PRD Agent — Example Plane Agent

An example Plane Agent built as a Cloudflare Worker. When mentioned in a work item comment, it generates a PRD (Product Requirements Document) using OpenAI and links it back to the issue as a Plane page.

Use this project as a reference for building your own Plane agents.

How It Works

Plane Agents are AI-powered apps that users can @mention in work item comments. Each mention triggers an Agent Run — a tracked interaction session where the agent can think, act, and respond.

This agent follows a simple tool-chain workflow:

1. Receive webhook — Plane sends an agent_run_activity event when the agent is mentioned
2. Get issue details — Fetches the work item title and description via the Plane SDK
3. Generate PRD — Sends the issue context to OpenAI to produce a structured PRD
4. Create page — Publishes the PRD as a Plane page in the project
5. Link to issue — Attaches the page URL back to the original work item

The agent communicates progress back to Plane using Agent Run Activities:

Project Structure

src/
├── index.ts                  # Entry point — routes requests (OAuth, webhooks, health)
└── lib/
    ├── helpers.ts            # OAuth configuration and API client setup
    ├── store.ts              # KV storage for workspace credentials
    ├── templates.ts          # HTML template for OAuth success page
    ├── types.ts              # TypeScript types (webhook payloads, tool names)
    ├── webhook-handler.ts    # Webhook processing and token refresh
    └── agent/
        ├── agentClient.ts    # Core agent loop — orchestrates LLM calls and tool execution
        ├── prompt.ts         # System prompt defining agent behavior
        └── tools.ts          # Tool implementations (Plane API + OpenAI)

Key Concepts

• AgentClient (agentClient.ts) — Implements a ReAct-style loop. Each LLM response is prefixed with THINKING:, ACTION:, RESPONSE:, ELICITATION:, or ERROR: to determine the next step. The loop runs until a terminal activity (response/error/elicitation) or max iterations (10).

• Tools (tools.ts) — Stateless functions called in sequence. Results are stored on the AgentClient instance so each tool can access the output of previous tools without parameter passing.

• Webhook handler (webhook-handler.ts) — Validates events, refreshes expired OAuth tokens, and hands off to the agent client.

Setup

Prerequisites

• Node.js v18+
• Cloudflare account
• Plane workspace
• OpenAI API key

1. Clone and install

git clone <repo-url>
cd prd-agent
npm install

2. Create a KV namespace

wrangler kv namespace create CREDENTIALS_KV

Copy the namespace ID into wrangler.jsonc.

3. Create a Plane App

1. Go to your Plane workspace settings
2. Navigate to Apps > Create App
3. Configure:
   • App Name: PRD Agent
   • Redirect URL: https://<your-worker-url>/oauth/callback
   • Webhook URL: https://<your-worker-url>/webhook
   • Enable App Mentions (this makes it an agent)
   • Enable the agent_run_user_prompt webhook event
4. Note the Client ID and Client Secret

4. Configure environment

Update wrangler.jsonc with your values:

Set secrets:

wrangler secret put PLANE_CLIENT_SECRET
wrangler secret put OPENAI_API_KEY

5. Deploy

6. Install the app

1. Visit https://<your-worker-url>/oauth/setup
2. Authorize the app for your workspace
3. The agent is now ready to use

Usage

Mention the agent in any work item comment:

@PRDAgent create a PRD for this issue

The agent will generate a PRD, create a page, and link it to the issue.

Development

npm run dev          # Start local dev server (wrangler dev)
npm run deploy       # Deploy to Cloudflare
npm run cf-typegen   # Generate TypeScript types for Cloudflare bindings

Environment Variables

API Endpoints

Learn More

• Plane Agents documentation
• Plane Node SDK
• Cloudflare Workers

License

MIT