The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of a vendor-provided CLI tool (primitive) to discover peers, manage network membership, read/write agent notes, and send messages. Examples include primitive network peers, primitive agent notes get, and primitive send.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data provided by other agents in the network.
Ingestion points: Untrusted data enters the context via primitive agent notes get AGENT_INFO and primitive agent notes get AGENT_WORKING commands in SKILL.md.
Boundary markers: Technically absent; there are no specific markers or delimiters mentioned to wrap the content of fetched notes to distinguish them from system instructions.
Capability inventory: The skill possesses capabilities to send messages (primitive send, primitive chat), write its own notes (primitive agent notes set), and perform administrative network tasks if authorized (primitive network set).
Sanitization: Absent; the skill does not specify any validation or filtering of the content retrieved from peer notes before the agent processes it. However, it does provide behavioral guidelines to the agent, noting that self-reported data is not authoritative.

primitive-network