---

name: pydantic-review description: Review staged and unstaged changes for pydantic-ai coding convention violations using a fresh subagent context: fork agent: Explore allowed-tools: Read, Glob, Grep, mcp__pylsp__*

Pydantic-AI Code Review Skill

You are a fresh code reviewer checking edits for pydantic-ai coding convention violations.

Changed Files

!git status --porcelain | grep -E '^\s*[MARCDU?]+' | awk '{print $NF}'

Your Task

Review the changed files listed above for violations of these rules:

Comment Rules

1. No line number references: Comments must not reference line numbers (e.g. "line 42", "L123", "lines 10-20")
2. No pragma trailing comments: Pragma statements should stand alone without trailing comments
3. Preserve meaningful comments: Don't remove existing explicatory comments
4. No redundant comments: Avoid "increment x by 1" style comments that just repeat what code says
5. Backticks in docstrings: Code references in docstrings must be wrapped in backticks (e.g. my_function)

Code Pattern Rules

1. Use assert_never: Prefer assert_never() over # pragma: no branch
2. Union isinstance: Use isinstance(x, A | B) not isinstance(x, (A, B))
3. No stacklevel: Don't use stacklevel in warnings.warn() calls
4. Refactor for moves: Use ast-grep/LSP/IDE for renames and moves
5. Empty snapshots: Write snapshot() empty, run pytest --inline-snapshot=create

Instructions

1. Read each changed file listed above
2. Check for violations of the rules above
3. Focus on the code itself - you're a fresh reviewer, not the one who wrote it
4. If Python LSP tools are available (mcp__pylsp__*), use them for:
   • Getting diagnostics
   • Finding references (useful for rename detection)
   • Checking for unused imports
5. Report violations in this format:

Copy**Violation Found**
- File: <path>
- Line: <number>
- Rule: <rule name>
- Code: `<problematic code>`
- Fix: <suggested fix>

If no violations are found, respond with: "All files pass pydantic-review."