messages
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes messages from external sources and other agents which are delivered directly into the agent's prompt iteration. This creates a surface for indirect prompt injection where malicious content could be used to influence the agent's actions.
- Ingestion points: According to SKILL.md, incoming messages from external sources and named channels are delivered inline in the iteration prompt.
- Boundary markers: The skill does not define or enforce boundary markers or delimiters for the external message content in its instructions or scripts.
- Capability inventory: The skill possesses the capability to send messages, reply to requests, and manage channel subscriptions via a Unix domain socket (TARIBOY_TOOLS_SOCKET), as seen in scripts/messages.py.
- Sanitization: No sanitization, escaping, or filtering of the message content is performed within scripts/messages.py before data is sent to the daemon.
Audit Metadata