human-taste-code

Pass

Audited by Gen Agent Trust Hub on Feb 27, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill analyzes untrusted user-provided code without establishing clear boundaries or instructions to ignore embedded directives, making it susceptible to indirect prompt injection.
  • Ingestion points: User-provided code snippets, class hierarchies, and system architectures processed in SKILL.md.
  • Boundary markers: Absent. Neither the evaluation rubric nor the output template uses delimiters or explicit warnings to treat the target code strictly as data.
  • Capability inventory: Across all files (SKILL.md, examples.md, and references/research-sources.md), the skill's functionality is limited to generating text-based analysis reports and does not involve subprocess execution, file system modifications, or network operations.
  • Sanitization: No input sanitization or filtering (e.g., stripping comments from code) is performed before the agent evaluates the content.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 27, 2026, 01:23 PM