Agent skills are becoming a new layer of software, but like code, they go stale. As APIs, frameworks, specifications, and best practices evolve, manually maintaining skills does not scale. This session shows how to turn an AI coding agent into a continuous maintainer for your skill library.

Using my open-source Web AI Agent Skills repository and GitHub Agentic Workflows, I’ll demonstrate a weekly workflow that researches changes, detects outdated or missing knowledge, proposes skill updates, and creates changes for human review. We’ll explore the architecture, guardrails, validation, and lessons learned, and show how this pattern can make skills versioned, testable, continuously improved, and treated as first-class software artifacts.

- Treat skills as software: design for versioning, validation, evaluation, distribution, and continuous maintenance, not one-time authoring.
- Automate skill maintenance: use agentic workflows to continuously detect knowledge drift, research changes, and propose updates.
- Build with guardrails: combine autonomous maintenance with constrained changes, automated checks, and human review to keep the skill supply chain trustworthy.

Keywords: Agent Skills, Context Engineering, AI Coding Agents, Agentic Workflows, AI-Native Development, Continuous AI, Skill Lifecycle, Autonomous Maintenance, Knowledge Drift, Context Drift, Developer Experience, AI Governance, Evaluation, GitHub Agentic Workflows, Web AI