paper-reading-workspace-presentation

Pass

Audited by Gen Agent Trust Hub on Jun 24, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill's React template (package.json) specifies standard, well-known dependencies including react, vite, lucide-react, and react-markdown which are downloaded from the npm registry during project setup.
  • [COMMAND_EXECUTION]: The skill instructs the agent to help the user run npm install and npm run dev to start the local presentation server once the scaffold is generated.
  • [DATA_EXFILTRATION]: The generated application enables local file preview using Vite's /@fs/ feature. Access is restricted to the workspace root by default through the server.fs.allow configuration in the generated vite.config.ts. Source files are indexed in source-index.json with their paths to facilitate local serving.
  • [PROMPT_INJECTION]: The skill processes external research artifacts to generate summaries, creating an indirect prompt injection surface.
  • Ingestion points: Research files (e.g., field_boundary.md, research_clusters.md) defined in references/presentation-display-map.md are read as sources.
  • Boundary markers: None explicitly defined in the generation rules in SKILL.md.
  • Capability: The skill writes generated metadata and summaries to the filesystem and guides the user to host a local dev server.
  • Sanitization: Content is rendered via react-markdown in the frontend; summaries are generated by the agent based on source artifacts.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 24, 2026, 03:30 AM
Security Audit — agent-trust-hub — paper-reading-workspace-presentation