experiment-design

Pass

Audited by Gen Agent Trust Hub on Apr 30, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: No malicious patterns such as remote code execution, obfuscation, or credential harvesting were detected. The skill is entirely instructional.
  • [DATA_EXPOSURE_AND_EXFILTRATION]: No exfiltration patterns detected. The skill defines a local project directory (~/cold-email-ai-skills/...) for saving experiment configurations, which is used for documentation and does not involve system-sensitive paths.
  • [INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface by ingesting user-provided hypothesis strings and variable names.
  • Ingestion points: User-defined hypotheses and experiment variables in SKILL.md.
  • Boundary markers: Not present in the suggested YAML template.
  • Capability inventory: File-write operations to local project paths; network operations are handled by external skills.
  • Sanitization: No sanitization or validation logic is specified for the user-supplied content.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 30, 2026, 10:21 AM
Security Audit — agent-trust-hub — experiment-design