> ## Documentation Index
> Fetch the complete documentation index at: https://docs.docex.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent Playbook

> Universal setup prompts for coding agents and human developers. Copy-paste these prompts into Codex, Claude Code, Cursor, or OpenCode to scaffold Docex integration.

# Agent Playbook

> Use these as the universal setup prompt set for coding agents and human developers. The primary path is agent-first.

## Primary setup command

```bash theme={null}
docex setup --use-case "vision analysis for security scanning" --framework auto --top-up 5 --json
```

The `--top-up` value is a USD wallet recharge amount. The agent should show the approval URL to the human, wait for wallet recharge approval, then wire the returned `DOCEX_API_KEY` and `DOCEX_BASE_URL` into the server runtime.

## Per-agent prompts

### Codex

```
Setup docex in this app for receipt scanning with server-side uploads and normalized JSON output
```

### Claude Code

```
Install and wire Docex into this codebase. Run docex setup, wait for browser approval, add env config, upload handling, and a server-side vision analysis flow for invoice images.
```

### OpenCode / Cursor

```
Add Docex to this project for vision and OCR analysis. Detect the framework, run docex setup, add API key wiring, and scaffold a working analysis route.
```

## Human quickstart

1. Ask your coding agent to run `docex setup`
2. Approve GitHub auth and a prepaid wallet recharge in the browser
3. Let the agent add `DOCEX_API_KEY` and `DOCEX_BASE_URL`
4. Call `run()` with a file and prompt

## Expected agent behavior

1. Detect the package manager from lockfiles or project conventions
2. Detect the app shape (Next.js App Router, Express, or unsupported)
3. Install `docexdev` with the user's package manager
4. Run `docex setup --use-case "<workflow>" --framework <auto|next|node> --top-up 5 --json`
5. Show the browser approval URL to the human if setup is still pending
6. Add `DOCEX_API_KEY` and `DOCEX_BASE_URL` to server-side environment wiring
7. Create or update a thin shared Docex client wrapper
8. Scaffold one server-side analysis entrypoint that accepts a file and prompt
9. Tailor the prompt and output handling for the user's vision use case
10. Run a smoke analysis or explain the exact command to run once a fixture is provided
11. Explain the exact files changed and how to test the integration

## Constraints

* Prefer prompt-guided analysis by default
* Keep output normalized JSON unless the user explicitly asks for plain text
* Use server-side Docex calls only — do not expose `DOCEX_API_KEY` to browsers
* Treat the dashboard as approval, billing, usage, and debugging support, not the primary setup path
* Preserve existing project conventions where possible
* If the stack is unsupported, do not fake a full scaffold. Add the package, env guidance, and the smallest working integration point