The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the wuji CLI tool to manage glove calibration. These commands are executed locally with user confirmation and structured monitoring, following the intended purpose of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill processes streaming JSONL output from the CLI to provide real-time guidance. It mitigates potential injection by validating the JSON schema and mapping tool-generated identifiers to a fixed set of instruction templates.
Ingestion points: Reads stdout from wuji calib hand-model --jsonl in SKILL.md.
Boundary markers: Uses JSON parsing to validate schema_version, operation, and calibration fields.
Capability inventory: Limited to wuji CLI operations and sending termination signals (Ctrl+C).
Sanitization: Maps step_name to specific textual instructions and sanitizes unknown pose IDs by replacing underscores.

wuji-cli-calib