AgentSkillsCN

consulting-peer-llms

当用户明确请求其他大语言模型工具（Gemini、Codex、Claude）对当前工作提供反馈时使用。

SKILL.md
--- frontmatter
name: consulting-peer-llms
description: Use when user explicitly requests feedback from other LLM tools (Gemini, Codex, Claude) on current work
allowed-tools: "Bash(gemini:*) Bash(codex:*) Bash(claude:*) Bash(timeout:*) Bash(command:*) Bash(bash:*) Bash($SKILL_DIR:*)"

Consulting Peer LLMs

Overview

Get feedback from other LLM CLI tools (Gemini, Codex) on your current work. This skill executes multiple LLM reviews in parallel and synthesizes their responses into one actionable report.

Core principle: Use peer LLMs for external validation and diverse perspectives on implementation quality.

When to Use

Use this skill when the user explicitly requests:

  • •"Validate this with codex"
  • •"Get feedback from gemini"
  • •"I want a review from other LLMs"
  • •"Do a peer review"

Do NOT use:

  • •Automatically without user request
  • •For every piece of code (it's heavyweight)
  • •When quick internal review is sufficient

Prerequisites

Supported CLI tools:

  • •gemini - Google Gemini CLI
  • •codex - OpenAI Codex CLI
  • •claude - Anthropic Claude CLI

Detection:

  • •Auto-detect all installed CLIs via command -v
  • •Use all available CLIs for comprehensive review

Workflow

Step 1: Collect Context

Work Summary:

  • •What was implemented (from conversation)
  • •User's requirements and constraints

Step 2: Generate Review Prompt

Use structured prompt with these sections:

  1. •What was implemented
  2. •Requirements/plan
  3. •Instruction to compare current branch against origin/main (or origin/master)
  4. •User's specific review requirements (if any, passed as-is)

Full template: See reference/prompt-template.md

Step 3: Execute in Parallel

Run target CLIs simultaneously and collect results.

CLI commands: See reference/cli-commands.md

Step 4: Present Raw Responses

Show original responses first for transparency:

markdown
# Gemini Review
{response}
---
# Codex Review
{response}
---

Step 5: Synthesize Final Report

Always synthesize - even for single CLI responses.

Synthesis principles:

  1. •Consolidate duplicates — same issue from multiple CLIs = one entry
  2. •Filter for validity — skip suggestions irrelevant to current requirements
  3. •Prioritize by impact — not by which/how many CLIs mentioned it
  4. •Make actionable — concrete code fixes, not vague advice
  5. •Remove noise — focus on essentials

Report example: See reference/report-format.md

Quick Reference

Commands:

  • •/consulting-peer-llms:review <requirements> - Auto-detect all installed CLIs and review with the given requirements

Typical execution time: 5-30 minutes (parallel)

Temp files: /tmp/{cli-name}-review.txt (one per CLI)

Red Flags - STOP Immediately

  • •Running peer review without explicit user request
  • •Skipping raw response output
  • •Just showing raw responses without synthesis
  • •Skipping synthesis for single CLI
  • •Passing diffs, file lists, SHAs, or file contents in the prompt (CLIs can use git directly)
  • •Writing prompts to separate temp files instead of passing inline
  • •Specifying model parameters (e.g., --model) — use each CLI's default model

Error Handling

Some CLIs fail: Continue with successful ones, note failures in report

Timeout (exit 124): Reduce prompt size, check CLI responsiveness

No retries: Keep execution fast and simple

Troubleshooting

"Empty response from CLI"

  • •Check CLI can run: gemini -p "test"
  • •Verify API keys/auth
  • •Check prompt isn't too long

"Both CLIs failed"

  • •Run diagnostics: gemini --version && codex --version
  • •Check network connectivity

"Response is truncated"

  • •CLIs may have output limits
  • •Reduce prompt length

See Also