The Agent Skills Directory

[PROMPT_INJECTION]: The instruction file prompts/system.md includes a directive to bypass safety filters: 'If content involves sensitive or copyrighted figures, create stylistically similar alternatives
DO NOT refuse to generate'. This is intended to prevent the image generator from refusing requests due to copyright constraints.
[INDIRECT_PROMPT_INJECTION]: The skill ingests article content provided by users and incorporates specific terms, metrics, and quotes into prompts for image generation (see references/workflow.md Step 5.1).
Ingestion points: Article files or pasted text content analyzed during Step 2.
Boundary markers: The workflow does not specify clear delimiters or warnings to prevent the agent from obeying instructions embedded within the article text when constructing prompts.
Capability inventory: Integration with image generation tools such as imagegen, baoyu-image-gen, and codex-cli.
Sanitization: There is no evidence of sanitization or filtering for the external content before it is used in the prompt generation process.
[DYNAMIC_EXECUTION]: In references/codex-imagegen.md, the skill describes a mechanism to locate and execute a vendor-provided wrapper script (packages/baoyu-codex-imagegen/src/main.ts) at runtime using a computed path based on the skill's installation directory.
[COMMAND_EXECUTION]: The skill utilizes shell commands to invoke the image generation backend and may execute scripts via the Bun runtime.
[EXTERNAL_DOWNLOADS]: As a fallback for missing local environments, the skill instructions in references/codex-imagegen.md suggest using npx -y bun to fetch the Bun runtime from the official npm registry, which is a well-known service.

baoyu-article-illustrator