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

# How to generate an implementation prompt

> Turn a feedback request into an agent-ready Markdown brief with Anthropic Claude, then copy it for Cursor or another coding agent.

An **implementation prompt** is a detailed Markdown brief generated from a request so a coding agent (for example Cursor or Claude Code) can implement the change and open a pull request.

Project Echo uses **Anthropic Claude Sonnet** to write the prompt (default API model id: `claude-sonnet-4-5`). The model reads the request **title**, **description**, **public comments**, and **internal notes**, then returns a structured handoff document.

## Generate from the request detail page

1. Open the request in the staff app (**Requests** → select the request).
2. Click the **⋯** (three-dot) menu next to the vote control.
3. Choose **Generate implementation prompt** (or **Regenerate implementation prompt** if one already exists).
4. Wait for generation to finish. A banner at the top of the screen confirms success or failure.
5. Open the **Implementation** tab (it appears after the first successful generation, between **Score** and **Settings** when scoring is enabled).
6. Review the Markdown, then click **Copy** to paste it into your coding agent.

Each successful run is also recorded on the **Activity** tab as “generated an implementation prompt.”

## What goes into the prompt

| Included | Notes |
| - | - |
| Title and description | Rich description text is flattened to plain text for the model |
| Public comments | Chronological, with author labels |
| Internal notes | Staff-only context is included so the agent gets full requirements |

Regenerating **replaces** the stored prompt. Previous activity rows remain in the history.

## Tips

* Add clear internal notes before generating so Claude has staff context (acceptance criteria, constraints, out of scope).
* After copying, ask your coding agent to implement from the Markdown and open a PR in your own repository. Project Echo does **not** write to GitHub for you.
* If generation fails, check the banner message and contact [support@projectecho.io](mailto:support@projectecho.io) if it keeps failing.

## Automate with API or MCP

Agents can also generate or fetch prompts programmatically. See [Generate implementation prompts with the API and MCP](/api-mcp-and-embed/generate-implementation-prompts).

Related: [Working from a request detail page](/requests/request-detail-page).