---
name: myai-dev-review
description: Review code for correctness bugs, security issues, and convention violations
---

Review a change before it merges.

Invoke the `myaidev-reviewer` skill for the full pipeline, including optional auto-fix. For
a read-only review, invoke the `dev-reviewer` agent.

## Usage

```bash
/myai-dev-review                          # reviews the working diff
/myai-dev-review --scope=branch           # reviews the whole branch vs base
/myai-dev-review --fix                    # applies the fixes it finds
/myai-dev-review --focus=security
```

## Options

| Option | Description | Default |
|--------|-------------|---------|
| `--scope` | `diff`, `branch`, `file`, `module` | `diff` |
| `--fix` | Apply fixes rather than only reporting | `false` |
| `--focus` | `security`, `performance`, `correctness`, `conventions` | all |

## What It Does

1. Establishes the codebase's conventions before judging anything as a violation
2. Reads the spec — a correct implementation of the wrong thing is the costliest miss
3. Looks for correctness bugs first, then security, then contract breaks, then conventions
4. Tries to disprove each finding before reporting it
5. Reports by severity, each with a concrete failure scenario and a specific fix

## Output

`{session_dir}/review.md`, with findings graded CRITICAL / WARNING / SUGGESTION / INFO.

Every finding names a `file:line`, the input or state that triggers the bad outcome, and
what to change. A finding without a reproducible scenario is downgraded, not reported as a
bug — false positives cost more than missed nits.