review

name	review
description	Review pending changes with external LLMs (Gemini, Codex, Claude, or all).
allowed-tools	Bash, Glob, Grep, Read, Task, mcp__consult-llm__consult_llm
disable-model-invocation	true

Review pending changes with external LLMs.

Arguments: $ARGUMENTS

Flags

Check arguments for reviewer flags (mutually exclusive):

• --gemini → review with Gemini only
• --codex → review with Codex only
• --claude → review with a Claude subagent
• No flag → review with both Gemini and Codex in parallel (default)

Strip flags from arguments. Remaining text is optional scope/focus for the review.

Phase 1: Gather Changes

1. Run git diff to get unstaged changes and git diff --cached for staged changes
2. If there are no changes, check git log main..HEAD for committed changes on the branch
3. If the user provided scope text (e.g. "auth changes"), use it to focus on relevant files
4. Read key files that are changed to understand context

Phase 2: Send for Review

Based on the flag, consult the reviewer(s):

If --gemini: Gemini only

Call mcp__consult-llm__consult_llm with:

• model: "gemini"
• prompt: Review prompt below
• task_mode: "review"
• git_diff: {"files": [<changed files>], "base_ref": "HEAD"}
• files: Array of relevant source files for context

If --codex: Codex only

Call mcp__consult-llm__consult_llm with:

• model: "openai"
• prompt: Review prompt below
• task_mode: "review"
• git_diff: {"files": [<changed files>], "base_ref": "HEAD"}
• files: Array of relevant source files for context

If --claude: Claude subagent

Use the Task tool with subagent_type: "general-purpose" and a prompt like:

Review the pending changes in this repository.

Run `git diff` and `git diff --cached` to see the changes, then read relevant files for context.

Consider:
- Any obvious bugs or edge cases missed?
- Code quality issues (error handling, naming, structure)?
- Deviations from best practices or existing patterns?
- Security concerns?

Provide specific, actionable feedback. Be concise.

If no flag (default): Both Gemini and Codex in parallel

Spawn BOTH as parallel subagents (Agent tool, subagent_type: "general-purpose", model: "sonnet"). NEVER run subagents in the background — always run them in the foreground so you can process their results immediately. Each subagent prompt must include the full review prompt, git_diff details, and file list so it can make the MCP call independently.

Gemini subagent — prompt must include:

• Call mcp__consult-llm__consult_llm with model: "gemini", prompt: the review prompt, task_mode: "review", git_diff: {"files": [<changed files>], "base_ref": "HEAD"}, files: [array of relevant source files]
• Return the COMPLETE response

Codex subagent — prompt must include:

• Call mcp__consult-llm__consult_llm with model: "openai", prompt: the review prompt, task_mode: "review", git_diff: {"files": [<changed files>], "base_ref": "HEAD"}, files: [array of relevant source files]
• Return the COMPLETE response

---

Review prompt:

Review these changes. Consider:
- Any obvious bugs or edge cases missed?
- Code quality issues (error handling, naming, structure)?
- Deviations from best practices or existing patterns?
- Security concerns?
- Anything that could break existing functionality?

Provide specific, actionable feedback. Be concise. Only flag issues worth fixing.

If the user provided scope/focus text, append it to the review prompt.

Phase 3: Present Results

Present the feedback to the user:

• If both reviewers were used, organize by reviewer (Gemini / Codex)
• Highlight common concerns flagged by both
• Note any conflicting feedback
• Ask if the user wants to apply any suggested fixes