The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses several shell commands to manage local development and assets.
It uses ssh -p 2222 -o ExitOnForwardFailure=yes -R $CRAWL_ID:80:localhost:$SERVER_PORT tunnel.tasting.dev to create a reverse tunnel, exposing the local development server to the vendor's verification engine.
it utilizes curl -fsSL to download CSS, HTML, and image artifacts from URLs provided by the Brand API.
It suggests using python3 -m http.server 3000 --bind 127.0.0.1 to serve static files locally.
It uses uuidgen to generate unique identifiers for crawl sessions.
[EXTERNAL_DOWNLOADS]: The skill fetches external assets (CSS, HTML, images) using curl. These downloads target URLs provided dynamically by the Brand API MCP tool during the extraction process.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from arbitrary, untrusted websites and incorporate it into a generated page.
Ingestion points: Content is pulled from external URLs via extract_website and get_website_result (including artifacts like CSS and HTML).
Boundary markers: None are specified; the agent is instead instructed to treat values as authoritative.
Capability inventory: The agent can write files, execute subprocesses (curl, ssh, python), and generate web content.
Sanitization: The instructions explicitly demand "verbatim" copying of SVG code and CSS rules (e.g., "Paste those rules into your verbatim", "Take the logo from there and never redraw one"). This lack of sanitization could allow malicious content (like XSS payloads in SVGs or CSS expressions) from a reference site to be injected into the agent's output.

brand-adherence