The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the herdr CLI tool to perform agent coordination tasks. This involves executing commands like herdr agent get, herdr agent read, and herdr agent prompt to interact with other agent panes and manage their transcripts.
[INDIRECT_PROMPT_INJECTION]: Because the skill processes messages from other agents, it creates an attack surface for indirect prompt injection. The skill provides clear instructions to mitigate this vulnerability:
Ingestion points: Peer messages are received and transcripts are read from other agents using the herdr agent read command in the SKILL.md file.
Boundary markers: The protocol requires a specific envelope format [herdr peer ...] to distinguish metadata from content.
Capability inventory: The skill uses herdr commands to discover sessions, inspect panes, and send prompts between agents.
Sanitization: The instructions explicitly mandate origin inspection for unexpected or consequential messages and require the agent to seek user approval before performing sensitive, destructive, or authority-changing actions.

herdr-agent-coordination