external-gitcode-ascend-triton-operator-env-config

Pass

Audited by Gen Agent Trust Hub on May 19, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill downloads the Miniconda installer from official Anaconda repositories (repo.anaconda.com) and specific PyTorch wheels from download.pytorch.org.
  • [REMOTE_CODE_EXECUTION]: Setup instructions include downloading and executing shell scripts for Miniconda installation (bash Miniconda3-latest-Linux-*.sh).
  • [COMMAND_EXECUTION]: The skill executes several shell commands for environment configuration, including sourcing CANN setup scripts (set_env.sh), checking NPU status (npu-smi), and using pip for package installation.
  • [COMMAND_EXECUTION]: The troubleshooting section includes instructions to install system-level build dependencies (such as g++, zlib1g-dev, and clang-15) using sudo apt install.
  • [COMMAND_EXECUTION]: A local Python script (scripts/01-vector-add.py) is executed to verify that the Triton environment and NPU acceleration are functioning as expected.
  • [PROMPT_INJECTION]: The skill contains a surface for indirect prompt injection by instructing the agent to read and follow requirements from external documentation at triton-ascend.readthedocs.io.
  • Ingestion points: Official online documentation for version matching and installation guides.
  • Boundary markers: Absent; the agent is directed to read the entire page without specific delimiters.
  • Capability inventory: The skill has access to file system writes, package installation (pip, apt), and network downloads (wget).
  • Sanitization: None; the agent is expected to extract and apply version numbers and instructions directly from the documentation.
Audit Metadata
Risk Level
SAFE
Analyzed
May 19, 2026, 06:27 AM
Security Audit — agent-trust-hub — external-gitcode-ascend-triton-operator-env-config