geo-fix-content
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted HTML from user-provided URLs, creating a surface for indirect prompt injection where malicious instructions could be embedded in the analyzed text.
- Ingestion points: Phase 1.1 in
SKILL.md(URL fetching and extraction). - Boundary markers: The 'Security: Untrusted Content Handling' section specifies a mental wrapper using
<untrusted-content>tags. - Capability inventory: Network read (content fetching), file write (creating markdown reports), and natural language processing.
- Sanitization: The skill provides explicit instructions to ignore and report instruction-like strings found in external content.
- [EXTERNAL_DOWNLOADS]: The skill performs runtime network requests to fetch content from arbitrary user-provided URLs.
- [COMMAND_EXECUTION]: The agent is instructed to execute network operations for content retrieval and file system operations to generate analysis reports.
- [PROMPT_INJECTION]: A static detector flag for instruction override was identified as a false positive. The strings in question are part of a defensive instruction intended to protect the agent from external data, not an attempt to compromise the agent.
Audit Metadata