article-study
Pass
Audited by Gen Agent Trust Hub on Aug 11, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill processes untrusted external data (URLs, local files, and PDFs) to generate educational content. This ingestion constitutes a surface for indirect prompt injection where malicious instructions in the source material could attempt to override agent logic.
- Ingestion points: Identified in SKILL.md and 教学法.md as links, local files, and PDF documents.
- Boundary markers: The skill requires a structured summary ('资料-<主题>.md') and explicitly mandates the separation of original text from agent-generated inferences (Red Line 5).
- Capability inventory: File system access for directory creation and courseware generation; network access for fetching articles.
- Sanitization: The skill utilizes local templates for HTML generation, minimizing raw injection risk, though specific escaping logic for user data in generated JS is not explicitly detailed.
- [COMMAND_EXECUTION]: The skill performs routine file system management, such as using 'mkdir -p' to create structured learning directories and writing Markdown and HTML files to the local environment. These actions are scoped to the skill's purpose of organizing study progress and artifacts.
Audit Metadata