Term
Agent Skills
Modular, reusable instruction packages (e.g. Claude Agent Skills) that an AI agent loads from disk on demand to perform specialized tasks — built around a SKILL.md holding metadata and instructions, plus optional scripts and reference files.
Agent Skills — in more detail
Agent Skills are self-contained, reusable capability packages for AI agents. Technically, a skill is a directory containing a central SKILL.md file that starts with YAML frontmatter (name and description), followed by the actual instructions. Alongside it can sit further Markdown references, scripts (e.g. Python), and resources.
The key mechanism is progressive disclosure, a three-level loading system. Level 1: at startup the agent loads only the name and description of every installed skill into its system prompt — just enough to recognize when a skill is relevant, without consuming context. Level 2: once a skill matches the task, the agent reads the full SKILL.md body. Level 3+: referenced extra files are pulled in only when needed. Files that aren’t required stay on disk and cost zero tokens.
Anthropic introduced Skills in October 2025 and opened the format as a cross-platform standard (agentskills.io). The same SKILL.md therefore works across Claude Code, Codex, and Gemini CLI.
Example / In practice
A PDF skill bundles several building blocks:
pdf-skill/
├── SKILL.md (main instructions + metadata)
├── FORMS.md (form-filling guide)
├── REFERENCE.md (detailed API reference)
└── scripts/
└── fill_form.pyIf the agent has to fill out a form, it reads SKILL.md, then specifically FORMS.md and fill_form.py — REFERENCE.md stays unloaded. This lets specialized expertise be encapsulated once and reused across many tasks and projects, instead of bloating a monolithic system prompt.
Distinction from similar terms
Tools / function calling give the agent individual executable functions with a defined schema that the model invokes at runtime. MCP (Model Context Protocol) standardizes how external tools and data sources are connected via a server. Agent Skills sit one level above: they provide no new interface but rather instructions, conventions, and example files — pre-packaged know-how the agent loads into context when needed. A skill can use tools or MCP servers, but does not replace them.
Compared with a monolithic system prompt, skills are modular, loaded on demand, and therefore token-efficient: instead of keeping everything permanently in context, only the relevant parts are added.
Sources
- Anthropic — Equipping agents for the real world with Agent Skills (anthropic.com/engineering)
- Claude Platform Docs — Agent Skills Overview (platform.claude.com/docs)
- agentskills.io — cross-platform Agent Skills standard
Entdecke mehr
AI Workflows by Keyword: How We Make Recurring Routines Enforceable
A typed keyword triggers a fixed AI routine — and every single step must be committed before the next one appears. Why that's the actual trick.
GlossarEnsemble / Multi-Model Orchestration
Ensemble means combining several deliberately varied LLM runs or models whose findings complement each other. Multi-model orchestration drives these runs via orchestrators with sub-agents, so the union of results is larger than any single run.
LexikonFunction Calling / Tool Use
How an LLM uses tools: define a tool as a schema, the model picks the function and arguments, the result returns to the chat — the basis of every agent.