The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes system commands to facilitate the development workflow.
Uses platform-specific commands (open, xdg-open, start) to launch the default web browser and display prototype viewers.
Relies on the GitHub CLI (gh) to perform repository authentication, clone external repositories, and create pull requests during the showcase publishing flow.
[EXTERNAL_DOWNLOADS]: The skill downloads content from external sources as part of its publishing feature.
Clones the stardust showcase repository from paolomoz/stardust-site on GitHub when the --publish-sample flag is used.
[DATA_EXFILTRATION]: The skill facilitates the transfer of project artifacts to external repositories.
The --publish-sample feature is designed to stage and push locally generated prototype HTML files and associated design metadata to a remote repository on GitHub.
[PROMPT_INJECTION]: The skill processes untrusted external data which constitutes an attack surface for Indirect Prompt Injection (Category 8).
Ingestion points: Reads captured website content from stardust/current/pages/<slug>.json and renders it into proposed redesigns. It also embeds live URLs in iframes with allow-scripts permissions for comparison.
Boundary markers: Implements structured provenance comment blocks (stardust:provenance) in generated HTML files to track input artifacts and direction.
Capability inventory: Performs file system writes (writing HTML and Markdown files), executes shell commands (gh, browser openers), and invokes the impeccable skill to perform complex creative tasks.
Sanitization: Employs a 'Content sourcing scan' and validation of token/attribute contracts, though it does not explicitly detail sanitization or escaping of executable content (e.g., <script> tags) present in the source website JSON data.

prototype