What are Agent Skills?
Agent Skills are reusable packages of instructions, and optionally scripts and reference files, that extend what an AI agent can do without modifying its code or model. In the format popularised by Anthropic, a skill is a folder containing a SKILL.md file with YAML frontmatter (name and description) followed by the instructions an agent loads when a task matches.
WHY IT MATTERS
Skills solve a distribution problem: how to give an agent procedural knowledge — house style, a deployment runbook, a document format — without retraining a model, rewriting a system prompt, or building a new integration. A skill is just files, so it can be versioned, reviewed, and shared like any other artefact.
The format relies on progressive disclosure. At startup the agent pre-loads only each skill's name and description; the full SKILL.md body, and any bundled references or scripts, are read into context only when a task actually matches. That keeps dozens of installed skills cheap until one is needed.
Skills complement rather than replace MCP tools. Tools give an agent capabilities — API access, file operations, queries — while skills give it procedures for using those capabilities well. The two compose: a skill can instruct an agent on which tools to call and in what order. Because skills can also bundle executable scripts, they carry a supply-chain consideration similar to installing any third-party code: teams should review skills from external sources before adoption, the same way they vet an MCP server.
HOW POLICYLAYER USES THIS
Skills shape how an agent decides to act; PolicyLayer governs what happens when it does. A skill may instruct an agent to call particular MCP tools, but when that traffic routes through the PolicyLayer gateway, each tools/call is still evaluated against deterministic policy — so a misjudged or malicious instruction package cannot reach tools the calling identity was never granted.