deep-research
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The main research script and its adapters use
importlib.utilto load localized extension modules (e.g.,reporting.py,workspace.py,run_metrics.py) located within the skill's internal directory. This dynamic loading is used solely for architectural modularity and state management. - [COMMAND_EXECUTION]: Evaluation and study scripts like
paired_evaluation.pyutilizesubprocess.runto invoke the skill's own research CLI (research.py). This allows for automated regression testing and blind comparison of research results within a controlled environment. - [EXTERNAL_DOWNLOADS]: The skill performs legitimate data retrieval from well-known academic and developer platforms, including Semantic Scholar, arXiv, OpenAlex, GitHub, and Kaggle. These requests use the
requestslibrary to fetch research content (JSON, HTML, PDF) for analysis and do not involve the execution of remote scripts. - [INDIRECT_PROMPT_INJECTION]: As the skill is designed to analyze web content, it possesses a vulnerability surface for indirect prompt injection. However, the instructions in
SKILL.mdexplicitly direct the agent to treat all analyzed materials as untrusted data and strictly forbid the execution of any commands, installations, or role-play instructions found within search results or files.
Audit Metadata