Keep evaluation useful after the first run

Read the operating agreement, maintenance guide, and managed workflows when applicable.

For bundled local tools and exact commands, read toolkit setup. Run the workspace installer for this skill; it preserves existing customized files. For the managed route, use concrete service operations after confirming the connected schema.

Establish ownership

Identify the existing CI, release process, production observability, reviewer, and baseline. Reuse them. Agree on which failures block a release, which merely flag review, and how infrastructure errors are handled. A failed request is never a silent passing check.

Regression checks

Create a small representative suite plus known-failure regressions. Pin cases, grader versions, app configuration, and independent confirmation rules. Separate fast deterministic checks from slower paid evaluation. Expose skipped or missing-credential runs rather than reporting a pass.

For managed PR workflows, inspect current Confident Actions setup and generated callback/workflow. Confirm it invokes the actual app and reports the intended check. For local workflows use the project's CI and runner, with credentials kept in its secret store. Do not expose paid secrets to untrusted pull-request code.

Production feedback