doc-processing
Pass
Audited by Gen Agent Trust Hub on Jun 23, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill includes a script to download documents from remote URLs using the well-known httpx library. This is the primary intended functionality for document retrieval from the web.
- [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection because it processes content from external files (PDFs, DOCX) into markdown and subsequently instructs the agent to read that content to extract metadata. An attacker could embed malicious instructions in a document to attempt to influence the agent's actions during the metadata enrichment stage.
- Ingestion points: Untrusted data enters the agent context via sources/docs//content.md after being converted from external files.
- Boundary markers: Absent. The agent is instructed to read the generated markdown files directly without specific delimiters or warnings to ignore embedded instructions.
- Capability inventory: The skill has the capability to create directories, write files (meta.md, content.md), and execute local Python scripts.
- Sanitization: Content is converted from binary formats to markdown without explicit sanitization or filtering of text patterns that might resemble instructions.
Audit Metadata