The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with the local system using shell commands, including qianwen skills search for discovering capabilities and qianwen skills install for adding new skills. It also performs a preflight version check using qianwen --version to ensure compatibility.\n- [EXTERNAL_DOWNLOADS]: The skill facilitates the installation or update of the @qianwenai/qianwen-cli package from the NPM registry when the tool is missing or outdated. This is an official vendor package required for the skill to operate.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external descriptions of published skills. It mitigates potential injection risks by instructing the agent to treat these descriptions as metadata only, while enforcing length and character restrictions on all search queries to prevent malicious interpolation into shell commands.\n
Ingestion points: Metadata and descriptions returned by the qianwen skills search command.\n
Boundary markers: Clear instructions in SKILL.md to disregard instructions within result descriptions.\n
Capability inventory: The skill has the ability to write to the filesystem via the installation command and access the network via the CLI.\n
Sanitization: Input validation restricts queries to 80 characters and a specific set of safe characters (alphanumeric, spaces, and select symbols).

qianwen-find-skills