| name | analyze-repo-context |
| description | Analyze a repository root and generate high-signal repository AGENTS.md / CLAUDE.md files with machine-readable app metadata. |
analyze-repo-context
Generate repository-wide context files for AI agents.
Output Contract
- Produce a markdown document suitable as
AGENTS.md / CLAUDE.md.
- The document must include:
- A fenced
yaml block with:
apps: list- each app entry containing
name, path, purpose, stack
- Human-readable sections:
- Project Overview
- Repo-Level Conventions
- Shared Tooling and Commands
- Cross-App Boundaries and Integration Rules
- Testing and Quality Gates
- Out of Scope
Quality Bar
- Be concrete, repo-grounded, and specific to observed code and tooling.
- Avoid generic advice; include enforceable conventions and actionable guidance.
- Prefer short, precise rules over broad narrative.
- If a fact is uncertain, state the uncertainty instead of inventing details.
Task-Specific Instructions
- Input must be a repository-level path. If no path is provided, or the path is missing/invalid, refuse and explain exactly what is wrong.
- Treat the input path as the scope root where output files must be written.
- Detect vendor directories at that root:
- If
.claude exists, generate CLAUDE.md.
- If any of
.codex, .cursor, .github exists, generate AGENTS.md.
- If none of these vendors exist, generate
AGENTS.md.
- Generate repository-level guidance and include a machine-readable fenced metadata block (
yaml preferred) with top-level apps: entries.
- Each app entry must include:
name, path, purpose, stack.
- Keep app paths repo-relative from the analyzed scope root.
- Output files must be created directly at the analyzed scope root (not inside
rnd/ or r3nd/).
- If both files are required, keep content semantically equivalent across both files.
- Keep the metadata block concise and valid so it can be parsed programmatically.
{{rnd/agents/shared/command-hygiene.md}}
