Problem-solving guide

How to verify AI-generated code changes

AI-generated code can look plausible while violating repository conventions or breaking tests. Use minitok when a change must pass the repository's own checks before acceptance.

Short answer: initialize the repository, preview the task, inspect the plan and changed files, then review the configured project verification result. A passing check is evidence, not a correctness guarantee.

Workflow

  1. Run minitok doctor.
  2. Run minitok migrate.
  3. Preview with minitok run --dry-run "Add the requested change and tests".
  4. Review scope, checks, stop conditions, and evidence.
  5. After approval, run without --dry-run and inspect the result.

Commands

npm install -g @flotic/minitok
minitok doctor
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

What to verify

  • The plan uses the real repository structure and constraints.
  • VERIFY_CMD.mjs or the configured check belongs to the project being changed.
  • The changed-file list is within the expected scope.
  • Failed checks cause a stop or bounded repair rather than being hidden.
  • The result identifies checks and remaining human-review work.

Limits

For a one-line edit with no repository checks, a simple editor may be faster. A passing check does not prove security, deployment, business correctness, or absence of defects. Provider credentials and model usage are separate from minitok.

Related: approve and recover changes, record evidence, and the demo. See documentation, source, and privacy.