research-scout

Pass

Audited by Gen Agent Trust Hub on Oct 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The script scripts/test_research_scout_scripts.py uses subprocess.run() to execute the local aggregation script during unit testing. This is standard behavior for verifying tool functionality in a development environment and does not involve executing remote or untrusted code.
  • [INDIRECT_PROMPT_INJECTION]: The skill inherently processes untrusted data from external sources such as research paper abstracts. This creates a potential surface for indirect prompt injection. However, the skill implements strong mitigations, including explicit instructions in SKILL.md and references/source-mix-and-compliance.md to ignore instructions embedded in papers, and the use of deterministic scripts and rule-based "trap filters" to verify the validity of findings.
  • [EXTERNAL_DOWNLOADS]: The various generator scripts in the scripts/ directory produce search URLs for trusted academic and industry sources, including arXiv, Hugging Face, Google Research, and Semantic Scholar. These references are used for legitimate research purposes and conform to the platform's trusted vendor guidelines.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 17, 2026, 08:26 AM