Manual Maker: Screen Recording to User Guide

Turning a repetitive workflow into written documentation is tedious and slow. Manual Maker automates it. Start recording and use your screen as you normally would — the extension logs your clicks, inputs, and menu selections in order, and captures the screen at every step. When you stop, Solar Pro 3 reads the action log plus button and field names from the DOM, writes a draft, and optionally self-checks the result to produce a step-by-step user manual. 🎬 How it works ▶️ Start recording — one click on the page you are documenting 🖱️ Click and type as usual — pressed buttons and filled fields are outlined in red in the capture. Actions in window.open() popup windows are recorded too ⏹️ Stop recording — the editor opens automatically ✍️ AI draft writing — Solar Pro 3 turns the action log, accessible names, and screen context into step titles and instructions 🤖 AI self-review (agent mode) — each draft step is compared against the original recording. Steps that are unsupported, vague, or inaccurate are rewritten with neighboring context (up to 2 rounds). The loop stops as soon as every step passes review 📤 Review and export — Markdown, Word, PDF, ZIP, or Claude Skill ZIP ✨ Why it is more accurate · Capture before the click: you record the screen the user actually pressed — not the page after it changed · Precise click targeting: in dense or overlapping layouts, the element actually under the cursor is chosen — not a wrapping container — so the right button or link is documented · No dropped steps when the page moves: a click or Enter that submits a form or opens a new page is still recorded — the step is committed even as the page navigates away · Reliable input capture: text you type is captured even when you immediately press Enter or submit; inputs inside web components (Shadow DOM) are recorded too · DOM-backed context: accessible names and surrounding context are sent with each action so the subject of every step stays grounded · Not a one-shot prompt: a draft → evaluate → rewrite-weak-steps agent loop reduces hallucination and gaps · Your edits are protected: steps you fix by hand are never overwritten by AI regeneration · Unnamed clicks handled automatically: clicks whose target can't be identified are dropped from the manual, with the reason shown (and restorable) 📋 Key features · Automatic action logging: click, double-click, right-click, text input, dropdown selection, checkbox, keyboard shortcuts (page navigation and scroll are optional in settings) · Solar Pro 3 manual generation: action log → Chat Completion draft → self-review (rewrite) pipeline. Adjust reasoning effort and agent mode in settings · Per-step editing: edit titles and descriptions, hide or delete steps, regenerate individually · Add your own steps: insert a new step anywhere between existing ones and write its title and description yourself; upload an image and it is auto-fitted to the captured screenshots' size (no cropping) · Screenshot annotations: click any capture to edit annotations anytime — highlight with boxes, ellipses, filled shapes, and freehand pen · Four document export formats: Markdown, Word (.docx), PDF, ZIP (with images and video) · Claude Skill export: package your manual as an upload-ready Claude Skill — skill.md plus step screenshots in resources/. The editor includes an upload checklist and copy-ready sample prompts for Claude chat · Save and replay the original recording (webm) 🔒 Privacy · Fields that look like passwords or national IDs are masked as •••••••• in the log; the original value is never stored anywhere · Those fields can also be blurred in the screenshots · Recordings, images, and video are stored only in your browser (IndexedDB). When you generate a manual or export a Claude Skill, only the action log and manual text (not screenshots) are sent to the Upstage API 💼 Ideal for any task where you have to explain something on screen: internal work manuals, software how-to guides, service tutorials, and new-hire onboarding docs. Update note (v1.0.14) What's new: · No lost steps on navigation — clicks (and Enter/form submits) that immediately open another page are no longer dropped. The screen is captured before the click and the step is committed even as the page moves away, so fast clicking that jumps pages stays reliable · Sturdier input capture — fast type → Enter → submit no longer loses the typed value; the input is committed the instant you move on (submit, Enter, or clicking elsewhere). Inputs inside web components (Shadow DOM) are now captured too · More precise click targeting — in dense or overlapping UIs, the element actually under the cursor is chosen instead of a wrapping container, so the correct button or link is documented · Slideshow (presentation mode) — review steps full-screen in the editor for walkthroughs and training · Keyboard & autoplay — move with arrow/Page keys, toggle autoplay with Space, and see a top progress bar · Dot & side-click navigation — jump to any step like a PowerPoint deck with side hit zones and bottom dots