Launching today
QualityLayer
Plan, build and review software with your coding agents
47 followers
Plan, build and review software with your coding agents
47 followers
QualityLayer takes each code change through seven steps: Discuss, Research, Plan, Outline, Implement, Verify and Review. You comment on diagrams and clickable screens and approve the Plan before code changes. Your agent builds it test first, and a fresh agent that did not write the code checks the result against your request. It works with Claude Code, Codex, Cursor and four more on your own subscription, and runs on your computer. https://github.com/maxritter/qualitylayer
Free Options
Launch Team
Atlassian LoomTalk it through in Loom. Ship with AI
Promoted
Hi Product Hunt, I'm Max. I built QualityLayer because coding agents write code fast, yet they don't work the way an experienced engineer would. The plan sits in a chat only I can see, and the agent that wrote the code also grades it.
QualityLayer wraps the agent you already use (Claude Code, Codex, Cursor, Grok, Antigravity, Pi or OpenCode) in seven steps: Discuss, Research, Plan, Outline, Implement, Verify and Review. During planning the agent reads the code and draws the design as diagrams and clickable screens. I comment on them and approve the plan before any code exists. Then it builds in small slices with tests first. A fresh agent that did not write the code checks the result against what we agreed. At the end I read the change and its proof, and I do the merge myself.
It runs on your computer on macOS, Windows and Linux, with the agent subscription you already have.
I'm around today and happy to hear what you think of the Plan step.
GitHub: https://github.com/maxritter/qualitylayer
a fresh agent that didn't write the code doing the check is the right call, the same agent grading its own work alwyas passes lol
seven steps feels heavy for a one line fix tho. can you skip straight to Implement for tiny changes??? or does every change go through the Plan step???
@thamibenjelloun Agreed, that's why QualityLayer has a separate Verify step. The same agent grading its own work almost always passes.
A one line fix probably doesn't need all seven steps. You can use a QL Chat for freeform work, or plan in Claude Code and implement with QualityLayer, or plan with QualityLayer and implement in Claude Code. Use the steps you need and skip the rest.
Each of our projects keeps a short STATUS.md and a dated LOG.md in the repo, so the next agent session starts from there and not from an old chat. Does the approved plan stay in the repo as a file after the merge, so a later session can read why it was built that way?
Small changes, tests first, then a separate review. Sounds like a thoughtful workflow. Good luck with the launch, Max!
@marianna_tymchuk Thanks a lot Marianna! This product is based on 18 months of working with different agents and harnesses on a daily base with my professional clients, so there is really a lot of work and thought put into it. I hope this helps other people to create better software as well.