The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided articles to determine illustration requirements, creating a surface for potential indirect instructions.\n
Ingestion points: Article text is read from files or pasted content as described in the workflow in SKILL.md.\n
Boundary markers: A multi-stage workflow (Analysis, Outline, Prompt Construction) separates initial content ingestion from final image generation.\n
Capability inventory: The skill uses write_file, terminal (for curl, cp, and mv), and image_generate.\n
Sanitization: Mandatory instructions are provided to scan for and strip secrets (API keys, tokens, credentials) from source content before processing.\n- [COMMAND_EXECUTION]: Shell commands are used via the terminal tool to manage local assets and download generated images. The skill specifies naming conventions like kebab-case for slugs and structured output directories to ensure these variables are populated safely by the agent during execution.\n- [SAFE]: The skill includes a proactive security principle to scan for and strip sensitive information before writing any data to disk, and its operations are strictly scoped to its intended functionality of asset management and image generation.

baoyu-article-illustrator