Pilot UX behavior contracts and auto-file violations #68

Outcome

Pilot typed UX behavior contracts on the OpenAgents issue tracker and file a forge issue automatically when a reviewed contract fails.

Current behavior

User expectations live in conversations, issue comments, and product copy. Regression tests cover selected behavior, but no standing contract connects a user-visible expectation to an oracle, a failure receipt, and a deduplicated tracker issue.

Source: docs/episode-triage.md, episodes 246 and 252.

Contract

• Start with a small reviewed registry for issue list, issue detail, project board, and notification journeys.
• Store each expectation, environment, oracle, falsifier, evidence policy, and owner as a typed artifact.
• Require human review before a mined conversation statement becomes a release gate.
• Run the contracts against the exact staging candidate.
• Create or update one issue per contract ID when an oracle fails.
• Link the failure evidence without copying private conversation content.
• Close or annotate the issue only after the same contract passes on a qualified candidate.

Acceptance criteria

• At least five tracker behavior contracts run as ordinary release checks.
• A deliberate violation files one deduplicated issue with the contract ID, candidate revision, and bounded evidence.
• Repeated failures update the existing issue instead of creating spam.
• Private source conversations never appear in public issue text or artifacts.
• A passing rerun records recovery evidence and follows the normal issue closure policy.

Verification

Test contract parsing, staging execution, deduplication, privacy, failure filing, recovery, and unauthorized mutation. Run the pilot through one exact candidate and mix precommit.

Dependencies

Build on the accepted-outcome contract and issue timeline work in project 3.