confidence-calibration

Pass

Audited by Gen Agent Trust Hub on Feb 19, 2026

Risk Level: SAFE
Full Analysis
  • [Prompt Injection] (SAFE): The skill instructions focus on self-calibration and do not contain patterns intended to bypass safety filters, override system instructions, or extract internal prompts.
  • [Data Exposure & Exfiltration] (SAFE): No sensitive file paths, hardcoded credentials, API keys, or network exfiltration patterns were identified.
  • [Remote Code Execution] (SAFE): There are no external downloads, package installations, or piped shell commands. The included Python snippet is a static dictionary mapping keywords to domains and does not involve dynamic execution.
  • [Indirect Prompt Injection] (SAFE): Although the skill processes user feedback (success/failure signals) to update performance stats, it lacks any dangerous capabilities (such as file-writing or network access) that could be exploited by an attacker via the ingestion point.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 19, 2026, 03:51 PM