Sutura
Product Hunt Juan González Ponce
Sutura launched today: a GitHub Action and CLI that tries to prove a CI fix is real. It reproduces the failure, rejects fake green checks, and hands humans an evidence-backed PR.
Based on reporting by Product Hunt, Juan González Ponce — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Sutura is trying to solve a problem plenty of teams know too well: CI goes green, but the bug never really left. The product launched today on Product Hunt as a GitHub Action and a CLI, built to reproduce a failure in an isolated sandbox before it claims anything is fixed.
The workflow is blunt. Sutura separates flaky failures from real ones, then searches for bounded repairs. It also tries to block the usual tricks that make a test suite look healthier than it is, including deleted tests, weakened assertions, and relaxed config.
And it does not trust its own first answer. The proposed fix gets pushed through an adversarial audit by NVIDIA Nemotron, with GPT-6 Astra and TypeSafe Jev as veto-only second opinions. That setup is there to catch green-wash, not to celebrate it.
The end result is an evidence-backed pull request for a human to review. Sutura does not auto-merge, which is the right instinct for a tool built around proving a fix instead of just chasing a checkmark.
My take — AI-written commentary, not fact-checked reporting
This is the right kind of boring: less “AI fixed it” theater, more proof. CI already has enough fake confidence floating around without another tool helping people cheat the scoreboard. The refusal to auto-merge is the detail that makes it feel serious instead of flashy.
Read more about this at: Product Hunt