Agent Skills

AI coding agents typically take the shortest path to implementing a feature, tending to skip essential senior engineer processes such as Spec, Test, and Review. This is a major cause of shipping unreliable software.

Agent Skills is a scaffolding framework that helps agents avoid skipping these essential steps. The core differentiator of this project is that it defines 'skills' not as mere prose, but as Workflows with clear steps and exit criteria.

Reflecting the software development lifecycle (SDLC), the project provides the following key stages:

• Define (/spec): Defining requirements and design
• Plan (/plan): Breaking down tasks and establishing a plan
• Build (/build): Implementation stage
• Verify (/test): Verification through testing
• Review (/review): Code review and catching errors
• Ship (/ship): Safe deployment

Rather than simply instructing agents to "test well," the goal is to inject an actionable process into the agent's context—such as "write a failing test first, make it pass with minimal code, then refactor"—to increase the reliability of the work.