amd-rocm-porting
Warn
Audited by Gen Agent Trust Hub on Feb 28, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill requires the execution of shell commands for system auditing and performance tuning. One specific command uses sudo to modify kernel NUMA balancing settings (sudo sh -c 'echo 0 > /proc/sys/kernel/numa_balancing'). Other commands include env, rocminfo, rocm-smi, hipcc, and grep.
- [EXTERNAL_DOWNLOADS]: The skill facilitates the installation of external Python packages via pip to setup the porting environment. It suggests installing individual packages with --no-deps and the local repository in editable mode.
- [PROMPT_INJECTION]: The skill demonstrates an attack surface for indirect prompt injection because it ingests untrusted repository content and possesses elevated execution capabilities. 1. Ingestion points: The skill reads source files (.cu, .py), dependency manifests (pyproject.toml, requirements.txt), and environment variables. 2. Boundary markers: No delimiters or ignore-instructions warnings are used when processing external code. 3. Capability inventory: The agent can execute sudo commands, install software, and modify environment configurations. 4. Sanitization: No sanitization or validation of the ingested codebase content is performed before it influences agent operations.
Audit Metadata