Workshop transcript from WorkOS developer experience engineers Nick Nisi and Zack Proser covering 'skills' — reusable, composable markdown-based instruction units for AI coding agents like Claude. Skills solve the problem of repeatedly re-injecting context into stateless LLM conversations. Key topics include: skill structure (YAML front matter with name/description for LLM routing, optional scripts for deterministic data injection), when to use skills vs. claude.md memory files, progressive disclosure to avoid bloating context windows, confidence scoring patterns, skill evaluation frameworks, team-scale skill sharing and governance challenges, and skill marketplaces. The workshop includes a hands-on 'repo roast' skill-building exercise and Q&A covering skill conflicts, sub-agents vs. skills, and organizational management of shared skill libraries.
Sort: