How to verify AI-generated test results
Check which project verification command ran, what changed, and what evidence remains before accepting an AI-assisted repository change.
Problem and general approach
An AI agent can report that a test passed without showing the repository, command, changed files, or review context. Run the repository's deterministic check, inspect the proposed scope, and preserve the result with the task record.
A passing check is evidence about that check at that point in time. It is not proof of correctness, security, deployment success, or production readiness.
Where ad-hoc workflows fall short
A terminal transcript or model summary can omit the exact command, hide retries, mix output from another directory, or lose a failed patch. Re-running tests without the task and changed-file context makes later review harder.
When minitok fits
Use minitok when an existing repository needs an explicit plan, project-specific checks, review, bounded repair, and an inspectable run record. It is not necessary for every one-line edit or unrelated test question.
cd your-repository
minitok migrate
minitok run --dry-run "Add a health-check endpoint, tests, and documentation. Preserve existing APIs."
minitok status
minitok run list
minitok run show <run-id>Inspect the configured project check, changed files, retries, stop reason, and remaining human review. The latest redacted summary is .minitok/last-run.json; structured evidence may be under .minitok/evidence/runs/.
Alternatives and limits
Run the repository test command directly when no planning or run record is needed. Provider credentials and model usage are separate from minitok.
Related: verification guide, evidence guide, case study, and documentation.