The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses several shell commands to manage dependencies, render assets, and publish styles. These include npx, npm install, and node for execution of a local rendering script. These operations are restricted to the local workspace and the vendor's official command-line interface.
[EXTERNAL_DOWNLOADS]: The skill instructs the environment to download necessary tools and libraries, specifically the Playwright browser automation suite (npx playwright install chromium) and the vendor's own CLI tools (npx gooseworks). These are well-known services or vendor-managed resources necessary for the skill's primary function.
[DYNAMIC_EXECUTION]: The workflow involves generating custom HTML and CSS at runtime based on the analysis of a reference image. This generated content is then executed within a headless browser (Playwright) to capture PNG screenshots. This is a core feature of the graphics pipeline and does not involve executing untrusted remote code.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect injection because it processes user-provided images which could theoretically contain malicious text instructions meant to influence the AI's analysis.
Ingestion points: The Read tool is used to process reference images provided via the --ref argument in SKILL.md.
Boundary markers: None identified in the prompt logic for image analysis.
Capability inventory: The skill can write files (gooseworks-style.json, PNGs) and perform network uploads via npx gooseworks styles publish.
Sanitization: There is no automated sanitization of text content extracted from images; however, the structured workflow and hardcoded "Standard Brief" for examples provide a significant guardrail against arbitrary instruction following.

goose-graphics-create-style