Can AI modify code or files during a consultation?

No. Consultations run in sandbox/read-only mode; no changes are made unless you explicitly authorize writes.

What output should I expect?

An executive summary, a prioritized findings section, your analysis, and a synthesis with concrete recommendations, formatted for reuse in reviews.

How do I choose a consultation type?

Use the Consultation Types guide to map your goal to Code Review, Architecture, Debugging, Security Audit, or Performance Review, and include relevant files and constraints in the prompt.

ai-consultation

npx machina-cli add skill jeffrigby/somepulp-agents/ai-consultation --openclaw

Files (1)

SKILL.md

3.3 KB

AI Consultation Skill

Methodology and templates for effective AI consultation workflows with external AI tools like Codex and Gemini.

When to Use

Getting second opinions from Codex or Gemini CLI
Code review consultations
Architecture decision validation
Security audits with multiple AI perspectives
Debugging with alternative diagnostic approaches
Performance reviews needing fresh insights

Consultation Types

Code Review / Quality Check - Focus on quality, security, performance, best practices
Architecture / Design Opinion - Focus on structure, patterns, scalability, tradeoffs
Debugging / Root Cause Analysis - Focus on finding causes, suggesting diagnostics
Security Audit - Focus on vulnerabilities, OWASP, auth, data exposure
Performance Review - Focus on bottlenecks, algorithms, optimization opportunities

Key Principles

Safety First

Always use sandbox/read-only mode for consultations
External AI tools should not modify files
Get user confirmation before any write operations

Critical Thinking

Never accept AI suggestions blindly
Compare external AI perspective with your own analysis
Validate recommendations before presenting
Acknowledge when uncertain

Effective Prompts

Be specific about files and components to analyze
Provide relevant context and constraints
Specify desired output format (line numbers, severity ratings)
Request prioritization by impact or severity

Resources

See the reference documents for detailed guidance:

references/prompt-templates.md - Ready-to-use prompt templates for all consultation types
references/examples.md - 12 concrete consultation examples with workflows
references/decision-tree.md - Comprehensive decision logic for choosing consultation types
references/consultation-checklist.md - Quality assurance checklist for consultations
references/cli-options.md - AI CLI options reference for Gemini and other tools

Quick Start

Basic Consultation Flow

Identify consultation type - What does the user need?
Prepare the prompt - Use templates, be specific
Execute safely - Always read-only/sandbox mode
Evaluate critically - Don't accept blindly
Synthesize perspectives - Combine AI insights with your own
Present clearly - Prioritize findings, provide action items

Example Prompt Structure

[ACTION VERB] [COMPONENT] for [PURPOSE].

Focus on:
- [ASPECT 1]
- [ASPECT 2]
- [ASPECT 3]

For each finding:
- Provide line numbers
- Rate severity (Critical/High/Medium/Low)
- Suggest remediation steps

Output Format

When presenting consultation results:

Executive Summary

Brief overview of key findings (2-3 sentences)

AI's Findings

Organized by priority:

Critical issues
High priority
Medium priority
Low priority

Your Analysis

Where you agree/disagree, additional insights

Synthesis & Recommendations

Combined recommendations with actionable next steps

Source

git clone https://github.com/jeffrigby/somepulp-agents/blob/main/plugins/second-opinion/skills/ai-consultation/SKILL.mdView on GitHub

Overview

The AI Consultation Skill provides structured workflows, prompt templates, and best practices for multi-model consultations (Codex, Gemini, etc.). It covers scenarios like second opinions, code reviews, security audits, and architecture analysis, emphasizing safe, read-only evaluation and clear output formats.

How This Skill Works

It organizes consultations into defined types (Code Review, Architecture, Debugging, Security Audit, Performance) and uses ready-made prompts and decision guidance. Interactions run in sandbox/read-only mode, with user confirmation required before any writes, and results are presented as an executive summary, prioritized findings, analysis, and actionable recommendations.

When to Use It

Getting second opinions from Codex or Gemini CLI
Code review consultations
Architecture decision validation
Security audits with multiple AI perspectives
Debugging with alternative diagnostic approaches
Performance reviews needing fresh insights

Quick Start

Step 1: Identify the consultation type and the scope of analysis (e.g., code review, architecture, security).
Step 2: Prepare a focused prompt using the templates and run in sandbox/read-only mode with clear constraints.
Step 3: Evaluate findings critically, synthesize external perspectives with your own analysis, and present actionable recommendations.

Best Practices

Always run consultations in sandbox/read-only mode to prevent unintended changes
External AI tools should not modify files; obtain user confirmation before any write operations
Be explicit about which files/components to analyze and provide necessary context
Compare external AI perspectives with your own analysis and validate recommendations
Request output in a structured format (line numbers, severity) and acknowledge uncertainty when present

Example Use Cases

Second opinion on an authentication flow in a microservices architecture
Code review for a Python data ingestion pipeline emphasizing security and performance
Architecture validation for an event-driven design with scalability tradeoffs
Security audit comparing OWASP considerations and data exposure risk across services
Debugging a memory leak using an alternative diagnostic approach from an external AI

Frequently Asked Questions

Add this skill to your agents