How to approve and recover repository changes
AI coding workflows need clear answers to when a generated change may be applied and what happens when verification or patch application fails.
Safety boundary: normal minitok runs do not require automatic acceptance.
--auto-accept bypasses the normal approval prompt and should be used only when intentional.Preview before approval
cd your-repository
minitok migrate
minitok run --dry-run "Refactor authentication and preserve the public API"
minitok statusInspect the plan, target repository, configured checks, proposed scope, and stop conditions before a real run.
Use an explicit approval record
minitok run "Refactor authentication and preserve the public API" \
--approval-file .minitok/approval.jsonApproval covers one complete generated change set. File-by-file cherry-picking or hunk-level selection is not currently supported.
Recover from failure
- Failed verification stops or guides bounded repair according to cycle, token, and time limits.
- If a run cannot apply its generated change set, inspect the preserved
.minitok/last-run.patchand apply it manually only after review. - Use
minitok run listandminitok run show <run-id>to inspect state; the latest structured evidence is also available at.minitok/last-run.json. - A stopped run is not an applied change or a deployment result.
When this helps
Use these boundaries for multi-file changes, project tests, second-person review, or an inspectable change record. A tiny local edit may be faster manually. minitok does not guarantee correctness or replace human review.
Related: verify changes, record evidence, and safe runs. See the source repository.