kb-retriever
Pass
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted local data (PDF, Excel, Markdown) and provides this content to the agent's context. This creates a surface for indirect prompt injection attacks where malicious instructions embedded in documents could influence agent behavior.
- Ingestion points: Files within the
knowledge/directory or user-provided paths. - Boundary markers: None explicitly mentioned in the instructions to separate retrieved data from the agent's system prompt or to instruct the agent to ignore instructions within the data.
- Capability inventory: File reading (
read_file), search (grep), PDF extraction (pdftotext,pdfplumber), and data analysis (pandas). - Sanitization: No specific sanitization or filtering of the extracted text is performed before it is added to the context.
- [COMMAND_EXECUTION]: The skill instructions direct the agent to execute various shell commands and Python scripts for file processing.
- Evidence: Instructions include using
test -dfor directory checks,pdftotextfor PDF extraction, and executing the providedscripts/convert_pdf_to_images.pyscript. While these are part of the intended functionality, they involve shell interaction with user-specified paths.
Audit Metadata