giuseppe-trisciuoglio/developer-kit

Comprehensive developer toolkit providing reusable skills for Java/Spring Boot, TypeScript/NestJS/React/Next.js, Python, PHP, AWS CloudFormation, AI/RAG, DevOps, and more.

Quality

90%

Does it follow best practices?

Impact

Pending

No eval scenarios have been run

Securityby

Risky

Do not use without reviewing

This version of the tile failed moderation

Moderation pipeline encountered an internal error

CLI Delegation Guide

Name: giuseppe-trisciuoglio/developer-kit
Rating: 90.85333333333334 (1 reviews)
Author: giuseppe-trisciuoglio

Guide for delegating tasks to external AI CLI tools using gemini, copilot-cli, codex, and qwen-coder skills.

Overview

CLI delegation enables Claude Code to offload specific tasks to external AI coding assistants. This is useful when:

A different model may be better suited for a particular task
You want a second opinion from another AI model
The task requires capabilities specific to another model family
You want to compare outputs from multiple models

Skills Summary

Skill	Model Family	Best For	Not Recommended For
`gemini`	Google Gemini	Large context analysis, fast iterations	Complex multi-file refactoring
`copilot-cli`	Claude, GPT, Gemini	Multi-model comparison, flexible model choice	Single-model focused tasks
`codex`	OpenAI (o-series, GPT-5.3-codex)	Advanced code generation, architectural design	Simple tasks, quick queries
`qwen-coder`	Alibaba Qwen	Deep reasoning, second opinion	Production code generation

Common Patterns

1. Basic Delegation

All CLI skills follow a similar pattern:

Convert user request to English prompt
Select appropriate model based on task complexity
Choose approval mode based on risk level
Execute command with appropriate flags
Present results for user review

2. Model Selection Guide

Task Complexity
│
├── Simple (< 50 lines, single file)
│   └── gemini-3-flash, qwen2.5-coder, o4-mini
│
├── Moderate (50-200 lines, 1-3 files)
│   └── gemini-3-pro, gpt-5.2, claude-sonnet-4.6
│
└── Complex (> 200 lines, architecture, multi-file)
    └── gpt-5.3-codex, o3, gemini-3-pro, qwq

3. Approval Mode Selection

Mode	Use Case	Risk Level
`plan`	Read-only analysis, security reviews	Low
`default`	General coding tasks with confirmation	Medium
`auto_edit`	Trusted modifications with oversight	High
`yolo`	Experimental tasks only	Very High

4. Prompt Engineering

All delegated prompts should be:

Written in English
Specific and outcome-driven
Include file paths and constraints
Define expected output format
Include acceptance criteria when applicable

Prompt Template:

Task: <clear objective>
Context: <project/module/files>
Constraints: <do/don't constraints>
Expected output: <format + depth>
Validation: <tests/checks to run>

Quick Reference

Gemini CLI

# Quick analysis with flash
gemini -p "Analyze architecture" -m gemini-3-flash

# Deep analysis with pro
gemini -p "Analyze architecture" -m gemini-3-pro --approval-mode plan

# Structured output
gemini -p "Return JSON refactoring opportunities" --output-format json

Copilot CLI

# Basic delegation
copilot -p "Refactor service" --model gpt-5.2 --allow-all-tools

# Multi-model comparison
copilot -p "Compare" --model claude-sonnet-4.6
copilot -p "Compare" --model gpt-5.2

# Session resume
copilot --resume <session-id>

Codex CLI

# Code generation
codex exec "Generate REST API" -m gpt-5.3-codex

# Code review
codex review "Review for security" -a on-request -s read-only

# Complex refactoring
codex exec "Refactor to SOLID" -m gpt-5.3-codex -a on-request -s workspace-write

Qwen Coder CLI

# Basic delegation
qwen -p "Analyze code" -m qwen2.5-coder

# Deep reasoning
qwen -p "Analyze architecture" -m qwq --approval-mode plan

# Session resume
qwen -c <session-id>

Security Considerations

Always Use Safe Defaults

Prefer plan mode for analysis tasks
Use read-only or workspace-write sandbox for Codex
Avoid yolo mode unless explicitly requested
Never include secrets in prompts

Treat Output as Untrusted

Review all generated code before applying
Run tests and linting on suggested changes
Validate security implications
Check for breaking changes

Error Handling

Error	Cause	Solution
CLI not found	Tool not installed	Install the CLI tool
Authentication failed	Invalid credentials	Run auth command (`nlm login`, `qwen auth`)
Permission denied	Insufficient permissions	Use appropriate `--allow-*` flags
Rate limit exceeded	API throttling	Wait and retry with backoff

Integration with Claude Code Workflow

┌─────────────────────────────────────────────────────┐
│ User Request                                        │
└─────────────────┬───────────────────────────────────┘
                  │
                  ▼
┌─────────────────────────────────────────────────────┐
│ Claude Code                                          │
│  - Understands request                              │
│  - Determines if delegation is beneficial          │
│  - Invokes appropriate skill                        │
└─────────────────┬───────────────────────────────────┘
                  │
                  ▼
┌─────────────────────────────────────────────────────┐
│ CLI Skill                                           │
│  - Converts to English prompt                       │
│  - Selects model and flags                          │
│  - Executes delegation                             │
└─────────────────┬───────────────────────────────────┘
                  │
                  ▼
┌─────────────────────────────────────────────────────┐
│ External CLI Result                                 │
│  - Presented to user for review                   │
│  - User confirms before applying changes          │
└─────────────────────────────────────────────────────┘

When NOT to Delegate

Do not delegate when:

The task is simple enough for direct implementation
The task requires knowledge of the current codebase that the external tool lacks
Network access to the external service is unavailable
Authentication credentials are not configured
The task involves sensitive data that should not leave the local environment