graph-it-refactor
Pass
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill follows security best practices by requiring a read-only analysis phase (Phase 1) to identify the impact of changes before invoking any write-capable tools (Phase 2). This prevents 'blind editing' and ensures the agent has full context before attempting changes.
- [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because it ingests and processes untrusted source code during its impact analysis phase. This is a common characteristic of developer tools and is mitigated by the skill's structured approach.
- Ingestion points: Reads source code and file structures through the graphitlive tools listed in SKILL.md.
- Boundary markers: No explicit delimiter instructions or 'ignore embedded instructions' warnings are provided for the tool outputs.
- Capability inventory: The skill uses write-capable tools such as edit_file and apply_patch during Phase 2 (SKILL.md).
- Sanitization: No specific content filtering or sanitization of tool outputs is mentioned.
Audit Metadata