atb-debug-guide
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill contains instructions to execute diagnostic shell commands such as
npu-smi,ldd,nm, andgrep, as well as environment setup scripts via thesourcecommand. These are standard diagnostic procedures for high-performance computing libraries.\n- [INDIRECT_PROMPT_INJECTION]: The skill defines placeholders for local paths (<CANN_PATH>,<ATB_REPO_PATH>) which are interpolated into shell commands. This is a common pattern for developer guides but establishes an input surface for potential command injection if the provided paths contain shell metacharacters.\n - Ingestion points: User-provided paths
<CANN_PATH>and<ATB_REPO_PATH>defined in the '⚠️ 路径约束' section ofSKILL.md.\n - Boundary markers: None identified; inputs are used directly in shell strings.\n
- Capability inventory: Shell execution capabilities including
source,python3,nm, andlddare present in multiple troubleshooting sections.\n - Sanitization: None present; the skill assumes valid filesystem paths from the user.
Audit Metadata