Documentation Index
Fetch the complete documentation index at: https://mintlify.com/bastndev/f1/llms.txt
Use this file to discover all available pages before exploring further.
The Prompt Composer is an in-panel overlay that gives you a rich text editor for crafting prompts before they reach the active agent. Instead of typing directly into the terminal, you compose in the overlay where you have access to file mentions, image attachments, live spellcheck, skill chips, and automatic translation — then send everything to the CLI in one keystroke. The composer integrates tightly with the host via message passing: spellcheck runs on the host dictionary, translations are fetched through the host network layer, and file paths are resolved against the open workspace.
Opening and Closing
| Action | Shortcut |
|---|
| Open / close the Prompt Composer | Shift+F1 |
| Toggle the composer (alternate binding) | Ctrl+Space |
| Execute the prompt and send to the CLI | Ctrl+Enter (⌘+Enter on macOS) |
| Recall previous prompt from history | ↑ (ArrowUp in an empty textarea) |
Ctrl+Space (the toggle binding) opens the composer if it is closed and closes it if it is open — useful when you want a quick dismiss without reaching for Shift+F1 again.
File Mentions with @
Type @ anywhere in the composer to open a file picker scoped to the current workspace. The extension queries workspace.listFiles and presents a fuzzy-filtered list. Selecting a file inserts an inline chip such as [Text #1 path/to/file.ts] that the host resolves to the actual file content before the prompt reaches the CLI.
File mention chips are recognized and protected by the spellchecker (via F1_MARKERS) so they are never flagged as misspelled words. The host preparer (src/my-cli/core/attachments/host-preparer.ts) scans the prompt for [Image #N], [Code #N], and [Text #N] markers and substitutes resolved file paths or image data before forwarding the text to the terminal.
Image Paste
Paste an image directly from the clipboard into the composer with Ctrl+V / ⌘+V. The webview reads the clipboard via the host (clipboard.read message), extracts the image data URL, and inserts an [Image #N] marker inline. When you submit the prompt, the host saves the image to globalStorageUri/my-cli-images/ and replaces the marker with the full file path that the agent can read:
// Supported image formats resolved from MIME type
case 'image/jpeg': return '.jpg';
case 'image/png': return '.png';
case 'image/gif': return '.gif';
case 'image/webp': return '.webp';
case 'image/svg+xml': return '.svg';
Live Spellcheck
The composer checks spelling in real time as you type, highlighting suspect words with inline underlines. Spellcheck is powered by the fixnow package (loaded dynamically in the host to avoid ESM/CJS conflicts) and supports four languages:
| Language | Code | Spellcheck | Accent-strict toggle |
|---|
| English | en | ✅ | No |
| Spanish | es | ✅ | Yes — toggle missing tildes (lenient by default) |
| Portuguese | pt | ✅ | No |
| Russian | ru | ✅ | No |
zh) does not have a trie-based checker and always passes without marking. Spellcheck is best-effort: if the host dictionary fails to load, the prompt is never blocked — the composer silently returns an empty issues list.
The spellchecker is warmed (dictionary pre-loaded into memory) automatically when the CLI Hub panel initializes, so the first keystroke is not slowed by trie decoding. The default warm language is Spanish (es).
Language Source Picker and Auto-Translation
The composer includes a language picker that sets the source language of your prompt. Selecting any language other than English causes the text to be automatically translated to English before it is sent to the CLI. This means you can write prompts in your native language and the agent always receives a well-formed English instruction.
| Language | Translates to English | Spellcheck |
|---|
🇪🇸 Spanish (es) | ✅ | ✅ |
🇨🇳 Chinese (zh) | ✅ | ❌ |
🇧🇷 Portuguese (pt) | ✅ | ✅ |
🇷🇺 Russian (ru) | ✅ | ✅ |
🌐 English (en) | No translation needed | ✅ |
src/my-cli/core/translation/host-prompt-translator.ts, which uses two providers with automatic fallback:
- MyMemory — preferred for short prompts (up to ~900 characters); chunks text into 450-byte segments
- Google Translate (unofficial) — preferred for longer prompts (>900 characters); chunks text into 1 500-byte segments
If one provider is rate-limited, the other is tried automatically. A 2-minute cooldown prevents hammering a rate-limited provider. Results are cached in memory so re-sending the same prompt incurs no additional network round-trip.
Skill Chips
Skills from the My Skills panel can be attached to a prompt to prepend structured context — such as an AGENTS.md instruction file or a custom workflow description — before the prompt text reaches the agent. Select skills in the My Skills panel and they appear as chips in the composer. The host resolves each skill’s content via workspace.listSkills and prepends it when the prompt is prepared.
Prompt History
Press ↑ (ArrowUp) in an empty textarea to recall the most recently sent prompt. Pressing ↑ again steps further back through history (up to 20 entries per agent, persisted in localStorage). Press ↓ to walk forward again — past the newest entry the textarea clears. Any edit exits browsing mode, so the arrow keys go back to moving the caret normally.