jetson-diagnostic
Pass
Audited by Gen Agent Trust Hub on Jun 22, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local shell scripts (
snapshot.sh,mem_summary.sh) that invoke NVIDIA-specific utilities liketegrastats,nvpmodel, andnvidia-smito gather hardware telemetry. It also utilizessudo -nfor non-interactive access to diagnostic information requiring elevated privileges. - [COMMAND_EXECUTION]: Diagnostic scripts read system-level status information from paths including
/proc/meminfo,/proc/device-tree/model, and/sys/kernel/debug/nvmap/, which are standard interfaces for monitoring hardware state on Jetson platforms. - [COMMAND_EXECUTION]: The
snapshot.shscript executes inlinepython3code to walk the/procfilesystem and extract memory usage data fromsmaps_rollup, providing a fallback mechanism for collecting process-level telemetry. - [PROMPT_INJECTION]: The skill creates an indirect prompt injection surface by processing output from various system tools. This is mitigated through the following evidence chain: (1) Ingestion points: data from command outputs and system status files; (2) Boundary markers: a well-defined JSON output contract provided in the documentation; (3) Capability inventory: local script execution and system-level file read access; (4) Sanitization: the
json_escapefunction ensures that all external content is safely encoded for JSON interpolation using Python or standard text processing tools.
Audit Metadata