Agent Skills is a packaging format for the reusable instructions and workflows an AI agent loads to perform a defined task. A skill is a directory containing a SKILL.md file, whose frontmatter and body the format specifies, optionally accompanied by scripts/, references/ and assets/ subdirectories. It sits alongside the Model Context Protocol as one of two component formats that extend agents: MCP defines how an agent connects to external tools and data, while Agent Skills define what instructions and procedures an agent carries with it.
The format originated at Anthropic, which released both it and the Model Context Protocol as open standards (Source: the-decoder.com). It is maintained as a separate specification at agentskills.io/specification.
Relation to Agent Plugins
The Agent Plugins 1.0.0 specification, published August 6, 2026 by a technical steering committee drawn from AWS, Cursor, Microsoft, OpenAI and Vercel, adopts Agent Skills as one of exactly two component types a conformant plugin may carry. It draws the boundary between the two documents explicitly: skills "MUST conform to the Agent Skills specification," which "is the source of truth for the SKILL.md format, frontmatter fields, and directory layout (scripts/, references/, assets/)," while Agent Plugins defines only how skills are discovered within a plugin — from a fixed skills/ directory, one skill per immediate child directory containing a SKILL.md regular file, with no recursive search deeper (Agent Plugins Specification v1.0.0 (Working Draft, August 2026)).
The Agent Plugins specification's stated reason for selecting Agent Skills and MCP over other proposed component types — commands, hooks, agents, rules and LSP servers — is that both "have established specifications outside this project and meaningful cross-client adoption," while the others "remain too client-specific for a stable portable contract" (Agent Plugins Specification v1.0.0 (Working Draft, August 2026)). A client may conform to Agent Plugins while supporting skills alone and no MCP servers.
Anthropic is not among the organisations on the Agent Plugins steering committee, an absence The Decoder noted alongside the observation that Anthropic originated both packaged formats and had separately added a plugin system to its own agent product (Source: the-decoder.com).
Deployment
Anthropic ships domain-specific skills with its vertical products. Claude for Healthcare, announced January 12, 2026, includes an Agent Skill for FHIR development covering cross-system interoperability and a prior-authorization review template that cross-references coverage requirements, clinical guidelines, patient records and appeal documents. The Claude for Life Sciences expansion added skills for scientific problem selection, instrument-data-to-Allotrope conversion, scVI-tools with Nextflow deployment for bioinformatics, and clinical-trial protocol draft generation accounting for FDA and NIH requirements, endpoint recommendations, competitive landscape and regulatory pathways (Source: anthropic.com). See Healthcare — AI Deployment.
Relationships
- related: Model Context Protocol (MCP) — the other component format packaged by Agent Plugins; complementary rather than competing.
- depended-on-by: Agent Plugins 1.0.0, Agent Plugins Specification v1.0.0 (Working Draft, August 2026) — the packaging specification that treats this format as its source of truth for skills.
- related: Agentic AI, Agent Architecture Patterns — the systems skills configure.
- related: Anthropic — the format's originator.
- deploys-in: Healthcare — AI Deployment — the documented vertical deployment.