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

# Understanding Your Claude Scope Generated Prompt Output

> Copy your agent-formatted prompt, configure output options, and save the session — the final step of the Claude Scope recording pipeline.

The **Output** page is the final step of the Claude Scope pipeline. By the time you arrive here, Claude Scope has analyzed your recorded frames with Vision AI, inspected your app's live accessibility tree with Playwright, and synthesized both sources into a single structured system prompt. Your job now is to review the prompt, choose the right format for your agent, toggle any output options, and save the session.

## How the prompt is built

Every prompt Claude Scope generates is assembled from two independent pipeline contributions, shown in the **Pipeline Contributions** section at the bottom of the page:

<CardGroup cols={2}>
  <Card title="Visual timeline" icon="eye">
    A Vision AI model analyzed each extracted frame from your recording. The result is a per-frame description of UI elements, interactions, and state changes — the visual narrative of what you recorded.
  </Card>

  <Card title="Structural inspection" icon="globe">
    Playwright visited the app URL you provided and captured a full ARIA accessibility snapshot of the live page. This gives the agent concrete element roles, names, and counts that the video alone cannot provide.
  </Card>
</CardGroup>

The prompt synthesis stage combines these two sources into a coherent, agent-ready system prompt.

## Choosing an agent format

The agent tabs at the top of the prompt code block reformat the same underlying prompt for each target agent. Click a tab to switch the view.

<Tabs>
  <Tab title="Claude Code">
    The **Claude Code** format structures the prompt as a `<system>` block with explicit role framing. It is optimized for Claude's instruction-following conventions and includes section headers that Claude Code parses well. Paste the full output directly into a Claude Code session or `CLAUDE.md` context file.
  </Tab>

  <Tab title="Codex">
    The **Codex** format strips Claude-specific XML framing and uses a flat Markdown structure. OpenAI Codex and the Responses API work best with concise, plain-text system prompts without nested XML tags.
  </Tab>

  <Tab title="Cursor">
    The **Cursor** format targets Cursor's `.cursorrules` convention. The prompt is written as a rule-set that Cursor can apply project-wide. It uses shorter imperative sentences and avoids verbose narrative sections.
  </Tab>

  <Tab title="Raw">
    **Raw** output shows the unformatted prompt exactly as Claude Scope synthesized it — no agent-specific wrapping, no reformatting. Use this if you want to inspect the full content or adapt it to a tool not listed here.
  </Tab>
</Tabs>

<Tip>
  When pasting into **Claude Code**, use the full Claude Code tab output and drop it into the `CLAUDE.md` file at your project root — Claude Code reads that file as persistent context for every session. When pasting into **Cursor**, use the Cursor tab output and paste it into `.cursorrules` so it applies automatically across the workspace.
</Tip>

## Output options

Three checkboxes below the prompt code block let you control what is included in the generated text. Toggling any option immediately updates the prompt.

| Option                        | Default | What it adds                                                                                                                                             |
| ----------------------------- | ------- | -------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Include screenshot bundle** | On      | Embeds Base64-encoded frame thumbnails directly in the prompt so the agent can reference visual context.                                                 |
| **Inline ARIA tree**          | On      | Includes the full Playwright ARIA accessibility tree in the prompt, giving the agent the complete structural snapshot of the live page.                  |
| **Include raw DOM diff**      | Off     | Appends the element diff summary between each pair of consecutive frames — useful when you want the agent to reason about exactly what changed and when. |

If prompt length is a concern, disabling **Include screenshot bundle** has the largest impact on size.

## Saving the session

Click **Save session** — available both in the top-right action area and in the bottom action row — to persist the session to your account. Claude Scope shows a confirmation toast and redirects you to the session detail page in the **Sessions** list, where you can revisit the prompt, frames, and inspection data at any time.

<Steps>
  <Step title="Select the agent tab">
    Click **Claude Code**, **Codex**, **Cursor**, or **Raw** depending on which tool you are working in.
  </Step>

  <Step title="Adjust output options">
    Check or uncheck **Include screenshot bundle**, **Inline ARIA tree**, and **Include raw DOM diff** to suit your needs. The prompt updates immediately.
  </Step>

  <Step title="Copy the prompt">
    Use the copy button on the code block to copy the prompt to your clipboard. Paste it into your AI coding agent of choice.
  </Step>

  <Step title="Save the session">
    Click **Save session** to save the recording, frames, inspection data, and prompt to your account. You can access saved sessions from the **Sessions** list in the sidebar.
  </Step>

  <Step title="Start a new recording (optional)">
    Click **New Recording** in the bottom action row to immediately begin another capture without returning to the dashboard.
  </Step>
</Steps>