Reads Markdown (.md), plain text (.txt), HTML (.html, .htm), Word (.docx) and PDF (.pdf), anywhere beneath the folder you point it at.
Reads nothing until you choose a folder. The chooser opens on your home directory, the one starting place no operating system asks you to approve. Point it at a Claude skills folder and the plugins tree beside it comes too, as a second root.
Opens a single document as well, from the second button. It reads that file and lists no directory around it, so whatever sits beside it stays unread.
Browse to any other folder instead, a notes folder or a project's documentation; the choice is remembered for next time.
Hidden and cache directories are passed over; so is a branch that leads to no document at all, so every branch you can open leads somewhere.
A folder it cannot open costs only itself rather than the whole tree.
A Word document is turned into HTML as it is read, so it reads like everything else here. HTML rather than Markdown because escaping into it is total: whatever the author typed stays text, where Markdown would read an indent as a block of code and a leading hyphen as a list. Word arranges pages with tables as often as it tabulates anything, so a table that is really layout gives up its contents as the paragraphs they are rather than being drawn as a table.
A PDF is read back into the document its page was laid out to be, with its pages marked. A PDF holds no headings, no paragraphs and no lists; it holds glyphs, each with a place, a size and a face. Ask it for the words alone and a CV arrives as a wall of monospace with every heading, every bold phrase and every bullet gone, so the sizes and the faces are read the way your eye reads them: larger than the body is a heading, bold is emphasis, a bullet opens an item, lines running on are one paragraph. That is a reading of the page rather than anything the file says, because the file does not say: no fonts, no rules and no pictures come across, while a form's grid becomes reading order, since a document read back as prose is prose. Where a PDF is a scan, a picture of a page rather than the words on it, no text can be taken and it says so instead of showing you nothing; a PDF that is password protected says that.
Opening a folder of them does not read them. Listing forty PDFs of twelve pages was measured at 7ms where extracting them all took 515ms; the text is fetched for the one document you pick.
Runs nothing a document carries. A page's scripts are dropped rather than executed, so a document somebody sent you cannot act on your machine; anything needing a script runtime is inert and shows as the plain text around it.
Fetches nothing on a document's behalf. A picture is read from this computer only: one named relatively is looked for in the document's own folder, so a picture stored beside the document is shown. One held at a web address is not fetched and not shown, nor is one on a network share; anything naming another machine is refused before it is read.
Follows only the links you would expect. A web or email link in a document is handed to your browser or mail program when you click it and a link to a place in the page moves you there. Any other kind of link does nothing, so a document cannot start some other program on your machine or swap the page for another file.