tear-sheet
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill instructions provide a set of JavaScript boilerplate functions and mandate that the agent generate and execute a Node.js script using the 'docx' library to assemble the final document. This involves runtime script generation based on provided code templates.
- [INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes untrusted narrative content from external sources (e.g., earnings transcripts, company summaries) via financial tools. While content quality rules require the agent to rewrite these narratives, the ingestion process remains a vector for indirect instruction injection.
- Ingestion points: Data is written to intermediate files in
/tmp/tear-sheet/(e.g.,earnings.txt,company-profile.txt). - Boundary markers: None explicitly defined in the instructions for the intermediate data storage or formatting.
- Capability inventory: The agent has file system access (read/write to /tmp and user outputs) and script execution capabilities.
- Sanitization: The skill enforces strict rules to rewrite and synthesize all external narrative data before inclusion in the final output.
- [COMMAND_EXECUTION]: The workflow requires the use of shell commands to manage temporary file directories and execute the document generation script.
Audit Metadata