The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external websites to derive brand kits.
Ingestion points: The references/capture-workflow.md file (Step 2) describes a process for walking and scraping the HTML and CSS of target websites.
Boundary markers: The workflow attempts to isolate raw HTML by using subagents for analysis, but the resulting visual observations are returned to the main context.
Capability inventory: The skill includes scripts for headless browser rendering (scripts/render.py), file system management (scripts/brand_cache.py), and subprocess execution in its test suite.
Sanitization: scripts/kit_validation.py performs validation on URLs, CSS values, and SVG structures, including checks to prevent external entity declarations in SVGs.
[DYNAMIC_EXECUTION]: The rendering script (scripts/render.py) uses the Playwright library to launch a headless Chromium instance to execute Javascript and capture screenshots of HTML content.
[EXTERNAL_DOWNLOADS]: The capture workflow fetches assets from external domains.
Evidence: references/capture-workflow.md details the use of curl to download screenshots, CSS files, and web fonts.
Evidence: scripts/render.py navigates to external URLs using a headless browser.
[COMMAND_EXECUTION]: The skill uses shell commands for testing and internal workflows.
Evidence: tests/test_check_adherence.py invokes subprocess.run to call the Python interpreter for adherence testing.
Evidence: references/capture-workflow.md documents bash pipelines using curl, grep, and base64 for processing brand assets.

get-brand-components