directory-federation

Pass

Audited by Gen Agent Trust Hub on Jun 29, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The code example demonstrates a tool that returns a shell command string (docker run -i --rm mcp/wikipedia-mcp) intended for direct execution by the client agent.
  • [PROMPT_INJECTION]: The skill architecture facilitates indirect prompt injection by design, where tool metadata from one source dictates subsequent execution logic.
  • Ingestion points: The return value of the search_available_tools_and_servers tool in SKILL.md.
  • Boundary markers: No specific delimiters or instructions are provided to the agent to treat the metadata as untrusted.
  • Capability inventory: The pattern assumes the agent has permissions to execute shell commands and establish network connections based on external data.
  • Sanitization: The example lacks validation or escaping for the returned command strings or server connection parameters.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 29, 2026, 08:36 PM
Security Audit — agent-trust-hub — directory-federation