The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates its workflow by executing shell scripts (scripts/export-gif.sh) and Node.js commands to automate browser actions, manage image processing, and handle file system operations.
[EXTERNAL_DOWNLOADS]: The installation and export scripts perform dynamic downloads from external sources, including unversioned Node.js packages (gifenc, sharp, playwright, jimp) from the npm registry and Chromium browser binaries via npx playwright install.
[INDIRECT_PROMPT_INJECTION]: The skill takes user-provided text prompts and interpolates them directly into HTML and CSS files for rendering. This creates a potential vulnerability where malicious input could influence the headless browser's behavior during the frame-capture process.
Ingestion points: User-provided prompt and style parameters processed in SKILL.md and references/ files.
Boundary markers: No sanitization or boundary markers are used when inserting user text into the HTML/CSS templates.
Capability inventory: The skill has capabilities for file system access (mkdir, writeFileSync), shell command execution (bash, node), and network communication (local server, external API calls).
Sanitization: None detected for interpolated user content.
[DATA_EXFILTRATION]: In the optional ai-generated mode, the skill is instructed to send a generated image (base frame) to an external third-party API (api.klingai.com) for processing.
[DYNAMIC_EXECUTION]: The skill instructions in SKILL.md utilize node -e to execute arbitrary JavaScript code at runtime for performing screenshots within the AI-generation workflow.

graphic-gif