The Agent Skills Directory

[COMMAND_EXECUTION]: Spawns a local browser process to render web pages. The skill identifies and launches local browser executables such as Chrome or Edge using node:child_process.spawn in scripts/cdp.ts to perform its content capture function.
[DYNAMIC_EXECUTION]: Executes JavaScript within the browser environment to extract and clean content. It utilizes the Chrome DevTools Protocol Runtime.evaluate method to run extraction scripts inside the page context as seen in scripts/cdp.ts.
[INDIRECT_PROMPT_INJECTION]: Ingests and processes untrusted data from arbitrary third-party websites which could contain malicious instructions.
Ingestion points: Fetches content from user-provided URLs in scripts/main.ts.
Boundary markers: The output Markdown is generated without specific delimiters or instructions to ignore embedded commands.
Capability inventory: The skill can launch local processes and write files to the disk.
Sanitization: Filters scripts and styles using standard libraries like turndown and linkedom, but the text content is not escaped against prompt injection attacks.
[EXTERNAL_DOWNLOADS]: The skill instructions in SKILL.md use npx to execute scripts, which may involve downloading external runtimes or packages from the official NPM registry.

baoyu-url-to-markdown