r3f-skill-smith

Pass

Audited by Gen Agent Trust Hub on Jun 26, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is designed to ingest findings and lessons from external code projects to update agent instructions, creating a surface for indirect prompt injection. \n
  • Ingestion points: Run signals, findings, and verified rules gathered from external scene jobs (SKILL.md). \n
  • Boundary markers: No explicit data isolation delimiters are implemented in the instruction set. \n
  • Capability inventory: File system read and write access to local markdown files and sibling skill checklists. \n
  • Sanitization: Implements a rigorous validation framework (references/lesson-promotion.md) that filters for generality and requires a mandatory human review step before updating instructions. \n- [SAFE]: The skill performs file system modifications to maintain its own logic and that of its sibling skills. This self-modification is the primary function of the tool and is constrained by a documented 'suggest-then-apply' human approval process, preventing autonomous or unauthorized changes to the agent's core checklists.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 26, 2026, 04:55 AM