self-improve

Pass

Audited by Gen Agent Trust Hub on Feb 18, 2026

Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
  • Indirect Prompt Injection (LOW): The skill is designed to ingest session context (which may contain untrusted data) and promote it to permanent instruction files.
  • Ingestion points: Session history, debugging insights, and pattern detection (SKILL.md).
  • Boundary markers: Strong mandatory human-in-the-loop requirement. SKILL.md states: 'Never auto-apply changes. Always get explicit approval.'
  • Capability inventory: Writing to CLAUDE.md, SKILL.md, and project memory paths via agent file tools.
  • Sanitization: No automated sanitization is described; security relies entirely on the user's manual review of proposed changes.
  • No Code (SAFE): The skill consists exclusively of instructional markdown files and contains no executable scripts, binaries, or package dependencies.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 18, 2026, 06:51 PM