

Before committing, run:

node scripts/validate.js

It checks that the manifests are valid JSON with required keys, that every SKILL.md, agents/*.md and commands/*.md (except its README) has name + description frontmatter, that every path referenced by plugin.json exists, that relative markdown links resolve and a skill's links stay inside that skill, and that any hooks/**/hooks.json is valid JSON. CI runs the same script on every push and PR, and a red gate blocks merge.

validate.js itself stays component-agnostic: it validates the repo's format, never one component's content. A component with its own invariants (a rule worded identically in several files, a closed vocabulary) puts them in scripts/checks/<component>.js, which the gate loads automatically. See scripts/checks/README.md.

Do not

  • Add hooks/MCP as a hard requirement, or otherwise break host-agnostic behaviour.
  • Reference files outside a skill's own directory from within that skill.
  • Wire a hook into .claude-plugin/plugin.json; hooks are opt-in only.
  • Bloat the repo docs or this file.
  • Commit with a failing scripts/validate.js.

Pointers

  • Repo overview and install: README.md
  • Per-tool setup: docs/getting-started.md
  • Adding a component: CONTRIBUTING.md

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free