agentsclimarketplace

Architecture review

Skill lx-wnk/skills/skills/architecture-review

Portable Agent Skills for AI coding agents (Claude Code, Codex, Cursor, Gemini) — agentskills.io-conformant, versioned, installable via skills.sh

Install
npx -y skills add lx-wnk/skills --skill architecture-review

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Review system-level architecture of a PR, branch, namespace, or whole project — boundary violations, dependency cycles, layer breaches, ADR compliance, cohesion of modules and bounded contexts. Make sure to use this skill whenever the user asks to review the architecture, audit module boundaries, check for circular dependencies, verify domain structure, or says things like "architektur-review", "prüfe die struktur", "check architecture", "review namespace", "audit boundaries", "are our modules clean". Use this skill for HIGH-LEVEL structural review only — SOLID, class design, and method-level issues belong in `component-review`.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

5.9 KB, ~1.2k tokens by cl100k_base, as published. Nobody here has run it

Architecture Review (System Level)

Audit structural health: boundaries, dependency direction, layering, ADR drift. Report findings — never modify code.

Scope

In scope: module/context boundaries, dependency cycles, layer violations, namespace cohesion, public-API surface leaks, shared-model pollution, drift from documented ADRs.

Out of scope: SOLID inside a class, naming of private methods, style, line-level bugs (→ component-review or general code review).

Examples

# Review architecture of a specific PR
/architecture-review pr 42

# Review a branch against main
/architecture-review branch feat/orders-refactor

# Focus on a single namespace/module
/architecture-review namespace src/Order

# Review the whole project
/architecture-review

Workflow

flowchart TD
  A["Parse target mode"] --> B["Load project context"]
  B --> C["Collect files in scope"]
  C --> D["Build dependency map"]
  D --> E["Run checks"]
  E --> F["Report findings"]

Phase 1: Parse Target

From $ARGUMENTS:

  • pr <N>gh pr diff <N>, gh pr view <N> --json headRefName,files, read full file context via git show <branch>:<file>
  • branch <name>git diff main...<name>, git ls-tree -r <name>, read files via git show <branch>:<file>
  • namespace <path> → recursively read files under <path>
  • empty → full-project review (sample entry points + top-level modules)

Phase 2: Load Project Context

Check in order:

  1. .agent-context/layer1-bootstrap.md, layer2-project-core.md
  2. .agent-context/decisions.json → ADRs that define intended structure
  3. docs/architecture/**/*.md
  4. CLAUDE.md, AGENTS.md, CONTRIBUTING.md
  5. Manifest files for stack detection

Never block on missing context — infer from code layout.

Phase 3: Build Dependency Map

Per detected stack, extract imports/uses:

  • PHP/Symfony/Shopwareuse statements, PSR-4 namespaces from composer.json
  • TypeScript/JavaScript/Vue/Nuxtimport statements, path aliases from tsconfig.json/nuxt.config
  • Goimport blocks, module path from go.mod
  • Java/Kotlinimport + package declarations
  • Pythonimport/from, top-level packages

Group imports by top-level module/namespace. Flag:

  • Imports that cross declared module boundaries
  • Imports going "upward" in a layered architecture
  • Imports into another module's internal/private sub-path

Phase 4: Architectural Checks

Run each relevant check and record findings with file:line:

1. Dependency Direction

  • Layered: higher layers may import lower, never the reverse
  • Hexagonal: adapters → ports → domain (never domain → adapters)
  • Modular: module A imports only B's public API, never internal/

2. Circular Dependencies

  • Build a module-level graph, detect cycles
  • Even acceptable cycles should be listed as risks

3. Boundary Cohesion

  • Does each module have a single clear responsibility?
  • Public API surface: how many files/classes does an outside caller touch?
  • "God module" symptom: one module imported by almost everything

4. Shared Model Pollution

  • Domain entities reused across bounded contexts without translation
  • Shared "Common" / "Util" module growing unbounded

5. ADR Drift

  • For each ADR in decisions.json or docs/architecture/adr/: does the current code still match?
  • Flag the specific ADR ID and the divergent file

6. Framework Convention Drift

  • Use WebFetch / Context7 for current-version best practices when the stack version matters
  • Example: Nuxt 3 server/client boundary, Symfony bundle structure, Go internal/

7. Cross-Cutting Concerns

  • Is auth enforced at a single seam or scattered?
  • Is logging/observability concentrated or repeated ad-hoc?

Phase 5: Report

Output in the user's language. Structure:

## Architecture Review — <scope>

**Risk Level:** LOW | MEDIUM | HIGH | CRITICAL

**Detected stack:** <stack>

**Checks run:** <list>

## Critical (must fix)

- **<Title>** — `path/to/file:line`
  - Problem: <what>
  - Why it matters: <impact>
  - Suggested direction: <high-level fix, not code>

## Warnings (should fix)

- ...

## Observations (worth knowing)

- ...

## ADR Compliance

| ADR                   | Status      | Notes                              |
| --------------------- | ----------- | ---------------------------------- |
| ADR-0003 Hexagonal    | ✅ ok       |                                    |
| ADR-0007 No shared DB | ⚠️ drifting | `OrderRepo` touches billing tables |

## Suggested Next Steps

1. ...
2. ...

Rules

  • Read-only. Never modify source files. Findings are reports.
  • Structural focus. If the only findings are code-quality issues, recommend component-review instead.
  • Evidence-based. Every finding needs a concrete file:line reference.
  • Prioritize. Not every boundary leak is critical — severity matters more than count.
  • Fallback, don't block. Missing ADRs or context layers → infer intent from code layout.
  • Ignore style. Linters and formatters own that territory.

Gives 0 of the 12 instructions most quality gates skills give in ~1.2k tokens

Counted across 1,195 of the 2,094 authors here whose files we hold, read 2026-08-06

  • read the output and check the exit codein 55 of 1195, across 14 files
  • verify requirements using a line-by-line checklistin 53 of 1195, across 12 files
  • identify the verification command proving the claimin 53 of 1195, across 12 files
  • run the full verification commandin 51 of 1195, across 11 files
  • verify output confirms the claimin 49 of 1195, across 10 files
  • check version control diff after agent delegationin 45 of 1195, across 5 files
  • state claim with evidencein 43 of 1195, across 3 files
  • run the test suitein 32 of 1195, across 24 files
  • keep state in memory by defaultin 27 of 1195, across 6 files
  • make prototype runnable with one commandin 26 of 1195, across 5 files
  • detect the package manager from lockfilesin 24 of 1195, across 5 files
  • produce a verification reportin 23 of 1195, across 12 files

Said here and by no other author read

  • never modify code
  • keep scope to structural issues
  • build a dependency map
  • flag imports crossing module boundaries
  • detect dependency cycles
  • check for ADR drift

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.