Governed skills, rules and commands for AI coding agents, served by a read-only MCP server.
io.github.event4u-app/agent-config MCP Server
This read-only MCP server serves “agent-config” content that defines governed skills, rules, and commands for AI coding agents. The server exposes configuration described by the provided name and description, intended for developer consumption via MCP-based tooling.
🛠️ Key Features
Provides governed skills
Provides rules
Provides commands
Read-only access
🚀 Use Cases
Supplying configuration for AI coding agents
Enforcing agent skills, rules, and command behavior via MCP
⚡ Developer Benefits
Centralizes agent configuration for AI coding agents through an MCP server
Uses a “read-only” model for safer configuration retrieval
⚠️ Limitations
Only read-only server behavior is indicated
No additional details (e.g., tool count, specific APIs, or formats) are provided in the source data
Installs into your user-level AI tool configuration; your repository only gets an agents/overrides/ folder for your own changes. Works with 20 AI coding tools.
Agent Config — teach your AI coding agents to check their work before they say done
Install once. Your AI coding tools are told to plan, test and verify before they say done; on Claude Code a turn that changed files without checking them can be stopped.
bash
npx -y @event4u/agent-config init
Free and MIT-licensed. Node 22.6 or newer on macOS, Linux or WSL2. Costs more tokens: 7.8× a run with no rules on Claude Haiku as installed, 2.9× with only the core rules. What it costs →
On a desktop this opens a browser wizard: pick your role and your tools, click Finish. Nothing is written before you do.
Try it in 30 seconds
Not ready to install anything? Drop one read-only checker into your repo and ask it whether your branch is really done:
bash
mkdir -p .claude/agents
curl -fsSL https://raw.githubusercontent.com/event4u-app/agent-config/main/docs/wedge/production-validator/production-validator.md \
-o .claude/agents/production-validator.md
# then in Claude Code: @production-validator check this branch is actually done
production-validator reads the code that would ship, hunts mocks and placeholders on that path and asks for proof that it ran. It changes nothing. What it checks →
What changes
Before. You ask your agent to add a CSV export. It writes a handler that
returns a hard-coded placeholder, the existing tests stay green and it replies
"Done — export implemented." Nothing ever ran the new code.
After. In Claude Code you ask @production-validator whether the branch is
really done. It reads the code that would ship, finds the placeholder and
answers NOT READY with the exact file and line. On a branch that is really
finished it answers READY and reports nothing. We tested this on two cases:
one with a placeholder left in, one finished.
On Claude Code the package can also refuse to let a turn end on "done" when the
agent changed files and ran nothing to check them. On every other tool that is
an instruction the agent is asked to follow, not a block.
What your agent is asked to do
Default behavior
With agent-config
Guess and edit blindly
Analyze code before changing it
Drift from project conventions
Follow detected stack conventions
Skip or invent tests
Write tests in the project's framework
Generic commit messages
Conventional Commits with scope + ticket links
Skip quality checks
Run the project's quality pipeline and fix reported errors
Open PRs without context
Structured PR descriptions from Jira / Linear / GitHub
Claim "done" without proof
Verify with real execution before claiming done
It changes what the agent is told to do, not how smart the model is. Re-measured in November 2026 after a scorer repair, the full install's discipline gain on Claude Haiku was not significant and Claude Sonnet showed no gain; only the small core rule set kept a significant gain on Haiku. We publish results like that one instead of hiding them. The whole story →
How it fits together
One install gives the agent you already use a set of skills, agent rules and slash commands. Rules that matter on every task are always on; the narrower ones load when a task needs them. See every slash command →
What it costs
More tokens. On Claude Haiku, on two scope-discipline tasks, the install as shipped cost 7.8× a run with no rules; the core rules alone (kernel plus downstream-changes) cost 2.9×. The discipline_profile setting does not yet change what a Claude install loads.
Blocking on Claude Code only and only while its hooks run. A host setting can switch them off and a user-installed mod can lift a refusal; agent-config doctor neighbours lists installed mods. Elsewhere the rules are instructions: they cut violations but do not remove them.
Network. By default nothing leaves your machine except the daily update check against the npm registry and a daily question to GitHub, through your own gh login, whether a pull request you pushed has merged. The update check is off in CI, without a terminal or with AGENT_CONFIG_NO_UPDATE_CHECK=1. Telemetry is off. The second-opinion feature is off until you set it up; then it sends work to other AI providers with your own API keys, which bill you. What leaves your machine →
Your machine. Node 22.6 or newer. Standing cost of the always-on rules per tool: table.
Your way out.agent-config uninstall --global removes only its own entries and leaves other tools' config untouched.
agent-config code-graph map --open draws a local map of your repository: who calls what, which tests cover it, data, interfaces and delivery. Pick a file and every area it reaches lights up. It runs offline and keeps file:line for every fact.
You start a piece of work, not a list of commands. These cover most days:
Command
What it does for you
/work "<what you want>"
Takes a free-form request through refine, plan, implement, test and verify, and reports back; it never commits for you.
/implement-ticket <ticket>
The same loop for a ticket: it refines the ticket first and stops to ask when something is ambiguous.
/review:changes
Reviews your local diff before the PR with focused reviewers (bugs, security, tests, quality, architecture, spec, overbuild) and merges their verdicts.
/commit
Stages your changes and commits them as logical Conventional Commits, after showing you the split.
agent-config doctor
Checks the install and reports anything missing or changed on disk, with the next command to fix it.
When a request is unclear the agent stops and asks with numbered options instead of guessing. Implement a change →
Use it in your project
Start with the Quickstart below; the full walk-through is on the site. Install →
Prove it
Every measured figure in this README carries a marker that points at a recorded, re-runnable piece of evidence. The build fails when one stops resolving. The proof page lists each claim with its evidence, including the runs where the package changed nothing, plus commands you run yourself on a fresh checkout. Verify it yourself →
Contribute
Skills, rules and commands are plain Markdown under src/; one command rebuilds every tool's copy and runs the checks. Development has the loop. Contributing guide →
Quickstart
bash
npx -y @event4u/agent-config init # browser wizard on a desktop
npx -y @event4u/agent-config init --tools=claude-code,cursor # no wizard: CI, scripts, servers
init finds the AI tools you have installed and selects them for you. In CI, without a terminal or with --yes or --tools=<ids> it installs without the wizard; both paths run the same installer. install is the same wizard under another name.
Installer default: user-global. agent-config init installs into your home directory — the package under
~/.event4u/agent-config/ and each AI tool's user-level folder such as
~/.claude/ or ~/.cursor/ — and writes no skills, rules or commands into
your repository, which only gets an agents/overrides/ folder for your own
changes plus a few gitignored bookkeeping files and a block each in
.gitignore and .gitattributes; agent-config init --project creates just
those in a repository and nothing else.
Then, in your AI tool:
text
/work "your first real task"
Later
Command
Change role, profile or tools
agent-config setup
Open the settings screen
agent-config config
Update to the latest release
agent-config upgrade
Check the install
agent-config doctor
Remove it
agent-config uninstall --global (--purge also deletes the files it recorded writing )
Agent Config is a set of skills, agent rules, slash commands and guidelines that it writes into each AI tool's own format. Skills use the open Agent Skills SKILL.md format with a few package-specific frontmatter keys.
It is not an agent. Your AI agent still runs the loop, calls the LLM and uses its own tools; the package changes what that agent is told to do. It does not lock you to a model or a vendor.
What's inside
How the counts in the badges at the top are counted — one counter, update_counts --check, re-derives all six from the tree and fails the build when one is off by one. Commands 214 counts every command file, including the ones in subfolders; all 213 are listed in the Commands section of the catalog page. Rules 122 counts the source rules, while the linked folder holds 121 because one rule is switched off and not shipped. Personas 29 leaves out the directory README. These are counts of files: none of them measures quality or adoption.
You do not get all of it at once. On tools without a way to hide skills, a new install ships a smaller set chosen for your role; Claude Code and Codex get every skill and show them as needed. Browse the catalog →
Customize
Rule level.--profile minimal | balanced | full at install sets the rule level: how many rules load on every task. It is separate from the role profile above; balanced is the default and full costs the most tokens.
Your own changes. Put them in agents/overrides/ in your repository instead of editing the installed files.
Settings.agent-config config opens the settings screen with search and reset.
Skill sets for your role. Add a set of skills for another domain without switching profile.
init detects 23 AI coding tools and works with 20 of them. What a tool gets depends on its tier:
18 tools get their own format written into their user-level folder. Which of rules, skills and commands each one reads is in the table below. Only Claude Code can block a turn; on the others the rules are instructions.
Codex CLI is one of those 18 but reads skills only.
GitHub Copilot gets a discovery plugin: a read-only MCP server that points to the install.
Claude Desktop gets skill ZIPs that you upload yourself under Settings → Customize → Skills.
Aider, Zed and JetBrains are export-only: you add the exported file to their config yourself.
Tool
Rules
Skills
Commands
How it works
Claude Code
✅
✅
✅
Reads .claude/
Cursor
✅
—
☑️
Reads .cursor/rules/ + commands via AGENTS.md
Cline
✅
—
☑️
Reads .clinerules/ + commands via AGENTS.md
Windsurf
✅
—
☑️
Reads .windsurfrules + commands via AGENTS.md
Gemini CLI
✅
—
☑️
Reads GEMINI.md
GitHub Copilot
✅
—
☑️
Discovery plugin (read-only MCP server, no hooks); project rules via agent-config export --tool=copilot-instructions into .github/copilot-instructions.md
Roo Code
✅
—
☑️
Auto-discovers .roo/rules/*.md + AGENTS.md
Codex CLI
—
✅
—
Reads skills from ~/.codex/skills/; rules and commands are delivered but not read, and no AGENTS.md is written (your project's own AGENTS.md is still read)
Continue.dev
✅
—
☑️
Auto-discovers .continue/rules/*.md + AGENTS.md
Aider
📌
—
—
Export-only: add the file to read: in .aider.conf.yml
Augment (VSCode/IntelliJ)
📌
—
—
Global-only; project writes marker
Claude Desktop
—
📦
📦
Skill ZIPs (commands bundled as skills) that you upload yourself under Settings → Customize → Skills
✅ native ☑️ text reference (in AGENTS.md, not a native slash command) 📌 marker or export only 📦 ZIP you upload yourself — not available
Plugins. Augment CLI can also install it as a plugin with rules, skills and commands. GitHub Copilot gets a discovery-only plugin: a read-only MCP server and a skill that points to the install, with no hooks. For per-project Copilot instructions run agent-config export --tool=copilot-instructions --output=.github/copilot-instructions.md. The Claude Code marketplace plugin is deprecated because the npm install already carries the content and the hooks; agent-config doctor flags a leftover plugin.
Teams. Every tool you install is recorded in agents/installed-tools.lock; a new team member runs npx @event4u/agent-config sync after cloning.
Cloud sessions. A Claude Code cloud session (claude.ai/code, claude --cloud) starts on a fresh clone and does not see the home-directory install, so it runs without agent-config skills, rules and hooks. See cloud sessions.
agent-switch runs several agent accounts on one machine, each in its own profile. The two compose: agent-switch keeps the accounts apart and agent-config shapes what the agents do inside them. Its settings stay scoped to the active profile. How they work together →
Who this is for
Anyone whose AI coding agent says "done" too early. The habits — analyse first, test, verify, commit cleanly — work in any stack. The deepest skill sets are for Laravel and modern PHP (Pest, PHPStan, Rector, Eloquent, Livewire), then Symfony, Next.js and React. Cross-stack skills cover API design, testing, security, databases, Docker, Git, CI, review, threat modeling and prompt engineering.
The skills come from production work on TypeScript and PHP codebases. Treat coverage on other stacks as promising, not proven. Tell us where it falls short. Discussions →
Legal and safety notes
Legal pack — not legal advice. The EU/DE legal skill set (contract, NDA and DPA review) is a research and drafting aid only. It does not replace a qualified lawyer and must not be relied on for a concrete matter. Read the legal notice (LEGAL_NOTICE.md) before use.
Start with agent-config doctor: it flags a binary missing from PATH, version mismatches and leftovers, each with a one-line fix. If a rule or hook did not fire, agent-config routing:doctor reports which checks are active and why. Upgrade problems such as a missing command or a skill listed twice: troubleshooting.
Development
Working on the package itself? Edit src/ only and rebuild the generated trees:
bash
task sync# regenerate dist/agent-src/ and .augment/
task generate-tools # regenerate .claude/, .cursor/, .clinerules/, .windsurfrules
task ci # full pipeline — green before PR
task test# unit + integration tests
task dev:setup # boot the onboarding wizard against the working tree
From a source checkout, run the CLI as ./agent-config <command>; npx @event4u/agent-config does not resolve there without npm link. Build it with npm run build:cli if dist/cli/agent-config.js is missing. The whole stack is TypeScript.
Node 22.6 or newer.npx -y @event4u/agent-config init is the install path; nothing else is needed.
Platform: macOS 12.3+, Linux and WSL2. Git Bash needs Developer Mode for symlinks. Contributors also need Task.
Windows. Native PowerShell and cmd are not supported for the file install; use WSL2. Without WSL2, point any MCP client at the built-in MCP server and the skills, rules and commands are available as prompts and resources:
The package is free, MIT and stays that way — no paid tier, no dual licensing. If it saves you time and you want to chip in, the GitHub Sponsor button at the top of the repo is the whole mechanism. If you would rather not, use it anyway; nothing here is gated on it.