Web Prompt 作者： Krzysztof Bogdan

Summarize an article. Pull a table out as CSV. Translate a page. Explain a stack trace.

Web Prompt runs your prompts against the page you are on, or against a single element you point at. You write the prompts; the extension aims them at whatever is in front of you and shows the answer in a panel beside the page.

Requires a free OpenRouter account and API key. You choose the model for each prompt from the several hundred OpenRouter carries, so you decide the cost and speed of every prompt you save.

Three buttons

• Full runs a prompt against the whole page.
• Pick lets you highlight and click one element, such as a table, a comment thread or a code block, and runs a prompt against just that. Esc cancels.
• Settings holds your API key and your prompt library.

What people use it for

• Compress: an article down to five bullets, a 400-comment thread down to the actual disagreement, a changelog down to what breaks.
• Extract to structure: an HTML table as CSV, a job posting as salary, stack and location, a spec sheet, a citation list.
• Translate: into another language, or out of legalese, medical and academic jargon into plain language.
• Explain: this code block, this stack trace on a bug tracker, what this table is actually saying.
• Enrich: expand acronyms, convert units and currencies, add the background the page assumes you already have.
• Rewrite: shorten, soften, formalise, de-buzzword, proofread.
• Critique: find the missing caveats in an argument, review a diff, ask what a page is not telling you.
• Classify: is this listing a scam, does this posting meet my criteria, is this worth reading at all.
• Convert: a page into markdown notes, a lesson into flashcards, a recipe into a shopping list.
• Draft: a reply to this thread, a cover letter from this posting.

Six prompts ship with it, namely Extract to markdown, Summarize, Table to CSV, Translate to English, Explain this and Plain language, so it does something useful the moment you add a key. You can change the model of any of them, disable them, reorder them, or duplicate one into a copy you are free to rewrite.

Prompt variables: $html$, $text$, $url$, $title$, $domain$, $date$ and $selection$. Every prompt also has optional prepend and append text that wraps the result instead of being sent to the model, which is useful for dropping output straight into your notes with a source link. Your prompts sync across your Firefox installs.

Privacy: the page content you act on goes to OpenRouter under your own API key, and nowhere else. There is no server of ours in the middle, no analytics and no telemetry. Your key and your prompts stay in Firefox sync storage.

Limits worth knowing before you install: one page at a time; single-turn, so there is no follow-up question and no chat; and it reads a page's text and HTML, not its images.