Definition
Agent skills are markdown (.md) documents packaging procedural knowledge for LLM agents — domain heuristics, tool-use policies, output constraints, and known failure modes — inserted into agent context before task execution.
Key Points
- External interface for adapting frozen models to enterprise workflows
- Traditionally hand-crafted or one-shot generated without systematic optimization
- skillopt treats skills as trainable parameters with validation gates
- Compact artifacts (300–2,000 tokens) portable across harnesses and models
Vercel Skills Ecosystem (2026)
npx skills addCLI — package manager for agent capabilities across 68+ agents (2026-07-06-vercel-skills-ai-agent-package-manager)- skills-sh directory: vercel-labs/skills ~25K GitHub stars; find-skills 2.3M installs
- Snyk ToxicSkills: 36.8% of 3,984 scanned skills flawed; 13.4% critical (2026-07-06-vercel-skills-snyk-toxicskills)
Optimization Challenges
Without systematic training:
- No step-size control → skill drift
- No validation → plausible edits regress performance
- No negative memory → repeated failed edits
Related
- skillopt
- text-space-optimization
- skills-sh
- vercel
- supply-chain-security
- coding-agents
- microsoft-agent-framework
- enterprise-ai-reliability