The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill prompts users for configuration variables such as BUILD_CMD and WINDOWS_IP, which are later interpolated into shell commands. While this is the intended functionality for build automation, it creates a surface where malicious configuration values could lead to command injection if the underlying agent does not sanitize inputs.
Ingestion points: User input requested in SKILL.md during the '初始化配置' (Initial Configuration) phase.
Boundary markers: Absent; the variables are used directly in shell strings.
Capability inventory: Access to Bash tool, Read, Write, and Edit. Ability to execute commands on remote devices via hdc.
Sanitization: None provided within the skill instructions.
Dynamic Execution (LOW): The skill dynamically generates shell commands by concatenating user-defined paths and commands (e.g., cd ${OH_ROOT}/ && ${BUILD_CMD} <TARGET_NAME>). This is consistent with the primary purpose of a build-and-test automation skill.

openharmony-ut