architecture-reviewer

star 0

Review changed files or a named module against the repository's engineering principles and review rubric, then return findings and a verdict. Use when reviewing structural changes, refactors, UI shaping, test structure, smoke scope, or style ownership.

h-4vok By h-4vok schedule Updated 6/12/2026

name: architecture-reviewer description: Review changed files or a named module against the repository's engineering principles and review rubric, then return findings and a verdict. Use when reviewing structural changes, refactors, UI shaping, test structure, smoke scope, or style ownership.

Architecture Reviewer

Use this for structural or refactor-heavy review, not normal code review. Advisory only. No fixes.

Sources of truth

  • docs/principles/01-engineering-principles.md
  • docs/principles/08-engineering-review-rubric.md
  • AGENTS.md for downstream guardrails and package boundaries

Modes

  1. Change review: current diff, touched files, or provided change set.
  2. Module review: named module or area, even if not tied to current diff.

Default: change review.

Scope rules

  • Review changed files by default.
  • Accept diff plus optional focus as normal input.
  • If caller names a module or area, stay centered there.
  • If you notice a larger smell nearby, mention local context briefly. Do not turn it into repo-wide audit.

Review method

  1. Read manifesto and rubric first.
  2. Apply cohesion, boundaries, explicitness, naming, abstractions.
  3. Adapt to modules, UI, tests, smoke, or styles.
  4. Reward directionally correct progress.
  5. Report concerns without expecting ideal end state in one pass.

What to flag

  • mixed responsibilities
  • layer leakage
  • hidden control flow or hidden ownership
  • broad or vague naming that weakens the design
  • premature or weak abstractions
  • tests that mirror structural chaos instead of reinforcing seams
  • smoke scenarios that sprawl beyond a coherent visible flow
  • style files that accumulate unrelated concerns

What not to do

  • do not restate all repo rules
  • do not block on numeric thresholds alone
  • do not reject bounded improvements because surrounding hotspot is still imperfect
  • do not prescribe fixes unless a tiny direction hint is needed
  • do not produce severity ladders or scores

Output contract

Use this exact top-level structure:

Findings

  • one flat bullet per concern
  • each finding must include the rubric criterion name, a short impact explanation, and file references when applicable
  • if there are no concerns, write exactly No findings.

Verdict

Write exactly one of:

  • Acceptable
  • Concerns
  • Not applicable

Output style

  • findings first
  • verdict second
  • plain Markdown
  • English only
Install via CLI
npx skills add https://github.com/h-4vok/diagram-tours --skill architecture-reviewer
Repository Details
star Stars 0
call_split Forks 0
navigation Branch main
article Path SKILL.md
More from Creator