self-reflection

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFENO_CODE
Full Analysis
  • No Code (SAFE): The provided files consist of documentation and configuration examples. No executable scripts, binaries, or source code were included in the skill package for analysis.\n- Indirect Prompt Injection (SAFE): The skill reads from a local memory file (~/workspace/memory/self-review.md). This ingestion point is used for reviewing past reflections, which is the primary purpose of the skill. Because the content is intended to be agent-generated and no high-risk capabilities are exposed, the risk is minimal.\n
  • Ingestion points: The read command accesses the reflection log file.\n
  • Boundary markers: None specified in the documentation.\n
  • Capability inventory: Commands are limited to logging and reading text-based reflections.\n
  • Sanitization: Not explicitly mentioned, but the data source is local.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 05:28 PM
Security Audit — agent-trust-hub — self-reflection