mirai
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill provides pedagogical examples and best practices for using the 'mirai' R framework. It does not contain hardcoded credentials, malicious scripts, or unauthorized network activities.
- [INDIRECT_PROMPT_INJECTION]: The skill enables an agent to construct and execute R expressions on background workers, creating a functional attack surface for processing untrusted code.
- Ingestion points: R expressions and data objects passed to functions like
mirai()andmirai_map()inSKILL.md. - Boundary markers: The documentation promotes the use of clean environments and explicit dependency passing (via
.args) to prevent unintended access to the host session's variables. - Capability inventory: The skill facilitates asynchronous R execution, local process management, and remote compute configuration via SSH and HPC cluster managers (Slurm, SGE, etc.).
- Sanitization: The skill provides instructions for correct R implementation but does not specify input validation or sanitization routines for the logic within the R expressions.
Audit Metadata