Agent
Reviews a context-relay document with fresh eyes and raises only substantive, non-trivial ambiguities a brand-new Claude instance couldn't resolve by reading the repo itself. Use when the context-relay skill is iteratively refining a relay doc. Returns either a numbered list of questions or the literal token…
Agent
Architecture lens for review-branch. Reviews the diff for layering/boundary violations. Operates in one of two modes: hex mode (when the project follows hexagonal architecture, detected by the orchestrating skill) checks ports/adapters discipline; otherwise general mode checks layer separation, dependency direction…
Agent
Test coverage lens for review-branch. Identifies new behavior introduced by the diff, evaluates whether the added tests actually exercise it, and flags meaningful gaps. Catches the 'Fake ignores the new parameter' anti-pattern and missing roundtrip tests on persistence boundaries.
Agent
Naming and API ergonomics lens for review-branch. Reviews the diff for naming consistency, public-surface design, parameter clarity, and symmetric-pair conventions. Compares new names against the codebase's existing vocabulary rather than abstract style guides.
Agent
Security lens for review-branch. Reviews the diff for trust-boundary issues, injection vectors, authn/authz regressions, secret handling, and resource-exhaustion risks. Reads full source files end-to-end and attempts to reproduce suspected bugs by running tests in the worktree.