The Agent Skills Directory

[DATA_EXPOSURE]: The skill reads from and writes to the local filesystem for configuration (e.g., EXTEND.md in .baoyu-skills/ or ~/.baoyu-skills/) and output storage. It contains proactive instructions to strip credentials, API keys, and secrets from any source material before including it in the output.
[COMMAND_EXECUTION]: The skill is designed to invoke external image generation tools (such as imagegen, image_generate, or baoyu-imagine) and user-interaction tools. This behavior is consistent with its stated purpose and is gated by a mandatory user confirmation policy.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content (text or files) and interpolates it into a structured prompt for an image generation backend.
Ingestion points: Untrusted data enters via source.md or direct text input from the user.
Boundary markers: The prompt template (references/base-prompt.md) uses variable interpolation ({{CONTENT}}) but does not implement specific boundary markers or 'ignore' instructions for the LLM within the template itself.
Capability inventory: The skill can write to the filesystem and invoke external image generation backends.
Sanitization: The workflow includes a specific step to redact secrets and credentials from the input data before processing, reducing the risk of accidental exposure.

baoyu-infographic