play-learn-lift

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: LOW
Full Analysis
  • [PROMPT_INJECTION] (SAFE): No instructions found that attempt to bypass AI safety filters or override system constraints. The methodology actually emphasizes a human-in-the-loop process (PROPOSE/COMMIT phases) for any rule or skill upgrades.
  • [DATA_EXFILTRATION] (SAFE): No evidence of network operations, hardcoded credentials, or unauthorized access to sensitive file paths (e.g., SSH keys, AWS configs).
  • [EXTERNAL_DOWNLOADS] (SAFE): The skill does not reference or download any external scripts or third-party packages. All links are relative internal references to the MOOLLM project.
  • [COMMAND_EXECUTION] (SAFE): No instances of shell command execution, subprocess spawning, or arbitrary code evaluation were detected.
  • [INDIRECT_PROMPT_INJECTION] (LOW/SAFE): While the skill involves processing session logs (untrusted data), it explicitly mitigates risks by mandating a human review gate for any 'LIFT' operations that would result in permanent skill or rule changes. This follows best practices for platform-legible self-evaluation.
Audit Metadata
Risk Level
LOW
Analyzed
Feb 17, 2026, 08:01 AM