optimize-c-code
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests and processes user-provided C source code and build files for optimization. However, this is the core intended functionality and the skill includes guidance on using sanitizers like ASan and UBSan to ensure correctness.\n
- Ingestion points: Processes C source code (.c, .h) and Makefiles.\n
- Boundary markers: Not explicitly defined for user-input code blocks.\n
- Capability inventory: Utilizes system compilers and benchmarking tools (hyperfine, perf) to execute code.\n
- Sanitization: The instructions emphasize running code under AddressSanitizer and UndefinedBehaviorSanitizer to detect flaws.\n- [DYNAMIC_EXECUTION]: The workflow requires the agent to generate code changes, compile them, and execute the resulting binaries to verify performance gains.\n
- Evidence: SKILL.md and verify.sh contain commands to build and run C benchmarks using standard compilers and measurement tools.
Audit Metadata