Give an agent an objective. Steer it until it lands.

A task answers one request. A goal keeps the agent working toward an outcome across many steps, commits each checkpoint to a draft pull request, and takes your corrections while it runs.

A running goal on northwind/storefront-web: Move checkout to the Payments SDK on Claude Opus 5.5, 37 minutes in, with its objective and two checkpoint commits on the draft PR
A goal in flight: its objective, model and elapsed time, and the checkpoints it has committed to the draft PR.

Product path overview

Pick the right unit of work.

Three ways to start work, one isolation model and one pull-request review at the end.

New task on northwind/storefront-web: one clear instruction in the prompt

New Task

One clear instruction. ProPR files the issue, runs the agent and opens the PR.

A generated plan broken into five ordered tasks, each a future GitHub issue

New Plan

A change you want to shape first. Approve the issue breakdown before the agent writes code. Planning

Start a goal on northwind/storefront-web: the objective, an outcome that takes many steps

New Goal

An outcome that takes many steps and judgment along the way. The agent works toward it while you watch and correct.

Two launch strategies.

Choose how the agent turns the objective into commits.

The agent implements directly

The agent works on the goal branch and checkpoints at each stable point. ProPR validates, commits and pushes every checkpoint to the draft PR, with visual previews attached.

Two checkpoint commits at a 15-minute target cadence, with the latest declaration: three files included, one held back

The agent orchestrates through ProPR

The agent breaks the objective into GitHub issues, runs them through ProPR and assembles the result into one draft PR.

An orchestrating goal's own breakdown: issues filed, one implemented, two monitored in parallel, the epic PR next

Stay in the loop while it runs.

The goal console shows what the agent is doing, its progress, logs and visual previews. When it needs a decision, it asks you.

A goal waiting on its operator: the agent asks whether offline photos upload over cellular, with suggested answers and Answer, Pause and Cancel
The agent asks before it guesses, with suggested answers.
Correct it mid-run. The next checkpoint carries it.A goal on Claude Opus 5.5 is about to swap the 3-D Secure modal for a hosted page. One correction stops it, and the next checkpoint on the draft PR keeps the modal.
The goal timeline records the correction verbatim, followed by the agent acting on it

Correct it mid-run

Send an instruction or ask a question. The timeline records your words next to what the agent did with them.

A paused goal with Resume and Cancel, and its draft PR: the same session picks up where it stopped

Pause, resume, cancel

Stop at a checkpoint, pick up the same session later, or switch the model between steps. Every checkpoint already sits on the branch.

Goal options: at most three parallel tasks, the orchestrate strategy, and Ultrafix on the result

Set the boundaries

Cap how many tasks an orchestrating goal runs in parallel, and let Ultrafix polish the result.

Visual previews published on the goal's draft PR, shown in the goal console

Review it like any PR

A goal ends in a draft pull request. Review, comment and merge in GitHub.

Which agents run goals.

Goals run on the agents built for long sessions.

AgentGoal supportCorrections
Claude CodeNativeLive, while the agent works
CodexNativeLive, while the agent works
Antigravity CLIResumable conversationAt the next safe boundary
OpenCode, Mistral VibeTasks and plans onlyThrough PR follow-ups

Launch goals from anywhere.

Start and steer goals from the web UI, the desktop app, the propr goal CLI commands, or your chat client through MCP.