freeze
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external decision records, which serves as an ingestion point for untrusted data. While it lacks explicit boundary markers or sanitization, its capabilities are restricted to local file updates necessary for its core function.
- Ingestion points: Reading decision record files mentioned in the workflow.
- Boundary markers: Absent in the provided instructions.
- Capability inventory: Performs read/write operations on markdown files and an index file at ~/.claude/freezes/active.md.
- Sanitization: No input validation or escaping is specified before updating files.
- [SAFE]: The skill's primary operations involve managing local state through file I/O within a dedicated application directory (~/.claude/freezes/). No network access, credential handling, or dangerous command executions were detected.
Audit Metadata