ljg-paper
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted research papers from sources like arXiv or URLs, which represents a potential attack surface for instructions meant to manipulate the agent.\n
- Ingestion points: PDF files, arXiv links, and paper URLs (ReadingGuide.md).\n
- Boundary markers: Output is structured using Markdown templates and research record maps (
template.md,paper-map.md).\n - Capability inventory: The skill writes summarize files to
~/Context/and uses a local script for structural validation.\n - Sanitization: The instructions explicitly direct the agent to scan for and refuse to execute any 'suspicious instructions' found within papers, such as requests to leak info or modify the system.\n- [COMMAND_EXECUTION]: The workflow requires the use of the Bun runtime to execute a local validation script (
validate_note.ts) to verify note formatting and data density.\n- [EXTERNAL_DOWNLOADS]: The skill fetches academic papers from external URLs and the arXiv repository to perform its core summarization function.
Audit Metadata