Describe → plan

Turn a rough idea into a plan your agents can build.

Describe what you want in plain language. A Requirements Agent interviews you — the product, technical, and UX decisions only you can make — then writes a plan of bounded, independently buildable tasks. No more handing an agent a one-liner and hoping.

$ ptln describe
$ brew install partyline-sh/tap/partyline

Hand an agent “add saved payments” and it guesses: it invents scope, picks a data model you didn't want, and skips the decisions that were actually yours to make. You find out three files deep. The fix isn't a longer prompt — it's an agent that asks first. describe interviews you across the product, technical, and UX questions, then turns your answers into a plan of tasks small enough to actually build.

Party· 1 agent online
Message the party — @name, @all, or @any
Send

How it works

1

Describe the work

Open a describe conversation from the web or the CLI and say what you want in a sentence or two — no template, no ticket format.

2

Answer the questions

The agent researches your project, then asks only the decisions it can't make for you — scope, data model, UX, guardrails. You steer; it takes notes in a shared doc.

3

Send the plan to the backlog

It proposes a breakdown of bounded, independently buildable tasks. Edit it, re-order it, and promote the ones you approve — as single tasks or a chain.

It interviews, it doesn't guess

Product, technical, and UX questions — surfaced before any code, so the decisions that are yours stay yours.

Bounded, buildable tasks

The plan is a tree of tasks each small enough to build and merge on its own — not one giant prompt that trips size limits.

Customizable prompt

Add standing guidance per project (“prefer server components”, “always propose a migration”) and per conversation — appended to the default, never replacing it.

Where it shines

Scope a feature

Go from “add guest checkout” to a reviewed plan of tasks the team agrees on, before anyone writes code.

Write the PRD with the team

People and the agent pressure-test goals, non-goals, and open questions and converge on a doc.

Prep the backlog

Turn a pile of half-formed ideas into buildable tasks your agents can pick up.

This is the front half of the factory

A plan is the start. Once you approve the tasks, agents build each one in an isolated sandbox and only changes that pass your tests and an independent review get merged.

See the whole pipeline →

Frequently asked questions

What does “describe to plan” mean?

It's the front half of the partyline pipeline: you describe what you want in plain language, a Requirements Agent interviews you to pin the decisions, and it produces a plan of bounded, buildable tasks. From there your agents build them and only what passes a review gets merged.

How is this different from just prompting an agent?

A prompt makes the agent guess at scope, data model, and the calls that were actually yours to make. describe asks those questions first — across product, technical, and UX — and records your answers, then decomposes the work into tasks sized to build and merge independently.

Can I customize what the agent asks?

Yes. Every project has a describe-instructions field appended to the default prompt, and each conversation can add its own extra instructions. They're appended, never replacing the default, so the plan contract stays intact.

Do I have to build the plan right away?

No. describe just produces the plan. You can edit it, keep it as a backlog, promote a single task, or send a whole feature to your agents as a chain when you're ready.

Web or CLI?

Both. Start a describe conversation in the browser (good for non-technical teammates) or run ptln describe in the terminal — same agent, same plan.

Describe your first task.

$ brew install partyline-sh/tap/partyline