Use this skill when an operator needs a repeatable, version-controlled way to keep AI coding-agent capabilities synchronized across multiple runtimes. CAPA reads a project-level capabilities.yaml, resolves and locks referenced capability content, then writes each agent’s native configuration files, including Cursor rules, Claude Code agents/settings, Codex AGENTS.md content, and MCP endpoint registrations.

## What the agent/operator does

1. Install CAPA on macOS, Linux, or Windows.
2. Run capa init in a repository to create a capabilities.yaml.
3. Declare skills, rules, tools, sub-agents, MCP servers, and plugins in that manifest.
4. Run capa install to resolve pinned content, update capabilities.lock, and generate provider-native files.
5. Use capa sh to list or invoke configured tools directly when an agent or operator needs a shell-facing tool entry point.
6. Review CAPA-managed marker blocks before committing generated configuration changes.

## When to invoke

Invoke this instead of hand-maintaining separate agent config files when a repository supports multiple AI coding agents, when teams want reproducible capability setup on clone, or when an operator needs to fan out the same approved tools and rules into several provider-specific formats. It is especially useful during repo onboarding, agent-environment standardization, and capability drift cleanup.

## Scope boundary

This is not a generic package-manager listing. Keep the workflow scoped to authoring and installing capabilities.yaml for AI coding-agent environments, validating generated native files, and using CAPA’s lock/cache behavior to keep capability setup reproducible. Do not broaden it into general dependency management or unrelated CLI orchestration.

## Requirements

– CAPA installed locally.
– A project repository whose agent capabilities should be managed.
– At least one supported agent runtime such as Cursor, Claude Code, Codex, Windsurf, GitHub Copilot, or Gemini CLI.
– Review discipline for generated provider files before commit.

## Installation

Install from the upstream release or review and run the installer published at capa.infragate.ai. Then run capa init and capa install in the target repository.

## Verification notes

Source metadata was verified from the exact GitHub repository and README. As of this intake run, infragate/capa had 692 GitHub stars, current push activity on 2026-08-01 UTC, a latest release 15 days old, MIT license in README badges, official docs at capa.infragate.ai, and README evidence for capabilities.yaml, capa install, generated native files for multiple AI coding agents, marker blocks, lockfile/cache behavior, filtered MCP endpoints, and capa sh.