Summary

This repository provides a library of domain-agnostic agent skills designed for building autonomous software factories. It addresses the challenge of breaking down complex goals into independently verifiable pieces, enabling agents to pursue objectives iteratively and reliably. The skills are targeted at engineers and researchers developing advanced AI agents for software development, offering a framework for robust, goal-driven automation.

How It Works

The project frames software development as a "factory" process where agents autonomously pursue goals. Its core approach treats unknowns as a "fog of war," systematically mapping and carving the problem space into isolated, verifiable territories. This involves recursively re-slicing the plan as implementation reveals new information. Each component must pass rigorous verification steps (architecture, code, visual review) before the process advances. This iterative cycle ensures that each step gets "less wrong" until the goal is achieved, with re-planning occurring mid-implementation when the spec proves stale.

Quick Start & Requirements

Installation involves adding the skills via npx skills add dzhng/skills or by copying skill folders into a compatible harness's skills directory (e.g., .claude/skills/). The primary requirement is a supporting "harness" that integrates with these skills, such as Claude Code, Codex, opencode, Cursor, or duet, among over 70 others. No specific hardware, OS, or non-default software prerequisites are detailed beyond the need for a compatible agent environment.

Highlighted Details

• Offers domain-agnostic skills for creating autonomous software development agents and factories.
• Emphasizes a methodology of breaking down goals into independently verifiable slices and iterative refinement.
• Includes skills for comprehensive workflow management: write-spec, implement-spec, refactor-clean, write-tests, write-docs, and visual review tools like compare-screenshots.
• Demonstrates capability through an unattended 1-day, 16-hour Codex run that autonomously pursued and completed a single goal.

Maintenance & Community

No specific details regarding maintainers, community channels (e.g., Discord, Slack), roadmaps, or sponsorships are present in the provided README.

Licensing & Compatibility

The project is released under the MIT license. This permissive license generally allows for commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

The effectiveness and functionality of these skills are inherently dependent on the capabilities and integration of the external "harnesses" (AI agent frameworks) they are used with. The README does not detail specific limitations of the skills themselves or unsupported platforms.