Agent Receipt

1. Gather only evidence available in the current run/repo.
2. Never infer a passed test, deployment, duration, or intervention count. Use unknown instead.
3. Write a local JSON manifest with: project, agent, duration_minutes, interventions, files_changed, tests_passed, tests_total, verification, deployed, skills_used, and evidence.
4. Run scripts/receipt.py <manifest> --out <public-result.json>. The public result contains evidence hashes and counts, never raw evidence values.
5. Run scripts/render.py <public-result.json> --svg <card.svg> [--png <card.png>].
6. Keep the source manifest and raw evidence local unless the user explicitly approves publishing selected evidence.
7. Publishing is a separate external action. Confirm destination and visibility immediately before uploading unless the current request already authorizes it.