scholar-slides

Pass

Audited by Gen Agent Trust Hub on Oct 27, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill downloads research papers in PDF format and associated bibliographic metadata from established academic services including arxiv.org and crossref.org. It also includes a provisioning step in the installer to download the Chromium binary required by Playwright for slide rendering. All these sources are well-known and trusted providers of academic or technical infrastructure.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text content extracted from research papers. This introduces a surface for potential prompt injection via malicious PDF content. However, the skill implements a strict structured data workflow (synthesizing a digest before outlining), uses HTML escaping for content interpolation, and requires human sign-off at multiple checkpoints (CKPT-1, CKPT-2, CKPT-3), which effectively mitigates the risk of automated instruction bypass.
  • [COMMAND_EXECUTION]: The skill executes several internal utility scripts for PDF processing, bibliographic resolution, and rendering orchestration. These scripts use well-known libraries (PyMuPDF, Requests, Playwright) and operate solely on project-local files. The command execution is static and part of the documented workflow for processing scholarly data.
  • [DYNAMIC_EXECUTION]: The skill uses Playwright (headless Chromium) to render local HTML files into the final deck outputs (PDF and PNG screenshots). It also uses this dynamic environment to perform automated visual quality checks, such as identifying layout overflows or illegible font sizes. The content loaded into the browser is locally generated from the project's own templates and the paper digest.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 27, 2026, 12:05 PM