Imported from baranovxyz/agent-context-kit (plugins/agent-ready-repo/skills/using-agent-ready-repo/SKILL.md). Install upstream with npx skills add baranovxyz/agent-context-kit --skill using-agent-ready-repo. Copyright stays with the author.

Using Agent Ready Repo

Use this skill as the routing gate for repository-memory work.

Core Model

Apply the layered model before editing:

1. AGENTS.md is the first-class portable instruction contract.
2. docs/ is durable memory.
3. Tool-specific files such as CLAUDE.md contain host-specific loading or capability wiring.
4. Skills encode repeatable procedures.
5. CLI, MCP, hooks, and CI expose executable contracts or enforcement.
6. Agents are autonomous workers for complex multi-step work.
7. Plugin commands are legacy flat skills; prefer skills/*/SKILL.md for new workflows.

Consult ${CLAUDE_PLUGIN_ROOT}/references/layered-architecture.md for details.

Routing Procedure

1. Identify the user's actual intent.
2. Classify the artifact using ${CLAUDE_PLUGIN_ROOT}/references/artifact-decision-matrix.md.
3. Choose the smallest applicable plugin skill:
   • use agent-ready-audit for repo-wide inspection or migration planning;
   • use agent-ready-migrate only for an explicitly approved migration plan;
   • use agent-ready-maintenance after project changes;
   • use monorepo-agent-context for workspace/package-level instruction layering;
   • use skill-and-agent-designer for repeated procedures or worker roles;
   • use tool-specific-wiring for Claude, Cursor, Copilot, Gemini, hooks, MCP, or host files;
   • use wrap-up-session to close out a session (audit, route findings, write continuation prompt, commit, PR, merge);
   • use resume-from-continuation to start a session from a previous continuation prompt.
4. Do not add content to AGENTS.md by default.
5. Prefer moving knowledge to the right layer over duplicating it.

Red Flags

Stop and classify before editing when seeing:

• a growing AGENTS.md;
• copied rules across CLAUDE.md, .cursor/rules, and .github/instructions;
• duplicated root and package-level instruction files in a monorepo;
• requirements living only in chat;
• a repeated checklist;
• a new role that would benefit from isolated context;
• a rule that must always run;
• an MCP or CLI decision framed as protocol preference instead of job-to-be-done.

Output

For small tasks, state the chosen layer and proceed.

For larger tasks, produce:

• selected skill or agent;
• affected artifacts;
• proposed changes;
• required verification.