mission.md

Mission - Reviewer Agent: Separate Builder from Marker

Goal

Build a reviewer loop that reads the builder's artifacts read-only and emits a review_report.json scored across five dimensions, totalling out of 10, with a verdict of pass, soft_fail, or hard_fail.

Inputs

• ReviewerInputs bundling diff, state, feedback, and verification verdict from prior lessons
• Rubric dimensions: problem fit, scope discipline, assumptions, verification quality, handoff readiness

Deliverables

• One scoring function per dimension (stub-grade for the lesson, deterministic)
• review_report.json writer with five scores, total, and verdict
• Two demo cases: a clean change and a "right tests, wrong problem" change

Acceptance

• python3 code/main.py exits zero
• The clean change scores at least 7 with verdict pass
• The wrong-problem change drops below 5 on at least one dimension and verdict flips to hard_fail

Out of scope

• Real LLM calls. The lesson stubs each dimension; the skill swaps in a model later.
• Editing the diff. The reviewer reads, scores, and reports. Patches are the builder's job next turn.

References

• docs/en.md - full lesson
• code/main.py - reference implementation
• outputs/skill-reviewer-agent.md - extracted skill