A Complete Guide to Input Methods: Voice, Files, and Prompt Templates
With the task list sorted out, this page answers "how to feed requirements and materials to it efficiently." Slow typing, messy materials, long chat records—WorkBuddy voice input and its siblings (audio-to-text, file dragging, chat record forwarding, prompt templates) exist for exactly these moments. In 10 minutes, turn the input box from a typing box into a "multimodal feeding port."
This is lesson 3 of "Daily Use," based on WorkBuddy v5.9.3, with operations following the official "New Task Bar" and "Create Task" documentation.
- Haven't gotten your first task working yet? Go back to 1.4 The First Task: From One Sentence to a Deliverable Result
- Want to see the full picture of this stage first? Go back to 2.0 Daily Use
What You'll Be Able to Do After This Lesson
- Dictate tasks by holding Mac left ⌥ / Windows ALT, with speech converted to text in real time and filled back into the input box
- Hand meeting recordings to audio-to-text, get a transcript with time ranges, and clearly explain its billing rules
- Feed file materials to it through three channels: dragging (up to 50 at a time), pasting screenshots, and the upload button
- Forward chat records from WeChat / WeCom directly as task attachments, add a sentence of explanation, and submit
- Get started quickly with prompt templates, use ↑ to recall the most recent 20 history entries, and have extra-long pastes collapse automatically so they don't blow up the input box
Your Current Struggles
- The requirement is crystal clear in your head, but typing it out takes five minutes—and it still misunderstands
- Two hours of recordings piled up after meetings, and the thought of manually transcribing and organizing them is a headache
- Eighty messages in the group chat before a plan was settled, ten screenshots taken, and pasting them one by one still keeps missing context
When to Use This Technique
When you need to:
- Speak out the requirements in your head quickly and turn them directly into executable tasks
- Hand over ready-made materials—recordings, spreadsheets, documents, screenshots, chat records—to it all at once
- Start with official templates or history entries, typing less and staying on track
And you don't want to:
- Slowly type out a long requirement description on the keyboard
- Copy and paste back and forth between WeChat and WorkBuddy
- Start from scratch every time wondering "how exactly should I phrase this"
🎒 Preparation Before Starting
- You have completed 2.0 Daily Use: Make WorkBuddy Your Office Companion, and gotten 1.4 The First Task working
- Your computer has a working microphone (for practicing voice input); have an audio file and an Excel or Word file on hand (for practicing file input)
- 10 minutes of reading time
💡 Will it cost credits?
Official rules: all conversations consume credits. Voice input, file dragging, and template filling are input aids, and the official documentation does not price them separately; audio-to-text is an explicitly paid capability, billed at about 0.6 credits per minute of actual audio duration, and a confirmation prompt appears before invoking it in free-model sessions. See 2.6 Credits and Pricing for the full rules.
Core Idea
Mental model: WorkBuddy's input box is not a typing box but a "multimodal feeding port." Whether it's dictated speech, dragged-in files, forwarded chat records, or official templates, everything first becomes an editable draft in the input box (text or attachment cards), and you decide when to send it.
| Input Method | How It Enters the Input Box | Can It Be Edited Before Sending |
|---|---|---|
| Keyboard text | Direct input | Yes |
| Voice input | Real-time transcription → model-adjusted text filled back in | Yes, the input box stays focused |
| Files / Screenshots | Attachment cards (file name, upload status, delete entry) | Can be deleted or supplemented |
| WeChat / WeCom chat records | Archive attachment | Can be deleted; if it fails, re-forward it |
| Prompt templates | Click to fill in the full prompt | Yes |
| Slash commands | Select from the completion list | Yes |
It is not the same thing as the web-based AI chat box you have used:
| Web-based AI chat box | WorkBuddy input box |
|---|---|
| Basically only typing and image upload | Voice, files, screenshots, chat history, templates, and commands are all supported |
| Uploading three to five files at once is a struggle | Drag and drop up to 50 files at a time |
| Pasting a long text blows up the input box | Texts over 3000 characters are automatically collapsed into items |
| Requirements sent yesterday have to be retyped today | ↑ One-click recall of the most recent 20 entries |
💡 When feeding materials, prioritize four types of information
The official recommendation is to clearly state the following when creating a task: what the goal is (what to generate / which file to process), what the input is (the data to analyze, reference documents), output format (Word, Excel, or PPT), and constraints (style, word count, deadline).
Follow along
Step 1: Use WorkBuddy voice input to dictate your task (Mac ⌥ / Windows ALT)
Why: When the requirement is in your head and your hands are slower than your mouth, speaking is fastest—WorkBuddy voice input converts your speech into text in real time and fills it into the input box, no typing needed.
Steps:
- Click the voice button in the input box to start recording; or use a shortcut—on Mac it is "click or long-press left ⌥ for voice input," and on Windows it is "click or long-press ALT for voice input" (hover over the voice button to see the shortcut hint)
- During recording: the input box displays the recognized text in real time, and the voice button and input box use animations to indicate that recording is in progress
- To end recording: click the voice button again, or release the long-pressed shortcut key
- After recording ends: the model adjusts the recognized text and fills the processed content into the input box; the input box stays focused, so you can directly edit or add to the text, or record another segment of speech
You should see: After you release the shortcut key, the organized text appears in the input box, the cursor stays in the input box, and you can edit it right away.
- I long-pressed ⌥ / ALT, spoke a task, and saw the text filled back into the input box
💡 Voice input ≠ audio-to-text transcription
Voice input means you speak into the microphone and it is converted to text in real time and filled into the input box (this step); audio-to-text means handing an existing recording file to it for transcription (next step, about 0.6 credits/minute). One speaks to the "input box," the other is handed to a "task" for processing.
Step 2: Hand a recording file to audio-to-text (about 0.6 credits/minute)
Why: Only remembering after the meeting that no minutes were taken, or having recorded a two-hour interview with no time to listen—just drop the recording file to it, and it can transcribe, summarize, and extract to-dos.
Steps: Upload a local audio file, or paste an audio link, then clearly state what you want it to do. The three official example tasks:
- Transcribe this recording into text.
- Summarize the core content of this meeting recording.
- Extract to-do items from the recording, and organize the owners and deadlines.
Know the rules once and for all:
| Item | Limit |
|---|---|
| Audio formats | wav, mp3, m4a, flv, mp4, wma, 3gp, amr, aac, ogg-opus (.opus or .ogg), flac |
| Single file | No more than 1 GB |
| Audio duration | No more than 5 hours |
| Supported languages | Chinese, English, Cantonese, and some dialects |
| Billing | Paid capability, about 0.6 credits/minute based on actual audio duration |
In a free-model session, a confirmation prompt appears before this capability is invoked, with three options: confirm, always allow for this session, and decline. If declined, the audio will not be uploaded or processed, and this transcription is canceled.
Transcription runs automatically in the background, and when finished WorkBuddy returns the result directly without you checking progress manually; what is returned is transcribed text with time ranges, making it easy to cross-check against the original or continue analysis. If the file is invalid, cannot be uploaded, or is in an unsupported format, it will point out the specific problem; transcription timeouts or failures return a failure explanation, and it will not pass off blank or incomplete content as a result.
You should see: After confirming the billing prompt, the conversation area subsequently returns a piece of transcribed text with time ranges.
- I uploaded an audio file and asked it to transcribe it, and got a result with time ranges
Step 3: Drag, paste, and upload: feed file materials to it
Why: Analyzing spreadsheets, editing documents, or looking at error screenshots all require giving it the file first. WorkBuddy offers three feeding channels—learn them all and pick the fastest one based on the materials at hand.
Method A: Drag and drop (recommended, fastest)
- Drag a local file to anywhere on the page—the home page or the conversation page both work; no need to drag precisely onto the input box
- A "release to add files" overlay appears on the page; after you release, the files are automatically added to the attachment area of the current input box
- Files are displayed as cards: file name, icon, upload status, and delete entry
- Up to 50 files at a time, with no limit on single file size; format rules are the same as uploading via the attachment button
- Non-compliant files show a specific error message at the top of the page (e.g., unsupported file type)
- Dragging out of the page or pressing ESC cancels, and the input box content is unaffected
Method B: Click the upload button
Click the upload button in the input box to select files; the format rules are exactly the same as drag and drop.
Method C: Paste a screenshot
Use Ctrl/Cmd + V to paste an image from the clipboard directly—error screenshots, design drafts, or parts of a spreadsheet, paste them in as you go.
Supported file formats (official statement): PDF, Word, TXT, Markdown, RTF, Excel, CSV, TSV, PPT, JPG, PNG, GIF, BMP, ZIP, RAR, and common code files.
⚠️ Adding ≠ sending
Whether dragged, uploaded, or pasted, attachments are not sent automatically after being added—you can continue typing text, add or remove attachments, and submit once your requirements and materials are all ready.
You should see: Drag an Excel file to anywhere in the window, and a "release to add files" overlay appears; after you release, a file card appears above the input box.
- I dragged a file into the page, saw a file card appear in the attachment area, and it was not sent automatically
💡 The fourth channel: @ references
Besides "stuffing" files in, you can also use @ references in the task description to cite files, documents, rules, and other information as context. To quote text, tables, or components from a model reply and follow up, see 2.1 Conversation Advanced: Multi-turn Adjustment, Rollback, and Referencing.
Step 4: Forward WeChat / WeCom chat history to WorkBuddy
Why: Requirements settled in a group chat, questions a customer raised on WeChat—no need to take screenshots one by one; just forward the entire chat history to it, and no context is lost.
Steps:
- In WeChat or WeCom, forward the chat history to WorkBuddy
- The WorkBuddy client will be invoked, and a chat history attachment is generated in the input box
- The attachment indicates its source (WeChat / WeCom chat history) and shows the file name, file size, or number of messages
- The chat history is delivered as a compressed archive; statuses such as "Parsing" or "Parsing failed" are shown on the attachment; if parsing fails, just delete the attachment and forward it again
- Whether WorkBuddy was not launched, already running, or running in the background, forwarding from the WeChat side can properly invoke it and be received
Just like files, an added attachment is not sent automatically—first add a line of explanation (for example, "Extract a requirements list from this chat history"), then submit.
You should see: Select the chat history in WeChat and forward it to WorkBuddy; WorkBuddy pops up automatically on the computer, and a chat history attachment appears in the input box.
- I forwarded a chat history from WeChat or WeCom, and the attachment appeared in the input box
Step 5: Extra-long pasting and ↑ recall: handling large blocks of text and repeated input
Why: Paste an entire report into the input box and it will be stretched to bursting; type the same requirement every day and, muscle memory aside, you still make mistakes. The official design offers two targeted solutions.
Extra-long pastes collapse automatically:
- When you paste text exceeding 3000 characters, the input box automatically collapses it into an expandable entry, keeping the input area clean
- The entry shows the opening snippet of the pasted content plus an ellipsis; pasting several blocks of long text in a row generates separate independent entries
- Hovering over an entry previews the original text and shows the hint "Click the capsule to expand the content for editing"
- After clicking, the original text fully expands into the input box for further editing, and the entry is removed automatically
- Pastes of no more than 3000 characters stay as-is, without triggering the collapse
Recalling historical input:
- When the input box is empty, press the ↑ arrow key to step back through recently sent content one by one (at most 20 entries are retained); press ↓ to go back in reverse
- Pressing ↓ on the newest entry clears the input box
- The recalled content can be sent directly, or edited before sending, reducing repeated input
You should see: Paste a 5000-word manuscript, and only a single entry with an ellipsis appears in the input box; after clearing the input box, press ↑ and the last message you sent is recalled.
- I pasted a block of extra-long text and saw it collapse into an entry; I used ↑ to recall a historical input
Step 6: Get started quickly with prompt templates and slash commands
Why: Not sure how to describe your requirement? The official side has prepared a set of prompt templates by scenario; repeated actions can be distilled into slash commands, invoked by typing /.
Prompt templates:
- The task creation page offers a set of prompt templates by scenario (daily office work, code development, design creativity)
- Hover over a template (or focus the template button with the keyboard), and the input box instantly shows a gray preview text; if the input box already contains text, the preview is appended to the end of the existing content, while the black original text remains unchanged
- The gray text is preview-only: it does not enter the draft, the undo history, or get sent; it is cleared automatically when the mouse moves away, focus is lost, or the scenario is switched
- Only by clicking the template is the full prompt actually filled into the input box, after which you can modify it as needed before sending
- Template content changes with the scenario and the selected "capsule": office and development scenarios show text-based tasks, while the design scenario shows image examples; with no capsule selected, recommended task prompts for the current scenario are shown, and after a capsule is selected, the corresponding prebuilt tasks are shown
Slash commands (project sessions):
- Type
/in the input box of a project session to see slash command completions for the current project (behavior consistent with CodeBuddy Code) - Project-level commands are defined in
.codebuddy/commands/in the project working directory and are available for completion as soon as the session is created, without sending a message first - Continue typing the command name to filter the list, which also shows command descriptions and parameter hints; nested directories are supported (e.g.,
team:review) - Project-level, user-level, and built-in commands are merged under a unified priority; after switching projects or creating a new session, the completion list contains only commands available in the current project
- When command files are added, modified, or deleted, the completion list refreshes automatically
You should see: Hover over a template and a gray preview appears in the input box; click and the content is officially filled in; type / in a project session and a command completion list pops up.
- I hovered to preview a template and clicked to fill it in; I typed / in a project session and saw the completion list
For detailed operations, refer to the official documentation: New Task Bar (Task Input and Guidance), Create Task.
Checkpoint ✅
- Can state the shortcut for voice input: long-press left ⌥ on Mac / long-press ALT on Windows; after recording ends, text is filled back automatically and the input box stays focused
- Have uploaded an audio file for speech-to-text, and know that billing is about 0.6 credits/minute, with a single file no larger than 1 GB and duration no longer than 5 hours
- Have added files by dragging: release anywhere on the page, at most 50 at a time, and they are not sent automatically after being added
- Have forwarded chat history from WeChat or WeCom and seen the parsing status on the attachment
- Have used ↑ to recall historical input, and know that pastes exceeding 3000 characters collapse automatically into entries
Pitfall Reminders
❌ Sending voice input without a quick check
Symptom: Press Enter right after dictation, and the sent task is full of wrong names, numbers, and proper nouns.
Cause: Voice input is not dictation—after recording ends, the model adjusts the recognized text before filling it back, but proper nouns and numbers may still be misrecognized; the official design deliberately keeps the input box focused precisely so you can make corrections.
Correct approach: After recording ends, glance over the recognition result first, make corrections, and then send.
❌ Audio file exceeds limits or format is unsupported, transcription fails
Symptom: Upload a 2 GB recording or an obscure-format file, and WorkBuddy reports a file problem, or the transcription returns a failure message.
Cause: Official limits—a single file must not exceed 1 GB, and audio duration must not exceed 5 hours; only wav, mp3, m4a, flv, mp4, wma, 3gp, amr, aac, ogg-opus (.opus or .ogg), and flac formats are supported.
The correct approach: check the file against the limitations table above; the official general advice for upload failures is to check the file format and network connection, and for large files, compress or split them before uploading again. A transcription timeout or failure returns a failure explanation rather than a blank result — just adjust according to the prompt and retry.
❌ Assuming the file must be dragged precisely onto the input box, and that it was "sent automatically" after dropping
Symptom one: carefully dragging the file to directly above the input box before daring to release. Symptom two: going off to do something else after dragging the file, only to come back and find the task never ran.
Reason: the official behavior is — dragging to anywhere on the page adds the file (a "Drop to add file" overlay appears); but after an attachment is added it is not sent automatically — you need to add a description and submit manually.
The correct approach: drag it anywhere in the window; after dropping, add a sentence about what you want it to do, then click send.
❌ Mistaking a template's gray preview for content that has been entered
Symptom: hovering to take a look at a template, deciding it fits, then switching away and sending — only to find the sent content contains no template at all.
Reason: the gray text is preview-only — it does not enter the draft, the undo history, or the message being sent; it clears automatically when the mouse moves away, focus is lost, or the scene is switched.
The correct approach: after reviewing the preview, click the template so its content is filled into the input box as regular text, then edit and send.
Advanced tips (optional)
🔧 Chat history + supplementary instructions combo. A forwarded chat history is not sent automatically — first add a line like "Extract the list of requirements from this chat history and organize it into a table" before submitting; this works far better than just forwarding a raw chat history.
🔧 Further processing of transcription results. Audio-to-text returns transcribed text with time ranges, which you can then ask it to summarize the core content and extract action items and owners — see A2 Meeting Minutes: Recording, Transcription, and Structured Minutes End to End for a full walkthrough.
🔧 Treat ↑ recall as "canned phrases." For the same type of task you send every day (e.g., daily report summaries), clear the input box, press ↑ to bring up the last one, change a couple of words, and send — saving you from retyping.
🔧 Build up slash commands for your team. Place command files in .codebuddy/commands/ within the project working directory, and everyone on the team can type / to complete them (nested directories such as team:review are supported), with behavior consistent with CodeBuddy Code.
Lesson summary
✅ Mental model: the input box is a "multimodal feeding port" — voice, files, screenshots, chat history, templates, and commands all go in through the same port, first becoming an editable draft, and you decide when to send.
✅ WorkBuddy Voice input: click the voice button or long-press the shortcut (left ⌥ on Mac / ALT on Windows); recognized text displays in real time during recording, and after you finish, the model adjusts and fills it back in while the input box stays focused so you can keep editing.
✅ Audio-to-text: upload an audio file or paste a link, about 0.6 credits/minute (free-model sessions show a confirmation prompt first); per file ≤ 1 GB, duration ≤ 5 hours, returning transcribed text with time ranges.
✅ Three file channels: drag and drop (anywhere on the page, up to 50 files at a time, no per-file size limit), the upload button, and Ctrl/Cmd+V to paste screenshots; none of them send automatically after adding.
✅ Direct chat-history input: forwarding from WeChat / WeCom invokes the client, archive attachments come with a parsing status, and they can be received whether the app is not started, started, or in the background.
✅ Text duo: pastes over 3000 characters automatically collapse into an item; press ↑ when the input box is empty to recall the most recent 20 history entries.
Input methods quick reference table:
| Input method | Key operation | Limits / billing |
|---|---|---|
| Voice input | Click the voice button, or long-press ⌥ / ALT | Real-time transcription with backfill, editable before sending |
| Audio-to-text | Upload an audio file or paste a link | About 0.6 credits/minute; ≤ 1 GB, ≤ 5 hours |
| Drag-and-drop files | Drag to anywhere on the page and release | Up to 50 files at a time; not sent automatically |
| Upload / paste screenshots | Upload button; Ctrl/Cmd+V | Format rules same as drag-and-drop |
| WeChat / WeCom chat history | Forward within WeChat/WeCom | Delivered as an archive, attachments carry a parsing status |
| Extra-long pastes | Paste directly | Over 3000 characters automatically collapses into an item |
| History input | Press ↑ when the input box is empty | Keeps up to 20 entries |
| Prompt templates | Hover to preview, click to fill in | Gray-text preview is not sent |
| Slash commands | Type / in a project session | Reads the project's .codebuddy/commands/ |
Next lesson preview
Now that you know how to feed in materials and requirements, the next lesson addresses "how it executes." In the next lesson you'll learn about 2.4 Three execution modes: Default (Agent), Plan, and Ask-only.
You will learn:
- What kinds of tasks each of the three modes — Default (Agent), Plan, and Ask-only — is suited for
- How Plan mode first produces a plan and waits for your confirmation before executing
- How Ask mode only answers without executing, and how to use it for consulting-type questions to save credits
- Two permission tiers: the difference between default permissions and full-access permissions