skill-refiner-acrazie
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [SAFE]: The skill operates as an interactive feedback refiner with a strictly scoped operational envelope. It prevents unauthorized file system access by requiring user approval for all write paths and ensures integrity by verifying file digests before processing target skills.- [INDIRECT_PROMPT_INJECTION]: As a tool that observes and records the output of other skills, it possesses an inherent attack surface for indirect prompt injection. This risk is effectively mitigated by the skill's instructions to record only minimized summaries, redact sensitive information such as secrets or personal data, and require explicit user approval for all consolidated findings.- [COMMAND_EXECUTION]: The provided Python script (scripts/test_satellite_contract.py) is a static analysis tool for verifying instruction consistency across the vendor's skills. It performs local file reads on hardcoded paths within the project structure and does not execute arbitrary shell commands, external scripts, or remote code.
Audit Metadata