The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses node:child_process.spawn to launch Google Chrome, Chromium, or Microsoft Edge. It also uses npx -y bun to execute the skill's TypeScript logic.
[EXTERNAL_DOWNLOADS]: The skill fetches external content from arbitrary URLs provided during execution to convert them to markdown.
[PROMPT_INJECTION]: The skill ingests untrusted data from the internet. Because the resulting markdown is processed by the agent, malicious instructions on a webpage could attempt an indirect prompt injection attack.
Ingestion points: Content is fetched from the URL provided to main.ts in the captureUrl function.
Boundary markers: The output format in scripts/html-to-markdown.ts uses YAML frontmatter but does not include explicit boundary markers or instructions for the agent to ignore embedded commands.
Capability inventory: The agent environment typically includes file system access and command execution capabilities.
Sanitization: The sanitizeHtml function in scripts/html-to-markdown.ts removes script and iframe tags, but the skill does not filter the resulting text content for natural language instructions.

baoyu-url-to-markdown