feat(review): restore the canonical task for every worker; keep it runtime-only

1faf416·

peyton-alt·1w ago·4 files·+19 added/-36 removed

Product decision revisited with the A/B evidence in hand: the built-in review brief was #1312's deliberate design, and the un-briefed variant ran ~35% faster but missed a real finding in the measured comparison. Depth wins — every worker (skill-bearing included) again receives the canonical profile task, restoring #1312's reviewer behavior exactly.

What stays from the earlier change: the built-in text remains a RUNTIME default — guided setup, first-run, and configure no longer persist it into saved profiles (that part was a defect: baked text masquerading as user config, freezing profiles on old wording and making the default uneditable). Users who want a lighter pass set their own shorter task.

Co-Authored-By: Claude Fable 5 noreply@anthropic.com

Sessions

5ad2ee054f94View transcript

Changes

4

• cmd/entire/cli/review

  • Mcmd.go+10/-9

  • Mcmd_test.go+8/-9

  • Mprofile.go-17

• docs/architecture

  • Mreview-command.md+1/-1

Behavior

• Reviewers run the profile task. A user-configured task reaches every worker verbatim; when task is empty, skill-bearing workers get NO injected brief (their skills define the behavior, matching a direct user invocation) and only skill-less workers fall back to the built-in default task.
• Reviewers run the profile task: the user-configured task when set, else the built-in default brief. The built-in text is a runtime default — setup never persists it into saved profiles, so profiles stay clean and pick up improved defaults.
• Multi-reviewer profiles run reviewers concurrently, then run one judge.
• Results are printed and saved locally; output: "trail" also posts findings to the branch trail.
• A bare non-interactive entire review does not auto-run a profile. Automation should pass a profile name.