The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data retrieved from external web pages (Substack dashboard), which may include user-controlled content such as post titles.
Ingestion points: Data is ingested via get_page_text and read_page MCP tools from the Substack stats and individual post pages.
Boundary markers: The workflow relies on parsing specific HTML structures (tables, headline numbers), though explicit LLM delimiters for the ingested text are not detailed.
Capability inventory: The skill has the ability to write files to the local corpus/stats/ directory, navigate the browser, and execute scripts via the javascript_tool.
Sanitization: External data is normalized into a structured WeekExport object based on a predefined schema before being archived.
[DYNAMIC_EXECUTION]: The skill documentation mentions the use of a javascript_tool as part of the browser automation workflow to parse the rendered dashboard and extract specific data points from the DOM.

fetch-substack-stats