Start with one sentence

What the handoff completes

1. Prepare — creates a safe project and installs its version-matched Codex and Claude Code workflows.
2. Create — turns the brief into readable MDX, speaker notes, a topic-specific visual system, and intentional motion.
3. Review — checks the source, builds the static presentation, and inspects the rendered states when browser tools are available.

The public prompt.md only starts the work. Once the project exists, its local instructions and skills take over so the agent always follows the contract for the installed Drever version.

When the brief is incomplete, the agent starts with one useful round: the topic first when it is missing, then common missing context such as the audience outcome and duration. Each round puts the concrete questions first and offers one escape at the end: Skip remaining questions — surprise me. Later rounds ask only topic-specific choices that would materially improve the result. Using the escape ends the interview immediately.

The work happens in ordinary files on your machine. Setup succeeds when the agent leaves a running local preview, a readable slides.mdx, and the requested web or PDF outputs ready for review.

Know what runs where

The website button copies text; it does not upload the brief or start a hosted generation job. Paste the prompt into the coding agent you already trust. That agent creates a local React and MDX project, installs packages from npm, and may use the network for research only when the brief and your agent permissions allow it.

Drever does not add telemetry to the generated project. Your chosen AI provider, any remote media, and the platform where you later publish the static build each keep their own privacy contract.

Give the agent a useful outcome

A single sentence is enough to begin. Include the audience, the change you want, and any required source material when you know them:

Create a short presentation for local residents about three proposals for a shared public space. Help them understand the trade-offs and decide which proposal to support. Use the planning brief and survey results as source material.

The agent can infer ordinary details and state its assumptions. Say Skip remaining questions — surprise me at any point to let it resolve every unanswered choice, or keep answering while the questions are useful.

npm create drever@latest my-slides
cd my-slides
npm run dev

npm run check
npm run build
npx playwright install chromium
npm run export