bridge-review

Warn

Audited by Gen Agent Trust Hub on Aug 22, 2026

Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from a "peer agent" via the agent-bridge tool. This peer output is untrusted and could contain malicious instructions aimed at manipulating the reviewer agent's subsequent actions.\n
  • Ingestion points: Data received from the peer agent via agent-bridge (SKILL.md).\n
  • Boundary markers: No explicit markers or "ignore instructions" directives are implemented to isolate the peer's output.\n
  • Capability inventory: The agent has permissions to read/write files and execute shell scripts (SKILL.md).\n
  • Sanitization: The skill lacks mechanisms to sanitize or validate the content of the peer review findings before use.\n- [DYNAMIC_EXECUTION]: The instructions direct the agent to "write and run temp script" to verify findings. This practice of generating and executing local code based on external input is a security risk.\n
  • Evidence: The skill body in SKILL.md explicitly calls for creating and running temporary scripts to validate whether a reported bug is real.\n- [COMMAND_EXECUTION]: The skill encourages the agent to "run the test" or "try the input" based on review findings, which involves executing arbitrary shell commands influenced by untrusted data.\n
  • Evidence: Section 3 ("Verify every finding") in SKILL.md describes these verification methods.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 22, 2026, 12:16 PM
Security Audit — agent-trust-hub — bridge-review