scratchpad-fetch
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- Indirect Prompt Injection (LOW): The skill is designed to ingest untrusted external data into the agent's context without sanitization.
- Ingestion points:
scripts/fetch_urls.shdownloads content from any user-provided URL usingcurl. - Boundary markers: The script uses simple Markdown headers (
## Source: $url) and separators (---), but lacks explicit instructions to the LLM to ignore any command-like text within the downloaded content. - Capability inventory: The agent has access to the
Bashtool, which could be exploited if the agent follows instructions found in a fetched document. - Sanitization: None. Raw content from the web is appended directly to the scratchpad files.
- External Downloads (LOW): The script performs network operations using
curl -sLto fetch data from non-whitelisted domains. While this is the primary purpose of the skill, it represents a standard risk associated with processing external web data.
Audit Metadata