Nothing leaves the page until //
Patkan takes no action until you deliberately end a sentence with two slashes.
For Anthropic Claude
Claude answers best when a prompt is split into labeled XML sections — role, context, task, constraints, output format. Writing that by hand every time is tedious. Patkan does it for you: type your rough thought on claude.ai, end it with //, and it becomes a structured Claude prompt in the blink of an eye.
Before → After
YOU TYPE
help me write a cold email to a founder about my analytics tool //
CLAUDE RECEIVES
<role>You are a senior B2B copywriter who writes concise, founder-to-founder cold email.</role> <context>The sender is reaching out about an analytics product. The recipient is a startup founder with limited time.</context> <task>Write a cold outreach email introducing the analytics tool.</task> <constraints> - Under 120 words - No buzzwords or hype - One clear call to action </constraints> <output_format>Subject line, then body. No sign-off placeholders.</output_format>
Open claude.ai and describe your task in plain, messy words. No structure needed.
Two slashes at the end of your sentence trigger Patkan. Nothing happens until then.
Your thought is replaced in place by a structured Claude prompt — role, context, task, constraints, and output format wrapped in the XML tags Claude follows best.
Patkan takes no action until you deliberately end a sentence with two slashes.
Claude gets XML. ChatGPT gets markdown. Gemini gets sections. Each model receives the structure it respects.
Installed unpacked as plain JavaScript — open the folder and inspect every line before you load it.
Questions
Anthropic's own prompting guidance recommends separating a prompt into clearly labeled XML sections — role, context, task, constraints, output format. Claude parses these boundaries more reliably than a wall of prose, which means fewer misunderstandings and better first answers. Patkan applies that structure for you automatically.
No. Patkan is a browser extension that runs inside the Claude tab. You type, you finish with //, and your text is transformed right in the input box — no copy-pasting, no switching tabs.
It compiles your rough thought into a structured prompt: it picks an expert persona, adds the context and constraints Claude needs, and wraps each section in XML tags. You can review and edit the result before sending it.
Yes. The same extension works inside ChatGPT, Gemini, Microsoft Copilot, Perplexity, and more — with the output format adapted to each model. This page just focuses on the Claude XML dialect.
10 transforms a day in ghost mode. No sign-up to start.