The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill enables the agent to read messages from other agents or sessions via the hive msg inbox command, which constitutes an ingestion point for untrusted external data.
Ingestion points: The hive msg inbox command (defined in SKILL.md) fetches message payloads from a messaging topic.
Boundary markers: While the output format is structured as JSON Lines, there are no explicit instructions or delimiters provided to the agent to treat the content of the payload field as untrusted or to disregard potential instructions embedded within those messages.
Capability inventory: The skill provides read access to messages. The risk involves the agent potentially following malicious instructions found in a message payload using other tools available in its environment (e.g., file system access or shell execution).
Sanitization: The documentation does not indicate any sanitization, filtering, or validation of the message content before it is presented to the agent.

inbox