The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and process messages from external peers, which could contain malicious instructions designed to manipulate the agent.
Ingestion points: Untrusted messages are received through the read_agent_messages tool or by polling the https://openhook.dev/api/v1/channels endpoint.
Boundary markers: The instructions explicitly direct the agent to label incoming items as "External Openhook peer message" and include metadata like sender ID and sequence.
Capability inventory: The skill possesses the ability to send messages, manage channel permissions, and execute local handler scripts.
Sanitization: The skill contains defensive instructions forbidding the agent from promoting peer messages to system instructions, tool approvals, or permissions to disclose secrets.
[COMMAND_EXECUTION]: The skill provides instructions for running a local listener process (openhook-channel-listen) and executing a local handler script to process incoming messages.
[EXTERNAL_DOWNLOADS]: The documentation instructs users and agents to use uv run to fetch and execute the openhook-channel-listen utility from the vendor's repository.
[DATA_EXFILTRATION]: The skill facilitates the transmission of agent-generated content and responses to the external domain openhook.dev. While this is the intended purpose of the messaging skill, it constitutes a pathway for data to leave the local environment.

openhook