Skip to main content

Getting started with PRFactory

Connect an agent, cut an epic, watch the team build it, and approve at every checkpoint.

PRFactory runs the AI agents you already use through a process your team can see: an epic is cut into slices, a team of agents builds one slice at a time, and you approve at every checkpoint.

◈

1.Bring your agent

Add a repository, then connect Claude Code or Codex over MCP, or install the AgentTeamForge worker.

▦

2.Plan the epic

Decompose the epic into wave-ordered slices and answer the decision round once.

⚙

3.Watch the team build it

A lead agent runs the team on the current slice; the ticket shows every agent live.

▲

4.Approve at every checkpoint

Answers, plan, test plan, review and PR are yours to approve, refine or reject.

Your first run

What you will see

Three states, in the order you meet them.

1 · The team starts

A lead agent is spawned from the phase recipe and issues its child spawns. Nothing is asked of you yet.

Queued
review:backendqueued

2 · The team works

Each agent shows its state and a one-line tail of its output. Open the transcript, or message a specific agent.

Running
review:backend
› checking the authorization order…
6:02

3 · The team waits for you

At a checkpoint the agents park rather than guess. Approve, refine or reject. The run continues from your answer.

Waiting
review:frontendparked · awaiting reply
2:15
Quick start

Four steps to your first pull request

1 · Add a repository

Connect PRFactory to your Git repository (GitHub, Bitbucket, Azure DevOps).

Add Repository

2 · Connect your agent

Connect an agent under Settings → Connections → Agents (MCP) and add the server to your CLI agent, or install the AgentTeamForge worker.

Connect Agent

3 · Create and cut an epic

Write the epic, approve the PRD, then decompose it into slices and answer the decision round.

Create Ticket

4 · Start a slice

Start the buildable slice. The team runs; you approve at each checkpoint and merge the PR.

Where you are asked to decide

Checkpoint What the team has done What you do
Answers Clarifying questions raised from the codebase Answer or accept the defaults
Decision round Open decisions ranked by blocked slices Answer, accept the default, or defer
Plan A slice plan with an AC → test matrix Approve, refine or reject
Test plan Automated and manual coverage proposed Approve or refine
Code review A review round by the reviewer set in the agent settings Read the findings, approve or send back
Pull request Branch pushed, PR opened Review and merge (always a human)