Coding Agents — Development skill for Claude Code

🤖 Coding Agents

A **multi-agent coding assistant**. Give it a task in plain English; a **Planner**, a **Coder** and a **Reviewer** (all powered by [Claude](https://www.anthropic.com)) collaborate inside a sandboxed workspace folder until the code is written, tested and approved.

$ coding-agents "Build a CLI todo app with add/list/done commands and tests"

🔁 round 1/3
🧭 planner finished: 1. Create todo.py with a JSON-backed TodoStore ...
  🛠️  coder: write_file {'path': 'todo.py', ...}
  🛠️  coder: write_file {'path': 'test_todo.py', ...}
  🛠️  coder: run_command {'command': 'python -m pytest -q'}
🛠️  coder finished: Implemented todo.py and tests; all 6 tests pass.
  🔍 reviewer: run_command {'command': 'python -m pytest -q'}
🔍 reviewer finished: VERDICT: APPROVED

APPROVED   Rounds used: 1

Table of contents

• How it works
• Quick start
• CLI options
• Use as a library
• Design notes
• Safety
• Project layout
• Testing
• Limitations and roadmap
• Contributing

How it works

flowchart LR
    T([Task]) --> P[Planner
read-only]
    P -->|plan| C[Coder
read, write, run]
    C -->|summary| R[Reviewer
read, run tests]
    R -->|VERDICT: APPROVED| D([Done])
    R -->|CHANGES_REQUESTED + feedback| C

Agent	Tools	Job
Planner	list_files, read_file	Turns the task into a short numbered plan, no code
Coder	list_files, read_file, write_file, run_command	Implements the plan, writes tests, runs them, fixes failures
Reviewer	list_files, read_file, run_command	Reads the code, runs the tests, ends with VERDICT: APPROVED or VERDICT: CHANGES_REQUESTED plus fixes

The Coder and Reviewer repeat until the Reviewer approves or `--max-rounds` is reached. Reviewer feedback is passed to