Vale construir sua versão de PromptDC?

Build me a personal Chrome extension for rewriting selected text with an LLM.
Use Manifest V3, vanilla HTML, CSS, and JavaScript, with no framework, bundler, package manager, or build step.
Create manifest.json, background.js, content.js, content.css, options.html, options.js, and simple generated placeholder icons.
Request only storage, contextMenus, activeTab, and scripting permissions, plus host permissions for the chosen LLM API.
Add an options page that stores provider, API key, and model in chrome.storage.local, never chrome.storage.sync.
Support OpenAI and Anthropic through direct fetch calls from the background service worker.
Use the OpenAI Responses API or Chat Completions API and the Anthropic Messages API with correct headers and response parsing.
Never log the API key or selected text, and show clear errors for a missing key, HTTP failure, rate limit, or malformed response.
Provide exactly two modes: Writing and Coding, with the last choice saved locally.
Writing mode must offer selectable profiles: Improve, Grammar fix, Shorten, Email, Social post, and Tone, each with its own system prompt that preserves facts and meaning, matches the input language, and returns only the revised text.
Coding's system prompt must turn a vague request into an implementation-ready specification with goal, context, requirements, constraints, acceptance criteria, and likely files, while preserving user intent and returning only the enhanced prompt.
Create a context menu named Enhance selected text that works on selection and editable contexts.
Also inject a small accessible floating button beside any nonempty text selection after mouseup or keyboard selection.
The floating UI must include a compact mode selector with the Writing profiles and Coding, an Enhance button, a loading state, and a dismiss control.
Capture selections in input and textarea elements with selectionStart and selectionEnd, and capture contenteditable selections with a cloned DOM Range.
For input and textarea, replace only the captured range, preserve surrounding text and caret position, and dispatch bubbling input and change events.
For contenteditable, restore the saved Range, replace only its contents with a text node, dispatch a bubbling input event, and avoid innerHTML assignment.
If the selection is read-only, detached, stale, or cannot be safely replaced, copy the result with navigator.clipboard.writeText and show a small Copied notification.
If no selection exists in an editable field, enhance the field's full current value; otherwise show a helpful Select some text message.
Pass requests from the content script to the service worker with chrome.runtime.sendMessage so page scripts never receive the API key.
Keep the interface unobtrusive, keyboard accessible, responsive, and isolated with prefixed CSS class names and a high z-index.
Include an options-page Test API button and a concise README with Load unpacked instructions, provider setup, security caveats, and a manual test checklist.
Handle extension reloads and restricted chrome:// pages gracefully, and remove injected UI when the selection collapses or Escape is pressed.
Deliberately leave out accounts, sync, per-site profiles, billing, telemetry, analytics, history, and any backend server.
Return every file with finished code, not pseudocode or TODOs, so loading the folder as an unpacked extension works immediately.