multi-llm-consult
Consult external LLMs (Gemini, OpenAI/Codex, Qwen) for second opinions, alternative plans, independent reviews, or delegated tasks. Use when a user asks for another model's perspective, wants to compare answers, or requests delegating a subtask to Gemini/Codex/Qwen.
When & Why to Use This Skill
The Multi-LLM Consult skill empowers Claude agents to cross-reference tasks and plans with external large language models such as Gemini, OpenAI (Codex), and Qwen. By facilitating second opinions, independent reviews, and task delegation, this skill enhances decision-making accuracy, reduces hallucinations, and provides a robust framework for multi-model collaboration within a single workflow.
Use Cases
- Case 1: Risk Assessment - Requesting a second opinion on complex refactoring plans from Gemini to identify potential architectural gaps or edge cases.
- Case 2: Independent Code Review - Delegating a logic review to Qwen or Codex to ensure code quality and security through a multi-perspective validation process.
- Case 3: Plan Verification - Comparing implementation strategies for new features across different models to determine the most efficient and scalable approach.
- Case 4: Specialized Delegation - Offloading specific subtasks to external models that may have specialized training or higher context windows for particular domains.
| name | multi-llm-consult |
|---|---|
| description | Consult external LLMs (Gemini, OpenAI/Codex, Qwen) for second opinions, alternative plans, independent reviews, or delegated tasks. Use when a user asks for another model's perspective, wants to compare answers, or requests delegating a subtask to Gemini/Codex/Qwen. |
Multi-LLM Consult
Overview
Use a bundled script to query external LLM providers with a sanitized prompt and return a concise comparison.
Setup
- Configure API keys in the TUI: open the Command Palette (Ctrl+P) and run Configure LLM Providers.
- Keys are stored in
settings.jsonunderllm_providers.
Workflow
- Identify the purpose (
second-opinion,plan,review,delegate). - Summarize the task and sanitize sensitive data before sending it out.
- Run the consult script with the chosen provider.
- Compare responses and reconcile with your own plan before acting.
Consult Script
Always run --help first:
python scripts/consult_llm.py --help
Example: second opinion
python scripts/consult_llm.py \
--provider gemini \
--purpose second-opinion \
--prompt "We plan to refactor module X. What risks or gaps do you see?"
Example: delegate a review
python scripts/consult_llm.py \
--provider qwen \
--purpose review \
--prompt-file /tmp/review_request.md \
--context-file /tmp/patch.diff
Example: plan check with Codex (OpenAI)
python scripts/consult_llm.py \
--provider codex \
--purpose plan \
--prompt "Draft a 5-step plan for implementing feature Y."
Output Handling
- Treat responses as advisory; verify against repo constraints and current state.
- Summarize the external response in 3-6 bullets before acting.
- If responses conflict, call out the differences explicitly and choose a path.
References
- Provider defaults and configuration:
references/providers.md