Three levels, one explicit boundary
- 1 Passive Runs inspect to check startup, lifecycle, catalogs, descriptions, and schemas without calling tools. This is the recommended baseline for every pull request.
- 2 Standard Adds repository-owned check scenarios that replay known inputs against exact selected tools. Start with read-only behavior or disposable test data.
- 3 Full Adds targeted break and reject runs for broader edge-case and invalid-input coverage. Use an isolated test server, disposable data, and explicit authority.
Choose the lightest mode that answers the question
Keep Passive on pull requests. Add Standard when known behavior needs replay. Reserve Full for isolated environments where broader active testing is worth the additional authority.
These are coverage labels—not a --mode switch
Your repository composes each level from explicit commands and reviewed scenarios. Full never means “call every tool”; every active tool, effect, case, and seed remains intentional.