Reviewer Selection

Always spawn the default reviewer gang, then add only the conditional personas that can challenge the change from a distinct angle.

Default Set

Spawn these for every Review Gang run:

• reviewers/general.md
• reviewers/tests.md
• reviewers/silent-failures.md
• reviewers/code-shape.md

The default gang is mandatory even for small diffs. The silent-failures lens is especially important when the diff touches error handling, retries, fallbacks, validation, auth, external services, or user-facing failures. The code-shape lens keeps module boundaries, single source of truth, and maintainability risk visible without turning cleanup into a style pass.

Add Conditional Personas

Add reviewers/types.md when:

• new types, schemas, or DTOs are introduced
• API contracts changed
• invariants moved into or out of types
• a refactor changed data boundaries
• the diff uses any, as, unknown, or other type escape hatches

Add reviewers/cleanup.md when:

• the diff is a refactor, migration, or generated-looking change with helper churn
• dead code, stale flags, duplicate branches, or unused exports look plausible
• the change introduces wrappers or abstractions that may add indirection without value
• you suspect the right fix is deletion or merging paths rather than adding more code

Add reviewers/comments.md when:

• docstrings or code comments changed materially
• large comment blocks were added
• the change leans on comments to explain tricky behavior

Shape-Based Add-Ons

• UI feature → default gang
• API / backend → default gang; add types for schema changes and cleanup for wide refactors
• State / migration / config → default gang; add types if invariants changed and cleanup if old paths may be left behind
• Refactor / cleanup → default gang + cleanup; add types when invariants changed
• Doc-heavy change → default gang + comments
• Tiny mechanical change → default gang only, unless comments/types/cleanup add a distinct concern

Do Not Add Everything Blindly

More conditional personas are not always better.

Avoid conditional persona spam when:

• the diff is tiny
• personas would repeat the same concern
• coordination overhead exceeds review value

Every added persona should contribute a distinct lens or evidence source.