literature-review

Pass

Audited by Gen Agent Trust Hub on May 14, 2026

Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
  • [PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection via retrieved research data. While inherent to the skill's purpose, it constitutes a potential attack vector if an external source contains malicious instructions.
  • Ingestion points: Data retrieved from Semantic Scholar, web searches, videos, and Wikipedia (SKILL.md).
  • Boundary markers: Absent; the template for the literature review does not employ delimiters to separate retrieved information from the agent's instructions.
  • Capability inventory: File system write access limited to creating markdown files within the literature/ directory.
  • Sanitization: No sanitization or verification of the external source content is specified before processing.
  • [NO_CODE]: This skill consists of natural language instructions and output templates. No executable scripts, binaries, or configuration files were included in the provided content.
Audit Metadata
Risk Level
SAFE
Analyzed
May 14, 2026, 08:52 AM