hunt-agent-mcp-exposure
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes several example shell commands using
curlto interact with JSON-RPC endpoints. These commands are templates for security researchers to enumerate resources and tools on target MCP and A2A servers. - [DATA_EXFILTRATION]: The provided Python probe utilizes the
urllib.requestmodule to perform POST requests against user-defined network endpoints. While designed for security auditing of remote targets, these represent network operations that ingest data from external sources. - [INDIRECT_PROMPT_INJECTION]: The skill acts as a surface for indirect prompt injection because it is designed to ingest and display raw, potentially untrusted data from external JSON-RPC interfaces.\n
- Ingestion points: Data is ingested via the
rpcfunction'surllib.request.urlopencall in the Python script (SKILL.md).\n - Boundary markers: The instructions do not define delimiters or specific safety warnings to prevent the agent from being influenced by malicious instructions embedded within the target's JSON-RPC responses.\n
- Capability inventory: The skill has the capability to perform network requests and execute system commands through the agent's tools (SKILL.md, references/a2a-protocol-attack.md).\n
- Sanitization: There is no evidence of sanitization or validation of the content returned from the external endpoints before it is processed by the agent.
Audit Metadata