The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local helper script scripts/discover.mjs using the Bash tool to identify documentation pages. This is a standard architectural pattern for complex logic in skills and does not involve executing untrusted remote code.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to discover and fetch documentation pages. It attempts to locate standard files like llms.txt, llms-full.txt, and sitemap.xml from the origin of the user-provided URL. This behavior is the primary purpose of the skill and is documented transparently.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks common to all web-scraping tools, as it ingests untrusted content from external websites.
Ingestion points: Web content is ingested via the WebFetch tool as described in SKILL.md (Step 4).
Boundary markers: The skill includes instructions to the agent to return the full page content as Markdown without summarizing, which acts as a basic instruction boundary.
Capability inventory: The skill possesses the Write tool to save files to the local disk and the Bash tool to execute its discovery script.
Sanitization: No explicit sanitization of the scraped Markdown content is performed before writing to the local file system. However, the files are stored as data and are not executed.

docs-to-md