compileiq-author-objective

Pass

Audited by Gen Agent Trust Hub on Oct 3, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill includes code templates in references/templates.md that use subprocess.run to call local compiler binaries such as nvcc and ptxas. This is necessary for the skill's intended purpose of optimizing kernel compilation.
  • [DYNAMIC_EXECUTION]: Template 5 in references/templates.md utilizes monkey-patching by overriding CompileCallable.__call__ to intercept library calls and inject specific compiler flags. This is a technical implementation designed to extend library functionality for optimization purposes.
  • [INDIRECT_PROMPT_INJECTION]: The skill defines an interaction surface where external configuration data enters the process via the config parameter in objective functions. While no specific boundary markers are used for this data, the ingestion is intended for compiler tuning, and the skill utilizes helper functions like save_compiler_config to handle the data format. The associated capabilities are restricted to local file operations and compiler execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 3, 2026, 05:25 PM
Security Audit — agent-trust-hub — compileiq-author-objective