The developer's new job: review the factory's artifacts before you merge
Part of the Software Factory series, a pattern by Greg Alton. The factory does the spec work, the tests, the code, the validation, and the review. You are the last station before merge. Your job isn't to write the code; it's to check the evidence the factory left behind.
- Time
- About 15 minutes
- Checked
- Tool
- AGENTS.md format
If a button name on the screen does not match this guide, follow the action. These products rename buttons often. This is an advanced guide: it uses code, and each technical term is explained the first time it appears. Model names and prices change, so check the linked docs before you ship.
Steps
Start from the Handoff comment
Open the ticket's Handoff comment. It links everything: the PR, the review comment, the disposition report, the Plan, Test Results, and Execution Summary, the commits, and the changed files. If it's missing, the factory isn't done.
Check the spec trail
In the Plan, confirm every task maps to a requirement in the spec, and the Deviation section is "none" or explains a spec change you agree with. Code that does something the spec doesn't mention is a red flag.
Check the test evidence
In Test Results, each requirement should have a test, and each test should show a real RED before GREEN. Look for exact commands and counts (passed, failed, skipped). A skipped or deleted test needs a reason. Run the QA steps yourself on anything risky.
Check the review and disposition
On the PR, read the Code Review and Security Review comment, then the Review Disposition Report. Every finding should be Accepted (with a commit), Partial (with a follow-up ticket), or Rejected (with evidence). No finding should be left unaccounted for.
Read the diff with that context
Now read the diff. Because you know what the spec wanted and what the tests prove, you can focus on design, naming, and anything the tests can't catch. This is much faster than reviewing cold.
Merge, or send it back down the line
If every gate holds, merge. If not, leave a review comment saying which station failed (spec, tests, code, validation, or review) and let the factory run that station again.
Tips
Treat missing artifacts as a failed build. 'The agent said it's done' is not evidence.
Spot-check one RED → GREEN claim per PR by running the test on the base branch. It keeps the factory honest.
When you find a problem the factory missed, add a rule to AGENTS.md so it can't happen again. That's how the factory gets better.
Common questions
Doesn't this just move work to the developer?
It moves the developer from typing code to judging evidence. Reviewing a well-documented PR takes minutes, and you review every change with the spec and test proof in front of you.
Can I let the factory merge small changes?
The pattern says no. The merge gate is what keeps a human accountable for what ships.
What's the fastest way to adopt this?
Start with the AGENTS.md from part 2 and just the plan, code, and validate stations. Add the tracker and review stations once those run smoothly.
More guides
AdvancedClaudeBuilding
Call the Claude API in Python and cut costs with prompt caching
Get an API key, make your first Claude Messages API call in Python, then add prompt caching so repeated long prompts cost less and run faster.
About 30 minutes using Claude Console, Claude API docs
AdvancedClaudeBuilding
Build your first MCP server in Python and use it in Claude Desktop
Write a small Model Context Protocol server with the official Python SDK, expose a tool, and connect it to Claude Desktop so Claude can call it.
About 35 minutes using MCP docs, Claude Desktop
AdvancedGrokBuilding
Build a simple RAG pipeline with the xAI Grok API
Answer questions from your own documents: split them into chunks, retrieve the best matches, and have Grok answer with citations via the xAI API.
About 40 minutes using xAI API docs, xAI Console
Occasional notes
Hear when a new guide is up
A short note when a guide is added. No account, and no emails until this form is connected to a real list.