todo2code

Preprompt and technical directives (ticket-002)

• Task title: Cross-repository semantic hardening
• Created: 2026-07-31T06:49:07Z
• Governance source: wellmanifest/new-project

Requirements and constraints

1. Test todo2code on real external repositories through deterministic, reproducible runs.
2. Capture a comparable baseline before changing semantic behavior.
3. Classify observed failures and select one shared, measurable defect.
4. Add an independent regression case before implementing its fix.
5. Apply one semantic change at a time and repeat gold plus corpus measurements.
6. Reject an attempted improvement when it increases noise or lacks measurable external benefit.
7. Preserve external repositories, secrets, untracked files and current user changes.
8. Keep raw command output in the provider-specific ticket log.

Referenced specifications

• docs/READINESS.md
• docs/TEST_REPORT.md
• evaluation/gold/README.md
• evaluation/gold/v2/dataset.json
• TODO.md
• Governance policy: wellmanifest/new-project/POLICY.md
• Governance procedure: wellmanifest/new-project/CONTRIBUTING.md

Execution boundary

The planning state is WAIT_FOR_APPROVAL. Under P-CORE-008, no source-code change or external benchmark execution begins until the user approves ai-codex.md and the project-level ticket entry in TODO.md.

This site is open source. Improve this page.