self-improvement

Pass

Audited by Gen Agent Trust Hub on Feb 20, 2026

Risk Level: SAFE
Full Analysis
  • PROMPT_INJECTION (SAFE): The skill uses reminders and hooks to prompt the agent to document its own activities. No instructions to bypass safety filters or ignore system prompts were found.
  • COMMAND_EXECUTION (SAFE): Bash scripts perform local file operations like directory creation and template instantiation. The extract-skill.sh script includes validation for user-provided skill names to prevent path traversal.
  • DATA_EXFILTRATION (SAFE): There are no network calls or attempts to access sensitive files. The error detector script reads standard environment variables provided by the agent environment for status checks only.
  • REMOTE_CODE_EXECUTION (SAFE): The skill does not download or execute code from external sources. All logic is contained within the local scripts and hook handlers.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 20, 2026, 03:48 PM