AgentSkillsCN

ask

通过 Gemini CLI 或 OrcaX 获取其他模型的第二意见。

SKILL.md
--- frontmatter
name: ask
description: Get second opinion from other models via gemini CLI or orcx.
allowed-tools: Read, Bash

Second Opinions from Other Models

Consult other models for second opinions.

Gemini (Preferred for Google Models)

Use the native gemini CLI for Gemini models (authenticated with subscription). Output goes to stdout; save to file to capture results.

bash
# One-shot query (output to stdout)
gemini "Your prompt here"

# Save response to file
gemini "Your prompt here" > response.md 2>/dev/null

# Pipe file context
cat file.py | gemini "Review this code" > review.md 2>/dev/null

# JSON output with stats
gemini -o json "Your prompt here" 2>/dev/null | jq -r .response

# Continue interactively after prompt
gemini -i "Start with this question"

orcx (Other Models)

Run orcx agents to see configured agents (DeepSeek, Grok, etc.).

Usage

bash
# Simple query (uses default model)
orcx "Your prompt here"

# With specific agent
orcx -a fast "Your prompt here"

# Pipe context
cat file.py | orcx "Review this code"

# Include file context
orcx -f file.py "Review this code"

# Save response to file
orcx -o response.md "Explain this concept"

# Explicit run subcommand (equivalent)
orcx run -a fast "Your prompt here"

Conversations

orcx stores conversations in SQLite (~/.config/orcx/conversations.db).

bash
# Continue last conversation
orcx -c "Follow-up question"

# Resume specific conversation by ID
orcx --resume abc1 "More context"

# Don't save this exchange
orcx --no-save "Quick question"

# List recent conversations
orcx conversations list

# Show conversation history
orcx conversations show abc1

# Clean old conversations (30+ days)
orcx conversations clean

When to Use

ToolUse for
geminiGoogle models (authenticated with subscription)
orcxDeepSeek, Grok, other OpenRouter models
  • •Second opinion on logic or approach
  • •Comparing model perspectives
  • •Leveraging model-specific strengths
  • •Fresh eyes on a problem

Workflow

  1. •Identify the specific question or code to review
  2. •Formulate clear, focused prompt
  3. •Include relevant context (pipe or -f for orcx, pipe or stdin for gemini)
  4. •Run via gemini or orcx
  5. •Save output to file if needed for reference
  6. •Compare response with your analysis
  7. •Synthesize insights