Cursor / Codex Setup¶
Use this guide to configure Cursor, Codex, and similar repository agents for safe, reproducible RQM work.
Recommended .cursorrules / agent instructions¶
You are an agent working in the RQM docs/code ecosystem.
Always:
- Read /llms.txt first.
- Verify API behavior via API Overview and Swagger UI.
- Use documented payload examples before drafting new payloads.
- Run validate-before-optimize flows for circuit workflows.
Never:
- Invent undocumented endpoints or undocumented schema fields.
- Treat internal optimizer IRs as public payload contracts.
- Present research hypotheses as production guarantees.
Output quality:
- Keep production behavior and research notes clearly separated.
- Add reproducible artifacts when possible (request examples, output reports, diffs, tests).
File reading order¶
/llms.txt/docs/api/overview.md/docs/platform/canonical-circuit-boundary.md/docs/platform/optimization-engine.md/docs/platform/execution-bridges.md/docs/products/rqm-studio.md- Relevant domain page(s) and examples
Safe edit boundaries¶
- Safe: docs copy, examples, guides, SDK snippets, integration recipes, testing docs.
- Caution: API route names, request/response schemas, and trust-language claims.
- Do not modify: production guarantees or endpoint inventories unless those changes are explicitly documented and verified.
Expected implementation behavior for RQM client code¶
- Start from documented public payload contracts.
- Perform validation before optimization or execution routing.
- Keep backend-specific assumptions in explicit bridge layers.
- Surface verification/trust artifacts whenever available.
- Fail conservatively when inputs are invalid or capabilities are unknown.
Testing expectations¶
- Run project checks (for this repo,
mkdocs build --strict). - Keep examples runnable or clearly marked as pseudocode.
- Ensure links, endpoint names, and payload fields match documentation.
Adding examples without inventing endpoints¶
- Pull endpoint names from API Overview and Swagger UI.
- Reuse documented payload shapes and adapt only values, not schema structure.
- If a needed endpoint is absent, document a gap and propose it as planned instead of fabricating it.