The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes a user-provided 'illustration brief' to generate Node.js code and HTML content. While this creates a vulnerability surface common to code-generation tools, the workflow is strictly defined to translate technical requirements into geometric primitives.
Ingestion point: The illustration brief variable in SKILL.md.
Boundary markers: Absent.
Capability inventory: Node.js process spawning, file writing, and browser script evaluation via Chrome DevTools.
Sanitization: Implicitly handled by the agent's logic for interpreting the brief.
[COMMAND_EXECUTION]: The skill instructions guide the agent to run Node.js build scripts and automate a local headless Chrome instance. These capabilities are used legitimately to generate figure geometry and perform visual regression tests such as checking line alignments and painter's order.
[OBFUSCATION]: The code utilizes Base64 encoding for packing complex geometry pair data and handling binary screenshot data from the browser. These are standard technical applications and were not used to hide malicious intent.
[NO_CODE]: The core 'kit' library is written using standard ESM and Node.js built-ins, with no external package dependencies, significantly reducing its attack surface.
[SAFE]: The skill follows security best practices for autonomous agents, such as working within isolated scratch directories to prevent unintentional modifications to the local worktree.

anatomy