Cap

Caps are modular units of functionality callable by agents, defined using the Agent Capability Protocol (ACP). They replace traditional app interfaces with composable service endpoints. Each Cap represents a discrete skill or function that agents can install, execute, and remove dynamically.

Cap Components

Caps are described through .acp.yaml files that contain:

• A JSON schema specifying expected inputs, outputs, and state structures.
• A system prompt guiding the LLM on how to use the capability.
• A tool manifest for runtime invocation.
• Metadata including ID, permissions, triggers, and compatibility.

Cap Lifecycle

Caps can be registered in a decentralized registry and discovered by agents at runtime. Users can:

• Install new capabilities on demand.
• Compose multiple Caps for complex workflows.
• Update or remove them without affecting global agent behavior.

For full specification, refer to NIP-7.