experiment-monitor

Pass

Audited by Gen Agent Trust Hub on Jun 21, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting and processing data from experiment specifications and metric files without strict boundaries.
  • Ingestion points: Data is read from experiments/active/* and metrics/daily/* via the state_read tool in SKILL.md.
  • Boundary markers: There are no explicit delimiters or instructions to the agent to ignore any embedded commands within the processed experiment specifications.
  • Capability inventory: The skill has the capability to write to the repository state (metrics/alerts.md) and generates instructions for the user based on the processed data.
  • Sanitization: No validation or sanitization is performed on the data extracted from the state files before it is used in logic or reports.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 21, 2026, 06:29 PM
Security Audit — agent-trust-hub — experiment-monitor