How-To Guides

Validate a new machine

npm run harness -- doctor
npm run harness -- init --profile local
npm run harness -- doctor --config harness.config.json

Fix one reported prerequisite at a time. If model services are intentionally offline, use the offline flags shown in the CLI reference.

Try the product without touching a project

npm run harness -- demo --list-tasks
npm run harness -- demo --dry-run --task ts-isUUID

Use a dry run to learn the task shape before pointing the harness at a real repository.

Keep a task bounded

1. Give the worker one outcome and a narrow file scope.
2. Name the exact verifier command in the task.
3. Run once and inspect the report.
4. Escalate or revise the task only after reading the evidence.

Recover from a failed run

1. Run npm run harness -- status to identify the job.
2. Run npm run harness -- report to inspect the structured result.
3. Check the verifier output and the repository diff.
4. If the model endpoint failed, restart the provider and rerun npm run harness -- doctor.
5. Retry the smallest reproducible task, not the entire project.

Connect a local model

Configure the provider and model in the generated config, then verify the endpoint with npm run harness -- doctor. Keep API keys in environment variables or the provider's credential store. The public docs intentionally do not publish private deployment URLs, keys, or internal routing policies.