← ClaudeAtlas

project-review-codebaselisted

Read-only review of production and test code across three dimensions — consistency, structure, architecture — via a multi-agent workflow that dedupes findings across dimensions and writes a standalone Markdown report with Mermaid diagrams; reports fixes, never edits.
hk9890/claude-dev-tools · ★ 0 · Code & Development · score 76
Install: claude install-skill hk9890/claude-dev-tools
Read-only codebase review across three dimensions — consistency, structure, and architecture. Launch the review workflow — do **not** review inline. The workflow returns a structured report plus a Markdown artifact; relay the report and write the artifact to a temp file. Test code is a subject here, not a backdrop: the dimensions cover test conventions and setup idioms, test-support layout and misplaced tests, and whether each module's failure modes are actually exercised. ## Run the workflow 1. Parse `$ARGUMENTS` as `[low|medium|high|ultra] [what-to-review]`. Both are optional. A leading `low` | `medium` | `high` | `ultra` token is the **level** (default `medium`); everything after it is **what to review** — a free-form description ("naming across the service layer") or a path. Default: the whole codebase. 2. `SKILL_DIR` is the **base directory for this skill**, given at the top of this file when the skill loads. It is absolute and install-correct — build every path below from it. 3. Invoke the **Workflow** tool: - `scriptPath`: `<SKILL_DIR>/workflows/review-codebase.js` - `args`: `{ "repoRoot": "<repo root, or the step-1 path>", "scope": "<the step-1 what-to-review, or empty>", "vocabFile": "<SKILL_DIR>/references/design-vocabulary.md", "level": "<the step-1 level>" }` - The workflow fans out one adversarial read-only agent per dimension (consistency, structure, architecture), then a synthesis stage dedupes and reconciles findings acro