crank-refine
Warn
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill utilizes a 'Probe' mechanism described as throwaway scripts generated in the OS temporary directory and executed at runtime to verify codebase facts, such as testing functions or performing database queries.
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided markdown artifacts and external research data to build an agenda that directs its subsequent actions and script generation.
- Ingestion points: Markdown artifact files and subagent returns from codebase exploration or web research.
- Boundary markers: No explicit markers are defined to isolate untrusted ingested data from the agent's logic.
- Capability inventory: The skill can write to the file system, dispatch subagents via the Agent tool, and execute generated scripts (probes).
- Sanitization: No sanitization or validation of the ingested content is specified before it influences the agent's behavior or code generation.
- [COMMAND_EXECUTION]: The skill executes shell commands like 'git rev-parse' to identify project structure and performs various command-line operations through its verification system.
Audit Metadata