learning-pack

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill retrieves information from external websites via multiple research agents in Stage 2 to build the learning package. This external data is untrusted and is processed into the final documentation.
  • Ingestion points: Web search results regarding system documentation, video metadata, and community experience (found in references/派单模板.md).
  • Boundary markers: The skill does not implement specific delimiters or safety instructions to prevent the agent from obeying instructions embedded in the search results.
  • Capability inventory: The skill has access to local shell execution via scripts/html2png.sh and scripts/md2pdf.sh, and it performs file operations within the ~/Downloads/ directory.
  • Sanitization: There is no explicit validation or escaping of the ingested search data before it is formatted into Markdown or HTML.
  • [COMMAND_EXECUTION]: The skill utilizes system-level shell scripts to automate document conversion. scripts/html2png.sh uses headless browsers for image capture, and scripts/md2pdf.sh employs pandoc and the Mermaid CLI to generate PDF files. These operations are intended parts of the skill's file-generation pipeline.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 09:50 AM
Security Audit — agent-trust-hub — learning-pack