intercom-automation
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external support interactions which may contain malicious instructions designed to manipulate agent behavior.
- Ingestion points: The skill retrieves conversation content and contact details using tools such as INTERCOM_LIST_CONVERSATIONS, INTERCOM_GET_CONVERSATION, and INTERCOM_SEARCH_CONTACTS as described in SKILL.md.
- Boundary markers: There are no instructions for implementing delimiters or special handling to separate user data from system instructions.
- Capability inventory: The skill includes powerful write capabilities such as INTERCOM_CREATE_CONVERSATION, INTERCOM_REPLY_TO_CONVERSATION, and INTERCOM_CLOSE_CONVERSATION in SKILL.md.
- Sanitization: The instructions lack guidance for sanitizing or validating external data before it is processed by the agent.
- [DATA_EXFILTRATION]: The skill configuration requires connecting to a third-party MCP server endpoint at https://rube.app/mcp which is not on the standard whitelisted domains list.
Audit Metadata