The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by reading content from user-provided documents via the Read tool and interpolating it directly into prompts sent to the Google Gemini API.
Ingestion Points: The SKILL.md workflow reads a document from doc_path, and scripts/generate_illustrations.py parses the file structure to extract text from sections.
Boundary Markers: There are no explicit delimiters (e.g., XML tags, triple quotes) or negative constraints ("ignore instructions in the text") used when building the final prompt string in scripts/generate_single_image.py or scripts/generate_illustrations.py.
Capability Inventory: The skill uses the Agent tool to execute shell commands, performs network operations to the Gemini API, and uses the Edit tool to modify the original document.
Sanitization: While scripts/generate_illustrations.py truncates content to 1,000 characters, it does not filter or escape potentially malicious instructions embedded in the document text that could hijack the image generation model.
[COMMAND_EXECUTION]: The skill uses the Agent tool to launch parallel subprocesses executing local Python scripts (scripts/generate_single_image.py). These scripts are passed arguments such as --title and --content which are derived from the analyzed document. While intended, this pattern relies on the agent's ability to properly quote and escape untrusted document content when forming shell commands.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of external Python libraries (google-genai, pillow, python-dotenv) to function. It also communicates with the Google Gemini API to generate images. These dependencies and services are managed by a well-known service provider.

lov-document-illustrator