The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted article content and user-supplied titles which are interpolated directly into the prompts/cover.md file used by image generation backends. This creates a surface where a malicious article could attempt to influence the image generator's output or bypass its safety filters through embedded instructions.
Ingestion points: Article source content (saved to source.md) and user-provided title/subtitle inputs interpolated into Step 3's prompt template.
Boundary markers: The prompt template uses structured headers (e.g., # Content Context), but does not include explicit instructions to the rendering model to ignore instructions embedded within the article text.
Capability inventory: The skill possesses the ability to write to the filesystem and execute shell commands to trigger image generation.
Sanitization: No explicit sanitization or filtering is described for the article content before it is placed into the generation prompt.
[DYNAMIC_EXECUTION]: The skill dynamically resolves the location of a wrapper script (codex-imagegen) at runtime by inspecting the $BAOYU_CODEX_IMAGEGEN_BIN environment variable or searching sibling directories (e.g., packages/baoyu-codex-imagegen/src/main.ts). The resolved path is then executed using bun or npx. While these are internal vendor tools, the use of runtime-computed paths for execution is a noted dynamic behavior.
[COMMAND_EXECUTION]: To perform its primary function, the skill executes external CLI tools and scripts including baoyu-image-gen, codex-cli, and bun. These tools are used to interface with local or remote image generation services. The instructions also suggest using npx -y bun as a fallback, which may involve fetching the bun package from the npm registry if not present locally.

baoyu-cover-image