synthesize-documents

Pass

Audited by Gen Agent Trust Hub on May 18, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill is designed to process external documents which presents a standard surface for indirect prompt injection. However, the instructions implement a rigorous 8-step workflow that forces the model to categorize and evaluate content rather than blindly executing it.
  • [DATA_EXFILTRATION]: No unauthorized data exfiltration patterns were found. The skill uses external search only for fact-checking conflicts or verifying factual claims as intended by the primary synthesis task.
  • [COMMAND_EXECUTION]: The skill does not perform arbitrary shell commands. It requests the creation of a new Markdown file to save the synthesis, which is a legitimate and scoped file operation consistent with its purpose.
  • [EXTERNAL_DOWNLOADS]: No external scripts or untrusted packages are downloaded or executed. The skill relies on the agent's built-in search and file access capabilities.
  • [SAFE]: The skill demonstrates best practices for content synthesis, including source inventory tracking and explicit disagreement mapping, which reduces the risk of accidental instruction following from processed data.
Audit Metadata
Risk Level
SAFE
Analyzed
May 18, 2026, 07:31 AM
Security Audit — agent-trust-hub — synthesize-documents