Documentation
Everything you need to get the most out of PromptForge — from installation to advanced settings.
Installation
PromptForge is not yet listed on the Chrome Web Store. To install the development version:
- Download the extension from the Chrome Web Store (coming soon)
- Open Chrome and go to
chrome://extensions - Enable Developer mode (toggle in the top right)
- Click Load unpacked and select the
pomptforge/folder - The ⚡ icon will appear in your toolbar
chrome://extensions, then hard-refresh the tab with Ctrl+Shift+R.How it works
PromptForge attaches a keyboard listener to the page in capture phase. When you press Enter in a supported chat input, it intercepts the event before the site processes it, opens the enhancement modal, and waits for you to build and confirm your prompt. When you click Enhance & Send, it injects the enhanced prompt into the input field and submits it.
Your original prompt is always shown in the modal so you never lose what you typed. All fields in every framework are optional — leave anything blank and it simply won't appear in the assembled output.
Supported sites
| Site | URL | Status |
|---|---|---|
| ChatGPT | chatgpt.com / chat.openai.com | ✅ Supported |
| Claude | claude.ai | ✅ Supported |
| Gemini | gemini.google.com | 🗓 V2 backlog |
| Copilot | copilot.microsoft.com | 🗓 V2 backlog |
RTF Framework Basic
Role / Task / Format — auto-triggered for short prompts below the word count threshold (default: 30 words).
| Field | Description | Pre-filled? |
|---|---|---|
| Role | Who should the AI be? PromptForge infers a role from domain keywords in your prompt. | If detected |
| Task | What needs doing? | ✅ Always |
| Format | How should the output look? e.g. bullet list, numbered steps, short paragraph. | If detected |
CREATE Framework Detailed
Character / Request / Examples / Adjustments / Types / Evaluation — based on the PMI Prompt Engineering Workbook (2024). Auto-triggered for long or complex prompts.
All six sections are collapsible. Click any header to expand. Leave sections closed to omit them from the output.
| Field | Description | Pre-filled? |
|---|---|---|
| Character | Define the AI persona or role. | If detected |
| Request | The task or objective. | ✅ Always |
| Examples | Paste text or upload a .txt or .pdf file. | No |
| Adjustments | Constraints, requirements, compliance notes. | No |
| Types of Output | Format or structure of the expected output. | If detected |
| Evaluation & Steps | Success criteria and step breakdown. | No |
SMART Framework Goal
Specific / Measurable / Achievable / Relevant / Time-bound — based on PMI prompt formula documentation. Auto-triggered when deadline or goal keywords are detected in your prompt.
SMART works in guide mode — it assembles a structured prompt in the preview pane that you can edit before sending.
| Field | Description | Pre-filled? |
|---|---|---|
| Specific | The exact objective. | ✅ Always |
| Measurable | How success will be measured. | No |
| Achievable | Feasibility of the task. | No |
| Relevant | Alignment with broader goals. | No |
| Time-bound | Deadline or timeline. | If detected |
Auto-classifier
The classifier runs on every intercepted prompt and selects a framework based on three signals, checked in priority order:
- Pinned default — if you've set a default framework in settings (not Auto), it always wins
- SMART keywords — deadline, goal, timeline, OKR, KPI, "by end of quarter" etc. → SMART
- Word count — at or above threshold → CREATE
- Complexity keywords — detailed, comprehensive, evaluate, stakeholder etc. → CREATE
- Fallback — everything else → RTF
You can always override the classifier's choice using the framework toggle inside the modal.
Modal overview
The modal slides up from the bottom of the page when you press Enter. It contains:
- Header — logo, framework badge, Send as-is button, close button
- Framework toggle — switch between enabled frameworks manually
- Classifier reason — explains why this framework was chosen
- Your prompt — read-only display of what you typed
- Framework form — pre-filled fields, all optional
- Quality checklist — collapsible 8-point self-check
- Preview pane — assembled prompt, editable before sending
- Footer — Preview, AI Assist, Save, Download, Enhance & Send
Quality checklist
A collapsible section with 8 yes/no checkboxes based on PMI effective prompt characteristics. As you check items, a colour bar updates from red → amber → green.
- Is the prompt specific and clear?
- Does it provide enough context without irrelevant detail?
- Does it include a role or character?
- Is the tone appropriate for the desired output?
- Are examples included where helpful?
- Is the language free from jargon or ambiguity?
- Is the output format or structure specified?
- Have you included a reliability/validation check?
AI Assist
The AI Assist button in the modal footer assembles your current prompt and wraps it in a meta-prompt:
"Review the following prompt and suggest improvements for clarity, specificity, and structure. Return an improved version."
The full meta-prompt is copied to your clipboard. Paste it into ChatGPT or Claude to get AI feedback on your prompt. No API calls are made from the extension itself.
Preview & edit
Click Preview to assemble all filled fields into the final prompt. The preview pane is a live textarea — you can edit the assembled prompt directly before sending. What you see in the preview is exactly what gets sent.
Settings reference
Access full settings via the ⚙ icon in the toolbar popup, or right-click the extension icon → Options.
| Setting | Default | Description |
|---|---|---|
| Enable PromptForge | On | Global on/off switch |
| Show quality checklist | On | Show/hide the 8-point checklist in the modal |
| Word count threshold | 30 words | Prompts at or above this count trigger CREATE instead of RTF |
| Minimum words to intercept | 4 words | Prompts below this count pass through without the modal — prevents interception of short follow-up replies |
| Default framework | Auto | Auto lets the classifier decide. Pin to RTF, CREATE, or SMART to always use that framework. |
| RTF enabled | On | Enable or disable the RTF framework |
| CREATE enabled | On | Enable or disable the CREATE framework |
| SMART enabled | On | Enable or disable the SMART framework |
| GuardianAI aware | Off | Shows a notice in the modal when GuardianAI is active. GuardianAI scans the enhanced prompt after Enhance & Send. |
Keyboard shortcuts
| Shortcut | Action |
|---|---|
| Enter | Intercept and open the PromptForge modal |
| Ctrl+Enter | Send without PromptForge (bypass) |
| Shift+Enter | New line (standard behaviour, not intercepted) |
| Alt+Shift+P | Disable PromptForge instantly. Re-enable from the popup. |
To change the Alt+Shift+P shortcut, go to chrome://extensions/shortcuts and find PromptForge.
Prompt history
Every prompt enhanced via Enhance & Send is automatically saved. Access history from the toolbar popup → Saved Prompts, or via the Options page → Saved Prompts tab.
Per-entry actions: Copy the enhanced prompt, Download as a .md file, or Delete.
Bulk actions: Export all as JSON (for backup), Clear all. History is capped at 500 entries.
Privacy & data
PromptForge is designed to be fully private:
- All data is stored in
chrome.storage.local— sandboxed to this extension on this machine only - No external API calls from the extension
- No telemetry, analytics, or usage tracking
- No accounts, no sign-in, no cloud sync
- The only network activity on supported sites is your own conversation with ChatGPT or Claude
Permissions used: storage, activeTab, scripting — the minimum required to function.