arch-reviewer

Composite: reviews an existing architecture (a code path, or a description) against Löwy's Method - layer placement, naming, Design Don'ts, use case call chains - and writes a scorecard. Delegates the

SKILL.md

GitHub

Part of the lowy plugin, which depends on the dmf plugin: invoke skills by their namespaced name (lowy:list-volatilities for this plugin, dmf:domain-modeling for DMF).

Review the following architecture against Löwy's Method: $ARGUMENTS

This is the interactive review: one lens (Löwy), with the user confirming the use cases. For a multi-lens audit with adversarial verification (Löwy, call chains, DMF, functional style), run /lowy:arch-audit instead.

Findings follow ${CLAUDE_PLUGIN_ROOT}/references/review-findings.md. When the Agent tool is not available (for example inside a subagent), do each delegated phase yourself with the skill named in its fallback line.

Phase 0: Scope

If cc-docs/architecture.md exists, read it and apply it as described in ${CLAUDE_PLUGIN_ROOT}/references/project-profile.md.

  • A path (or nothing, inside a repository): review the code. Docs are claims to verify against it.
  • A description: review it as given, and state that the findings are unverified against code.

Phase 1: Structural Review

Delegate to the lowy:lowy-reviewer agent with the target, and with the description when there is no code. It builds the service inventory from the code and checks layers, naming, the 12 Design Don'ts, closed architecture, the Managers-to-Engines ratio, expendability, Utilities and decomposition smells.

Fallback: the lowy:classify-structure skill on an inventory you build from the code.

Show the user the inventory and the findings table before going on. If the inventory misplaces a service, correct it with the user and send the correction back to the agent.

Phase 2: Use Case Validation

Agree on 2-6 core use cases with the user. When reviewing code, propose candidates from the Managers' public operations, and confirm them before tracing.

Delegate to the lowy:call-chain-validator agent with the target and the confirmed use cases. It returns one chain per use case, with every hop checked, a verdict, the symmetry comparison, and any use case the services cannot satisfy.

Fallback: the lowy:validate-use-cases skill.

Phase 3: Summary Report

Merge the two results. Drop duplicates (the same rule on the same lines), keep the stronger evidence, and keep every finding's status: ❌ new violation, ⚠️ known debt, or ✅ approved deviation, per the project profile.

Architecture Scorecard

  • Layer compliance: ✅/⚠️/❌
  • Naming compliance: ✅/⚠️/❌
  • Design Don'ts: X violations found
  • Use case coverage: X/Y core use cases validated
  • Symmetry: ✅/⚠️/❌
  • Functional decomposition risk: Low/Medium/High

Recommended Actions

Prioritized list of changes, from most critical to least.

Profile Gaps and Not Checked

Copy both sections from the agents' results: proposals for the profile (never applied without the user), and what could not be verified.

Save the report as cc-docs/reviews/<target>-lowy-review-<YYYY-MM-DD>.md, where <target> is the kebab-case name of the reviewed subsystem or system.

If the review found mechanical violations (layer direction, Design Don'ts #2 and #6-#12, naming), offer /lowy:fitness-tests so they cannot come back unnoticed.