Do I need to supervise every step of the process?

No, this skill is designed for automated execution. It only escalates to a human if the agent fails to converge on a solution after three review cycles or if it encounters an ambiguity requiring human input.

What happens if a subagent fails a code review?

The implementer subagent is provided with the reviewer's feedback and must fix the issues. The reviewer then re-evaluates the code. This cycle repeats until both specification and quality standards are met.

Does it support Test-Driven Development (TDD)?

Yes, subagents are instructed to follow TDD principles for every task, ensuring that functionality is verified by tests before being submitted for review.

Which other skills are required to use arc-agent-driven?

It requires arc-using-worktrees for workspace isolation and is designed to work with arc-writing-tasks for the initial task list creation.

How does arc-agent-driven differ from standard task execution?

Unlike standard execution, it dispatches a fresh subagent for every task to avoid context drift and enforces a two-stage review process (spec then quality) before proceeding to the next item.

Arc Agent-Driven Workflow

Name: Arc Agent-Driven Workflow
Author: GregoryHo

byGregoryHo

0•

Productivity & Workflow

Executes complex task lists using isolated subagents and a rigorous two-stage review process to ensure high-quality, spec-compliant code.

arc-agent-driven is an advanced orchestration skill for Claude Code that automates the execution of development task lists. It operates by dispatching a fresh, context-isolated subagent for every individual task, which prevents context pollution and ensures focused implementation. The workflow enforces a strict two-stage verification process—validating specification compliance first, followed by a code quality review—to maintain professional standards without requiring constant human intervention. This skill is ideal for executing large feature epics where tasks are relatively independent, integrating seamlessly with TDD practices and worktree management for a scalable AI-driven development experience.

Key Features

01Isolated subagent dispatch per task to eliminate context drift and confusion

02Mandatory two-stage review cycle covering specification compliance and code quality

03Worktree integration for managing isolated development environments

040 GitHub stars

05Automated feedback loops that allow subagents to ask clarifying questions

06Built-in support for Test-Driven Development (TDD) during implementation

Use Cases

01Ensuring high-quality code contributions through multi-stage AI reviewer subagents

02Automating the sequential implementation of a large feature set or epic

03Executing independent software tasks with strict quality gates and no human-in-the-loop

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add gregoryho/arcforge arc-agent-driven

For use in Claude.ai and ChatGPT

Download Skill