research-plan
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The initialization script (scripts/initialize.py) includes a security validation check that prevents directory traversal and symbolic link attacks by ensuring all task-specific directories are children of the current working directory.
- [EXTERNAL_DOWNLOADS]: The skill is configured to interact with well-known and reputable scientific research APIs and databases, including Semantic Scholar, PubMed, IEEE Xplore, and arXiv, for fetching metadata and research documents.
- [COMMAND_EXECUTION]: The skill uses the GitHub CLI tool (gh) to facilitate bug reporting to the vendor's repository (huangwb8/bensz-bugs) when explicitly requested by the user, representing standard vendor-specific functionality.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted research data from external PDFs and search APIs (Ingestion Points: search results and PDFs described in SKILL.md). While explicit boundary markers are absent, the skill's capabilities (Inventory: file-write in scripts/initialize.py and BibTeX generation) are mitigated by robust sanitization routines in scripts/utils.py and scripts/bibtex.py, as well as schema validation in scripts/validate.py.
Audit Metadata