スキル一覧に戻る
vrazdalovschi

consult-agents

by vrazdalovschi

0🍴 0📅 2026年1月22日
GitHubで見るManusで実行

SKILL.md


name: consult-agents description: Use when you need validation, analysis, or research from multiple AI agents before implementing. Gathers perspectives from Gemini and Codex, synthesizes feedback.

Consult Agents

Get validation, analysis, or research feedback from Gemini and Codex before Claude Code implements.

Core principle: Claude Code implements. External agents validate and research.

When to Use

  • Validate an approach before implementing
  • Research a topic from multiple perspectives
  • Get code review feedback on a design
  • Confirm understanding of requirements
  • Aggregate opinions on trade-offs

Quick Reference

Task TypeGemini ModeCodex Mode
Analyze code/files onlygemini -p "..."--sandbox read-only
Research with web accessgemini -p "..." (has web search)--sandbox danger-full-access
Deep repo analysistmux interactive--sandbox read-only

Workflow

1. Prepare question/context (what you need validated)
2. Query Gemini → capture response
3. Query Codex → capture response
4. Synthesize: agreements, disagreements, insights
5. Claude Code decides and implements

Parallel Execution

Run both agents simultaneously for faster feedback:

# Terminal 1: Gemini
gemini -p "Analyze this approach: [context]. What are the trade-offs?" 2>/dev/null

# Terminal 2: Codex
echo "Analyze this approach: [context]. What are the trade-offs?" | \
  codex exec --sandbox read-only --skip-git-repo-check -m gpt-5.2 2>/dev/null

Or use background jobs:

gemini -p "..." > /tmp/gemini-response.txt 2>/dev/null &
echo "..." | codex exec --sandbox read-only ... > /tmp/codex-response.txt 2>/dev/null &
wait

Synthesis Template

After gathering responses, synthesize:

## Agent Responses

### Gemini
[Summary of Gemini's perspective]

### Codex
[Summary of Codex's perspective]

## Synthesis

**Agreements:**
- [What both agents agree on]

**Disagreements:**
- [Where they differ and why]

**Key Insights:**
- [Unique valuable points from either]

## Decision
[Claude Code's decision based on aggregated feedback]

Prompt Templates

Validation

Review this [design/approach/code]:

[CONTEXT]

Questions:
1. Is this approach sound?
2. What are the risks?
3. What would you do differently?

Research

Research [TOPIC] focusing on:
1. Current best practices
2. Common pitfalls
3. Recommended tools/libraries

Context: [Why you need this]

Code Review

Review this code for:
- Correctness
- Performance concerns
- Security issues
- Maintainability

[CODE]

Common Mistakes

MistakeFix
Asking agents to implementThey validate; Claude Code implements
Not providing enough contextInclude relevant code, constraints, goals
Ignoring disagreementsDisagreements reveal important trade-offs
Blindly following consensusUse judgment; agents can both be wrong

When NOT to Use

  • Simple, obvious implementations
  • Time-critical changes (adds latency)
  • Trivial questions (overkill)

スコア

総合スコア

50/100

リポジトリの品質指標に基づく評価

SKILL.md

SKILL.mdファイルが含まれている

+20
LICENSE

ライセンスが設定されている

0/10
説明文

100文字以上の説明がある

0/10
人気

GitHub Stars 100以上

0/15
最近の活動

3ヶ月以内に更新がある

0/10
フォーク

10回以上フォークされている

0/5
Issue管理

オープンIssueが50未満

+5
言語

プログラミング言語が設定されている

+5
タグ

1つ以上のタグが設定されている

0/5

レビュー

💬

レビュー機能は近日公開予定です