nc://markdown-new

v0.1.0Pinned
ThomSharktools/nodecode-markdown-new: Reads, searches, crawls and converts public web pages to Markdown through the Markdown.new service. - a rewrite of nosliwhtes/hermes-markdown-new (MIT), credited to nosliwhtes7ee4d20· 9 Nov1 commit
test1 filecell.lisp6.9 KBtools.lisp16.6 KBwire.lisp20.7 KB

nodecode-markdown-new

Markdown New: reads, searches, crawls and converts public web pages to Markdown, as one tool the model calls. A Nodecode rewrite of nosliwhtes/hermes-markdown-new by nosliwhtes.

Get it

Tell the agent "add the markdown-new cell", or run nodecode add nc://markdown-new.

Set it up

Nothing. The Markdown.new service is public and keyless, so the cell works as installed; a page, a query or a site is sent to it and its Markdown comes back.

The section is optional, and every member has a default:

Member Default What it bounds
base_url https://markdown.new the service address; another instance if you run one
timeout_seconds 60 one request
poll_interval_seconds 2 how often crawl_run asks a job for its status
poll_timeout_seconds 300 how long crawl_run waits before handing the job id back
max_chars 20000 how much of one document an answer carries
max_upload_bytes 10485760 the largest local file an upload may send

/markdown-new with no argument prints what the cell is connected to and what it can be asked, and (help :markdown-new) is the model's guide.

Use it

Ask for what you want; the model picks the operation:

Operation Use
url_convert a page you already have the address of, as Markdown
search_extract search the web, with the extracted content of the results and their urls
crawl_start begin a bounded crawl of a site; answers a job id
crawl_status how far a job got, and how many pages it read
crawl_download a finished job's Markdown
crawl_run start, wait and download in one call
crawl_cancel ask the service to stop a job
file_convert a public PDF's address, or a local document, as Markdown

From the keyboard, /markdown-new <url> reads one page and shows it.

What it does not send

The service reads public pages only, so a URL that points inside your network (localhost, 192.168.…, a bare hostname) and a URL carrying a credential in a query parameter (?access_token=…) are refused before anything leaves. An upload must be a regular file that is not a secret store — no .env, no .ssh, .gnupg or .aws path, no .pem, .key or wallet file — and no larger than max_upload_bytes.

What comes back is untrusted text: the tool's own description says so, and a page's instructions are the page's, not yours.

Test

From the cells repository root:

.github/test-cell.sh tools/nodecode-markdown-new

Every request is a stubbed nlk:http, so nothing reaches Markdown.new and no network is needed.

MIT licensed, as the original is.

Add nc://markdown-new

  1. 1

    Install Nodecode

    Once per machine.

    $ curl -fsSL https://nodecode.ai/install.sh | sh
  2. 2

    Add the cell

    Pinned at 7ee4d20.

    $ nodecode add nc://markdown-new

    Or tell the agent add the markdown new cell.