agents-mdlisted

Create, maintain, and audit a repo's AI-agent contract — one canonical AGENTS.md body plus a CLAUDE.md pointer stub that imports it. Use to write, update, audit, or drift-check AGENTS.md or CLAUDE.md, or to collapse a dual-body pair into the single-body model. Defers net-new repo scanning to native /init. Not for prose docs — use `docit` (public) or `knowledge` (internal).
jwogrady/spark · ★ 0 · AI & Automation · score 66

Install: claude install-skill jwogrady/spark

# agents-md — the agent behavioral-contract skill A repo speaks to its AI contributors through **one canonical contract body**: - **`AGENTS.md`** — the contract. Tool-agnostic: repo map, commands, workflow, and the doctrine every AI agent must follow. Any agent reads it directly. - **`CLAUDE.md`** — a pointer stub whose body is `@AGENTS.md`, so Claude Code imports the same contract other tools read. Claude-specific notes may follow the import, but only when genuinely tool-specific. One body means the two files **cannot drift** — the dual-body model this skill previously maintained required editing every rule twice, and the two paid a truth-pass tax on every change. This skill owns both files. ## Division with native `/init` Claude Code's native `/init` already *creates* a first `CLAUDE.md` by scanning a repo. **Do not reimplement that.** This skill's jobs are the ones `/init` does not do: - **Author `AGENTS.md`** — the canonical body has no native generator. When `/init` produced a full `CLAUDE.md`, move its body into `AGENTS.md` and leave the `@AGENTS.md` stub behind. - **Maintain & audit** the contract — patch missing sections, refresh stale content, inject Spark's required doctrine (attribution, GitHub guardrails, agent safety). - **Migrate** a legacy dual-body pair (full `CLAUDE.md` + full `AGENTS.md`) to the single-body model: merge, dedupe, keep the stricter rule on conflict, and present the diff before writing. ## What the contract carries The