Task Builder

Last verified: 2026-05-06 · Drift risk: low

No API key required. Tell us what you want done and which AI app you already use. We'll route you to the right workflow, give you a prompt to paste, and show you exactly what to click.

The Task Builder itself runs entirely in your browser. It does not send your answers to an API, analytics service, telemetry service, or account connection. Optional Save Draft uses localStorage only. The surrounding documentation site may still load normal site assets such as fonts, search indexes, and GitHub Pages/MkDocs metadata.

Tell us about you and the task

Beginner Mode is on by default. Developer Mode is for people who want API keys, CLIs, MCP, and local scripts.

Beginner Power User Builder Developer

Save draft Load draft Reset Export JSON Import JSON

1. Which AI do you already use most?

ChatGPT Claude Gemini Grok Perplexity GitHub Copilot Something else I don't know

2. What is your comfort level?

Beginner — no code, no API keys Comfortable with apps Power user Developer

Comfort affects how much detail we show. Comfort = Developer auto-switches Mode to Developer.

3. What do you want to do?

Write Summarize Research Plan Learn Organize Prepare for meetings Manage email/calendar Build a website/app Automate a repeated task Analyze documents/data Work with code Build an agent Build a skill / custom assistant Maximize the AI subscription I already pay for Other

4. The task in your own words

What task do you want done? What final output should the AI produce? Frequency One-time Repeated manually Scheduled (later, after manual run) Event-triggered (later, after manual run)

5. (Developer Mode) Other surfaces you have access to

In Beginner Mode we use only your primary app. Switch to Developer Mode at the top to expose API/CLI/MCP/local-script surfaces.

Claude (claude.ai) Claude Projects Claude Desktop Claude Code (CLI) ChatGPT (chatgpt.com) ChatGPT Projects Custom GPTs Codex CLI OpenAI API / Agents SDK Gemini app / Gems Gemini CLI Google Antigravity Google AI Studio / Gemini API Grok / xAI API GitHub Copilot cloud agent Perplexity Local Python / shell scripts MCP server (build your own)

6. Inputs and outputs

What does the task touch?

Files on my machine A code repository Email Calendar A web browser External websites Local apps APIs A database Documents (PDFs, slides, sheets)

Where do inputs come from? Where should outputs go?

7. Sensitivity and authority

Is any data sensitive? No (public or my own non-sensitive notes) Confidential business data Financial Legal Clinical / PHI Other regulated data Should the AI act, or only draft? Draft only — humans take all final actions Take actions, but require human approval per action Take actions automatically within a strict allowlist Which actions always require human approval?

8. Quality and preferences

What does good output look like? (paste an example or describe it) What mistakes should the AI avoid? Tone, style, and preferences to remember What counts as success? What should happen when information is missing or uncertain? Ask me a clarifying question Flag the uncertainty in the output and continue Skip the uncertain part and continue

Recommended workflow

Generated locally from your answers. Updates as you type.

Pick your AI in section 1 to see a recommendation

Beginner Mode uses only your primary app and the safest shape that fits your task.

Prompt What to click Use this in… Learning ladder What good output looks like Practice exercise Free vs paid Maximize my subscription Level up If your plan doesn't have this System / project instructions Memory & preferences Tool allowlist Approval gates Eval checklist Red-team cases Repeat-run playbook Troubleshooting Expert expansion Full JSON

What the Task Builder gives you

For every task you describe, the Task Builder produces:

1. Recommendation — a one-paragraph plain-English suggestion, scoped to the AI app you already use.
2. Prompt — copy-paste-ready, with your constraints and preferences baked in.
3. What to click — exact step-by-step instructions for the surface you chose.
4. Level up this workflow — the next useful capability ladder for your app (saved prompt → Project → memory → custom assistant → skill → scheduled task → automation → agent).
5. If your plan doesn't have this — a fallback that works on a free or basic plan.
6. System / project instructions — the long-form persistent prompt for a Project / GPT / Gem.
7. Memory and preferences — a portable preference block you can reuse across tools.
8. Tool allowlist and approval gates — least-privilege guardrails.
9. Eval checklist and red-team probes — a starter set of tests.
10. Repeat-run playbook and troubleshooting — how to run this every time, and how to fix it when it breaks.
11. Exportable JSON — the whole thing, for backup or sharing.

Privacy

• The form runs in your browser only. There are no network calls.
• Save Draft uses localStorage only. Nothing else is persisted.
• Disconnect your network and the Task Builder still works the same.

Want a different surface?

Switch your AI in section 1, your comfort level, or Mode at the top — the output regenerates instantly.

Where to go next

• Mastery — beginner-to-expert tracks for each major AI.
• Surface router — the underlying decision matrix in long form.
• Recipes — 41 worked examples you can clone.
• Starter kits — copyable scaffolding (Developer Mode).