refactor-live-truth
Pass
Audited by Gen Agent Trust Hub on Jul 6, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: No malicious injection patterns or safety bypass attempts were detected. The skill uses instructional language to enforce a verification protocol rather than to override system guardrails.
- [DATA_EXFILTRATION]: No data exfiltration patterns were found. While the skill accesses local project files and internal framework memory files (e.g., sessions.jsonl), it does not contain network operations or patterns associated with sending data to external domains.
- [REMOTE_CODE_EXECUTION]: No remote code execution or unauthorized package installation patterns were detected. The included Node.js script (scripts/checklist.mjs) is a benign utility for printing the skill's checklist in JSON format.
- [INDIRECT_PROMPT_INJECTION]: The skill explicitly manages the risks of processing untrusted data from documentation or comments by treating them as unverified 'claims'.
- Ingestion points: README.md, code comments, and .refactor-chain/memory/sessions.jsonl.
- Boundary markers: The skill instructs the agent to use a 'witness' (code/CLI output) to validate any 'claim' (prose documentation).
- Capability inventory: The skill utilizes standard project inspection tools such as grep, ls, and cat.
- Sanitization: The verification requirement serves as a functional validation of external data before it is acted upon.
Audit Metadata