cross-review-project

Pass

Audited by Gen Agent Trust Hub on Jun 20, 2026

Risk Level: SAFEREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill instructs the installation and execution of the cross-review-mcp package using npx to serve as a communication broker.
  • [DATA_EXFILTRATION]: Project source code, architectural patterns, and dependency graphs are systematically read and transmitted as JSON-encoded briefings and findings to an external MCP broker.
  • [COMMAND_EXECUTION]: Bash commands are used to modify the agent's environment by adding the third-party MCP broker to the configuration.
  • [PROMPT_INJECTION]: The agent is instructed to ingest and process untrusted data from a peer agent, creating a surface for indirect prompt injection.
  • Ingestion points: Reads peer briefings and findings from the broker and peer source code via the Read tool.
  • Boundary markers: Absent; peer-provided content is not explicitly wrapped in delimiters or safety instructions.
  • Capability inventory: Agent has access to Bash, Read, Write, and Edit tools.
  • Sanitization: Absent; external content is processed as valid input for the review cycle.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 20, 2026, 05:54 PM
Security Audit — agent-trust-hub — cross-review-project