## What it is
agent-skills is the whole rollcall family in one install — skill-rollcall, mcp-rollcall, settings-effective, and git-attribution. Each one makes a step that a coding agent does silently legible again — and each reads the files of whichever agent you're in: Claude Code, Codex, Cursor, Gemini CLI, Copilot, VS Code, Windsurf, OpenCode.
## The through-line
Every coding agent registers skills, starts MCP servers, merges settings files, and signs commits — and tells you nothing about any of it until something's missing. Each also keeps those things somewhere different (~/.claude, ~/.codex, ~/.agents, ~/.cursor, ~/.gemini). Each tool here rebuilds one of those silent steps from disk, with the reasons attached, then repairs or reports:
- skill-rollcall — which skills registered for which host, which are new, which will never show up and why, and whether a skill folder passes the Agent Skills spec.
- mcp-rollcall — which MCP servers will fail to connect and exactly why, and what each costs in tools and context, read from every host's own config shape.
- settings-effective — every Claude Code setting actually in effect, which file decided it, and why the value you set isn't applying (runs from any host).
- git-attribution — whether AI co-author trailers are landing in your history, where each agent's switch lives (Codex's is a workspace policy with nothing local to flip), which commits carry them, and how to scrub and block them.
## How it's built
Every tool is a single stdlib-only Python script that also runs as a plain CLI — read-only unless you pass --apply, with --json and --strict for tooling and CI. This repo is the family in one install; each skill also lives in its own repo, which is the source of truth. The copies here are synced from upstream, pinned in a lock file, and CI fails if they drift. The layout is the Agent Skills standard, so one npx skills add Londopy/agent-skills installs into any of 79 agents.