second-opinion

Request a Claude peer review of diffs, commits, or design plans from Codex.

4|Updated Jun 9, 2026
One-click install
npx skills add https://github.com/laicluse/agent-fieldkit --skill second-opinion-laicluse
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: second-opinion
Source: https://github.com/laicluse/agent-fieldkit/tree/main/.agents/plugins/generated/intervision/skills/second-opinion
Command: npx skills add https://github.com/laicluse/agent-fieldkit --skill second-opinion-laicluse

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When working alone, a coding agent can miss bugs, risks, and design flaws in its own work. This Skill brings Claude in as an independent peer reviewer via the claude -p CLI, giving you a genuinely different read on recent changes or plans before you commit to them. ## Core Features & Use Cases - Diff and Commit Review: Pipes the actual git status, staged/unstaged diffs, branch ranges, or single commits to Claude for peer review, with findings reported first and no file edits allowed. - Design Consultation: Sends a design plan or open trade-off to Claude before any code is written, keeping the uncertain parts intact for an honest critique. - Multi-Round Dialogue: Supports follow-up exchanges that quote the peer's prior findings alongside your evidence, resolving or sharpening each disagreement. - Use Case: After finishing a feature branch, pipe git diff main...HEAD to Claude, receive a list of bugs and missing tests, then push back on one flagged race condition with evidence from your locking code. ## Quick Start Ask the agent to get a second opinion from Claude on the current uncommitted changes in this repository.

Frequently Asked Questions about second-opinion

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I get a second opinion on my code changes from another AI?▼

Pipe your actual git status and diff into `claude -p` with tools disabled, and ask it to report bugs, risks, and missing tests first. The Skill provides ready-made commands for uncommitted changes, branch ranges, and single commits.

How to review a design plan with Claude before writing code?▼

Paste the full design, including the uncertain trade-offs, into a quoted heredoc passed to `claude -p` on stdin. Keeping the shaky parts in ensures the peer review is honest rather than flattering.

Does this skill work without the Claude CLI installed?▼

No. The claude CLI must be installed and logged in, since the peer review runs through `claude -p`. The Skill checks for the binary first and stops with a clear message if it is missing.

Why does the Claude review take so long with no output?▼

Claude is materially slower on substantial reviews and may produce no partial output for many minutes while still working. Allow at least 20 minutes and keep waiting while the process is alive; silence alone is not failure evidence.

Can the peer reviewer edit my files during the review?▼

No. The handoff runs with `--tools ""`, which removes file and shell tools from the Claude process, and the prompt explicitly instructs it not to edit files. The peer proposes findings; you decide what to fix.