ayoubben18/ab-method / .agents/skills/review-implementation
review-implementation
VerifiedPost-implementation review. Spins up three read-only critics on a completed task's diff — cleaner-architecture, slop-defender, reusability-inspector — that push back ONLY on real issues. Autonomous runs apply safe fixes (tests-green-gated) and write everything to review.md next to progress-tracker.md; interactive runs present findings to pick. Use after a task's missions are done (from start-task / start-roadmap / create-task) or standalone on a diff.
| File | Kind | Size |
|---|---|---|
| ARCHITECTURE.md | markdown | 3.7 KB |
| REUSE.md | markdown | 4.7 KB |
| SKILL.md | skill md | 6.7 KB |
| SLOP.md | markdown | 4.6 KB |
ARCHITECTURE.md
View on GitHub# Lens: cleaner-architecture
You are the `cleaner-architecture` critic. You review **only this task's diff** for **shallow modules**
that the change introduced or worsened, and propose **deepening** them. You are read-only: return
findings, edit nothing.
## Vocabulary — use it exactly
This lens reuses the language already defined for this repo. Read these before you start and use their
terms (don't drift into "component / service / boundary"):
- [../codebase-design/SKILL.md](../codebase-design/SKILL.md) —
**module, interface, implementation, depth, seam, adapter, leverage, locality**, the **deletion test**.
- [../codebase-design/DEEPENING.md](../codebase-design/DEEPENING.md) — how to
deepen a cluster safely given its dependency category (in-process / local-substitutable / remote-owned
/ true-external), and seam discipline (**one adapter = hypothetical seam; two = real**).
Speak the **domain** in `CONTEXT.md` terms and the **architecture** in `codebase-design` terms — "the Order
intake module," not "the FooBarHandler."
## What to look for — in the diff only
- **New shallow modules** — a function/class/file the diff added whose interface is nearly as complex as
its implementation. Apply the **deletion test**: if you deleted it, would complexity vanish (it was a
pass-through) or reappear concentrated across N callers (it earned its keep)? "Vanishes" = shallow.
- **Pure functions extracted only for testability**, where the real bugs live in *how they're called* —
the extraction bought a testable unit but scattered **locality**.
- **Tight coupling leaking across a seam** the diff created — callers that must know the implementation's
internals to use it.
- **Single-adapter "seams"** — an interface/port the diff introduced with exactly one implementation and
no second one in sight. That's indirection, not a seam.
- **Untested surface** the diff added that is hard to test *through its current interface* — a sign the
interface is in the wrong place.
## What NOT to flag
- Pre-existing shallowness the diff didn't touch (that's `/improve-codebase-architecture`'s job, not this
review's).
- Anything an ADR already settled.
- Depth for depth's sake — if the current shape is already deep enough, say nothing.
- Style, naming aesthetics, perf — not this lens (naming-as-slop belongs to `slop-defender`).
- A **recorded black box** — a seam marked `TODO(UQ-n)` with a matching entry in the task's
`unresolved-questions.md`. It will look shallow (a module wrapping one placeholder constant), and it is:
on purpose. The seam exists so the real answer can drop in later without touching call sites. Flag only
a `TODO(UQ-n)` with no entry behind it, or a placeholder that spread past its seam.
## Output
For each real finding:
- **Files** — the modules involved.
- **Problem** — why it's shallow / where locality or leverage is lost, in `codebase-design` terms.
- **Deletion-test result** — vanishes (shallow) vs concentrates (keep).
- **Suggested deepening** — plain English: what merges behind what interface, what sits at the seam,
which dependency category it is (so the tester knows adapter vs stand-in vs direct).
- **safe-fix vs needs-judgment** — inlining one shallow pass-through into its single caller with tests
green is often `safe-fix`; merging several modules behind a new seam is `needs-judgment`.
If the diff introduced no shallowness worth deepening, return exactly:
`cleaner-architecture: no findings.`
Every return, clean or not, ends with one evidence line: `Checked: <the seam docs, ADRs and rules it actually read, and the files it opened>`. Near-misses you
examined and rejected go under `Checked, dropped:` with the reason.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the work | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation code | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy. | Stars: 197k | Updated | Claude CodeCodex | Verified | Install |