Code Quality FAQ

What is code quality review?

A one-call code review: the CodeReviewWorkflow delegates to four specialized SDK subagents — security, quality, performance, architecture — and synthesizes their findings into a single report with an overall 0–100 health score, per-domain findings, and a prioritized list of next steps.

When should I run a code quality review?

Before opening pull requests, after large refactors, when inheriting unfamiliar code, or any time you want a health read on a codebase. It is the everyday breadth review.

How do I start a review?

Four ways: the /code-quality skill in a Claude Code conversation, the CLI (attune workflow run code-review --path src/ — note the slug), the code_review MCP tool, or the Python API (await CodeReviewWorkflow().execute(path="src/")).

What's the difference between quick, standard, and deep reviews?

depth sets the agent-turn budget — quick 10, standard 20 (the default), deep 40. All four passes run at every depth; a bigger budget means more thorough passes, not different ones. (One nuance: the /code-quality skill routes a "deep" request to the separate deep-review workflow.)

What do the health scores mean?

The Summary section opens with an overall 0–100 health score synthesized from the four passes — higher is healthier. It is an LLM judgment, not a measurement: treat it as a trend signal and read the per-domain findings for the substance.

Can I fix issues automatically?

The workflow reports; it never modifies files. In a Claude Code session, ask the agent to apply specific fixes from the report — mechanical ones (unused imports, style) are quick wins; structural findings need manual judgment.

How do I focus on specific types of issues?

There's no focus parameter — all four passes always run. Narrow the path, use deep-review (which has focus), or drill into the returned report conversationally ("just show me the security findings").

What if I want to compare different parts of my code?

Run the workflow once per path (e.g. src/auth/ then src/api/) and compare the health scores and finding counts — the report is per-run, one path at a time.

Where can I learn more?

Say "tell me more" (the coach skill goes progressively deeper: concept → procedural → reference), or open this feature's quickstart and task guides.

Why does attune workflow run code-quality say "unknown workflow"?

The registered slug is code-review. The code-quality name is the skill / help topic; run the workflow as attune workflow run code-review.

What's the difference between code-quality and deep-review?

Code-quality runs four passes — security, quality, performance, architecture — and has no narrowing knob. Deep-review swaps performance and architecture for a test-gap pass and lets you narrow with focus. Use code-quality for the everyday maintainability read, deep-review when test coverage is the concern.

How do I run only the security pass?

You can't — code-quality has no focus. Either narrow the path, or use deep-review with focus=["security"].

Which calls are async?

execute is the only public method and it is a coroutine — await it or use asyncio.run.

Does a clean review mean the code is good?

No. Findings are LLM predictions, not proofs, and a clean pass is not a guarantee — treat the review as one input.