The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a Makefile to orchestrate complex operations including Docker image management (make pull), GPU pipeline execution (make run-pipeline), and host-side utility installation (make ffmpeg-install). These commands are standard for managing the Physical AI Data Factory environment and are triggered by explicit user requests for operational tasks.
[INDIRECT_PROMPT_INJECTION]: The skill has a defined workflow for 'Context Understanding' and 'Distribution Analysis' that involves reading natural-language captions generated by Vision-Language Models (VLMs) from processed data files (metas/v0/*.json). This data is then used to synthesize new prompts and configurations.
Ingestion points: Ingestion occurs in references/context-understanding.md and references/distribution-analysis.md where the agent reads the output of previous pipeline stages.
Boundary markers: The skill does not employ specific delimiters for this ingested AI data, though it instructs the agent to perform 'synthesis' rather than direct execution of instructions found within the data.
Capability inventory: The skill possesses the capability to execute shell commands and modify local configuration files through the make utility.
Sanitization: There is no explicit sanitization or filtering mentioned for the ingested VLM captions before they are used for reasoning.

paidf-curation-and-retrieval