The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the wuji CLI to perform hardware operations, including scanning for devices and reading/writing device parameters.
[EXTERNAL_DOWNLOADS]: The wuji update command enables the retrieval and installation of updated CLI binaries from the vendor.
[PERSISTENCE]: The wuji completions command generates and saves scripts to the user's shell completion directory, ensuring the tool's integration remains active across shell sessions.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external physical hardware, which is then processed by the agent.
Ingestion points: Sensor data and device configuration parameters are read via the wuji get and wuji sub commands in SKILL.md.
Boundary markers: The instructions favor using --json and --jsonl flags for structured data output, though they lack explicit safety delimiters for the LLM.
Capability inventory: The skill has the ability to write configuration data back to the hardware via the wuji set command in SKILL.md.
Sanitization: Data is retrieved in structured formats like JSON, providing inherent schema separation.

wuji-cli-base