MCAF skill
A framework for working with coding agents: a root AGENTS.md, 18 mcaf skills and tests as the decision makers.
- 55 ★ on the repository
- Last commit
- CC-BY-4.0
Minimal access
Why this level
- The set consists of instructions and templates, but during install the agent downloads files from the tutorial address.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/mcafDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The set consists of instructions and templates, but during install the agent downloads files from the tutorial address.
This is third-party code. Review the repository files before installing.
What it does
MCAF (Managed Code Coding AI Framework) describes how to run AI-assisted development predictably. Context, verification and instructions live in the repository: code, architecture docs, ADRs, root and local `AGENTS.md` files, skills. Integration tests and static analysis decide, not opinions. Version 1.3 is built around skills. The bootstrap set holds a root `AGENTS.md` template, a `CLAUDE.md` wrapper and a tutorial page that lets an agent download the needed skill folders itself. Skills prefixed `mcaf-` cover feature specs, ADRs, architecture overview, CI/CD, testing, code review, observability, a security baseline, non-functional requirements and ML project delivery.
Who it is for. Teams that pin agent working rules in the repository and want one skill set for specs, tests and review.
Good fit when
- A root `AGENTS.md` and shared working order for agents are introduced in an existing repository.
- Before a large feature the agent writes a spec with REQ and AC identifiers, and only then code.
- A multi-project solution gets a local `AGENTS.md` for each project.
- Shared review checklists, ADRs and security baselines are needed for every team agent.
Not a fit when
- The process only gets in the way of a small script or a one-off experiment.
- Rules should arrive from a one-command installer: version 1.3 has no install script.
- The team does not keep decisions and specs inside the repository.
Example request
Install or update MCAF in this repository following the guide at mcaf.managed-code.com/tutorial and do not weaken the rules already in AGENTS.md.Limitations
There is no shell installer: installation follows the tutorial address, the agent downloads templates and skill folders itself and the agent must be restarted afterwards. Some .NET skills live on a separate site. The framework demands discipline: specs with REQ and AC, ADRs and test verification are mandatory before agents write code. The README covers concepts, while the install procedure sits in TUTORIAL.md. Rules already in `AGENTS.md` must be merged by the agent without weakening them. The CC-BY-4.0 license requires attribution when reusing the texts.
How to disable. Delete the `mcaf-*` skill folders from the agent skills directory and the template blocks from `AGENTS.md` and `CLAUDE.md`. Rules the team chose to keep can stay.
FAQ
How do I install MCAF in Claude Code?
Paste the first prompt from `TUTORIAL.md` to the agent. It downloads the `AGENTS.md` template, the `CLAUDE.md` wrapper and the needed `mcaf-*` skills, after which you restart Claude Code.
Which skills are included?
Eighteen skills prefixed `mcaf-`: ADRs, feature specs, architecture overview, CI/CD, testing, code review, observability, security, non-functional requirements, UI/UX and more.
What are the three elements of MCAF?
Context, verification and instructions. Context lives in the repository, tests and static analysis verify, and the root and local `AGENTS.md` files set instructions.
Does MCAF suit Codex and Cursor?
The base lives in `AGENTS.md`, which many agents read. A `CLAUDE.md` wrapper is added for Claude Code; for other agents you must check the skills yourself.
Can I take only some skills?
Yes, skills sit in separate folders. The process is still designed as a chain: spec, ADR, tests and review complement each other.
Discussion
No questions or reviews yet.
Related
SkillsOver twenty skills for coding agents: MCP and hooks, plan mode, a bug-fix protocol, prompting and a bash safety hook
AI Code Toolkit
aicode-toolkit
Plugins and MCP servers for coding agents: template-based projects, patterns before a file edit and review against YAML rules.
Kode CLI is a terminal coding agent that reads AGENTS.md and CLAUDE.md, connects MCP servers and works with many models.
A library of 30-plus instruction skills for coding agents: planning, building, testing, hallucination guards and shipping
Work modes for AI coding agents: a light solo mode for daily tasks and /man with a plan, checks and review for big changes
Tasks, docs, specs and decisions live as markdown files in your repository, and the agent reads them through the CLI and MCP
README in short
The concepts document is split into context, verification and instructions. The install procedure lives separately in TUTORIAL.md. The `skills` directory holds a `SKILL.md` folder for every process.