cited-web-research
Cited web research
Searches and reads the web through browser and pplx in the hedwigai MCP's read-only shell, filing each original under its own URL so the citation and the copy match.
Before you start
This skill uses the hedwigai MCP's bash and create_workbook tools. If they aren't available, tell the user to connect hedwigai — in Claude, from https://claude.ai/directory/hedwigai; in Claude Code, /plugin marketplace add un-seen/hedwigai-skills then /plugin install hedwigai@hedwigai; anywhere else, add https://mcp.hedwigai.com/ as a remote MCP server.
browser may be switched off on the account. If the shell says so, ask the user before running programs enable browser: enabling a metered program means its calls are charged. programs list shows what each program bills. Run a study with no arguments, or <program> --help, before the first call; don't guess flags. Output is JSON: use jq in the same shell, and keep work in /tmp, the only writable place.
Studies
| Step | Study |
|---|---|
| Ranked results for a question (titles, URLs, dates; Google operators work) | browser search "<query>" or pplx search web "<query>" (--recency-filter week, --domains, --country) |
| The passages of pages that bear on the question, each page held first | pplx content snippets "<query>" <url> [url…] |
| The original document, held under its URL | browser asset <URL>, then read it with pdftext <URL>, doctext or xlsx |
| A page that needs clicks or a login | browser session open <URL>, session act …, then session save to make the page citable |