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
Risk Level
SAFE
Analyzed
Sep 15, 2026, 03:36 PM
Security Audit — agent-trust-hub — geo-fix-content