| name | eng-review |
| description | Third pass of the eng pipeline. An engineering-manager review of a .plan doc: rates the plan across architecture, edge cases, test coverage, performance, and — critically — consistency with how THIS codebase already builds things (APIs, components, patterns), so the eventual code matches house style. Scores each dimension 0-10, explains what would make each a 10, and folds the fixes back into the doc. Use when a grilled .plan doc is ready for architectural review before defining tests. Triggers: "eng review", "review the plan", "review the architecture", "lock in the plan". |
eng-review — pressure-test the plan and align it to house conventions
You are the third pass of the pipeline. The doc has been scoped and grilled.
Your job is to catch architecture and consistency problems before tests and
code, and to leave the plan provably better.
First, read the doc schema at ${CLAUDE_PLUGIN_ROOT}/SPEC.md, the target
ticket doc, and the ## Codebase Touchpoints / ## Conventions & Patterns it
references. If multiple tickets, review each (or ask which).
Review dimensions — rate each 0-10
For every dimension: give a score, state the one or two things that would make
it a 10, then apply those fixes to the doc (don't just critique).
- Architecture & approach — is the design sound, simple, and the right
shape for the problem? Any over- or under-engineering?
- Consistency with the codebase — THE dimension you must not skip. Does the
plan match how this repo already does these things? Re-read the real code:
- How are APIs/endpoints/services defined here? Does the plan follow it?
- How are React components structured (props, state, styling, file layout)?
- Naming, error handling, logging, config, folder conventions, test style.
Cite
file:line for the pattern the plan should follow. The goal: code that
looks like the team wrote it, not generic boilerplate.
- Edge cases & failure modes — empty/duplicate/malformed inputs, partial
failure, retries, idempotency, race conditions.
- Test coverage — will the planned tests actually prove the acceptance
criteria? Gaps?
- Performance & scale — N+1s, unbounded loops/queries, large payloads,
anything that won't hold at real data sizes.
- Data & migration safety — schema changes, backfills, rollback,
backward compatibility.
Steps
- Walk the dimensions interactively with the user — surface the real issues,
give an opinionated recommendation for each, let them weigh in on judgment
calls. Don't just dump a report.
- Fold accepted changes back into the relevant sections (Scope, Touchpoints,
Conventions & Patterns, Acceptance Criteria, Test Plan).
- Write the
## Eng Review Verdict section: dimension scores, what a 10 needs,
and what you changed.
- Set frontmatter
phase: reviewed (update the README table). End with:
"Run /tdd next to define the test contract."
