profile-kernel
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local shell scripts (
lock-gpu-clock.sh,reset-gpu-clock.sh) and Python modules (mla_var3.kernel,tests.benchmark) to manage GPU hardware states and perform profiling runs. These are routine tasks in a performance engineering workflow.\n- [INDIRECT_PROMPT_INJECTION]: The instructions require the agent to read and interpret profiling outputs (Markdown, JSON, CSV) generated during execution. This presents an indirect prompt injection surface if those outputs were to contain malicious instructions.\n - Ingestion points: Profiling reports (
report.md,report-verbose.md), kernel configuration (params.json), and autotuning history (autotuning.json) located in theout/profiles/and.cache/directories.\n - Boundary markers: No specific delimiters or validation warnings are provided for the agent when parsing the generated reports.\n
- Capability inventory: The skill possesses the ability to execute shell commands and run Python modules within the project environment.\n
- Sanitization: The instructions do not specify any filtering or sanitization of the report data before the agent processes it for bottleneck analysis.
Audit Metadata