Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-09-08 · high
The open-source tool from Semaphore's team screens issues before it drafts a pull request, betting selection matters more than code generation.
Why now: The project is early enough in beta, with roughly as many open pull requests as open issues, that the selection bet hasn't been tested at scale yet.
Semaphore's team open-sourced SuperPlane. It scans a backlog and picks out the issues it judges safe to attempt without a human watching. Then it drives the pick to a review-ready pull request.
Most autonomous-PR tools skip that judgment call and will generate a patch for any issue you point them at. Deciding which issues are safe to hand to an unattended agent is a harder problem than writing the patch. Get it wrong and you burn compute on a ticket that needed a design decision, not a diff.
SuperPlane calls itself "an open source factory for one-shot engineering." The repo sits at 6,166 stars, licensed Apache 2.0, and is still labeled beta.
It carries 322 open PRs against 284 open issues, a close split that suggests plenty of attempts don't close clean on the first pass. The repo doesn't spell out what counts as a "safe" issue. It's unclear whether that's a fixed rule set, a learned model, or a heuristic that improves with use.
Whether the screening step cuts down on bad picks, or just moves the failure from patch generation to issue selection, isn't answered yet. Watch the PR-to-issue ratio as the project moves past beta. It's an early read on how often the selection step misses, and whether that gap narrows with more use.
Each link below shares sources, entities, or timing with this story.
2,465 stars, Apache-2.0, targeting LLM and VLM post-training, describing itself as co-evolving with upstream slime, pushed within the hour. That 6.6:1 PR-to-issue inversion is the most PR-heavy backlog in today's sample, and a backlog that lopsided usually means an internal te...
Splitting the open-issues count via the GitHub search API gives a 1:2.8 ratio where most trending agent repos this week run PR-heavy. Still shipping daily, v1.18.27 on September 2, pushed within the hour. For anyone evaluating it as a base, the backlog shape says user-reported...
Nearly two contributions per complaint, the most PR-heavy backlog I've seen on this beat (GitHub). Set against ruvnet/ruflo, which splits 289 open PRs against 604 open issues at 70,203 stars, the two largest agent meta-harnesses have inverted contribution shapes at similar act...
The Rust sandbox runtime for autonomous agents is at 8,465 stars and cut v0.0.116 on August 28, still on a 0.0.x line five months after its February 24 creation (GitHub). A 1:2.8 PR-to-issue ratio reverses the PR-heavy pattern across every other trending agent repo this week....
PR #24448 adds Q2_0 to ggml for CPU (ARM NEON plus scalar fallback), completing the Q1_0/Q2_0/Q4_0/Q8_0 family, primarily to serve PrismML's Apache-2.0 Ternary Bonsai models. Format packs 2 bits per weight with one fp16 scale per 64 weights mapping {0,1,2,3} to {-1,0,+1,+2}·d....
When the meter's running hot, the obvious move is a cheaper model that's actually good. Mistral shipped one. Devstral 2 (123B, modified MIT) scores 72.2% on SWE-bench Verified. Devstral Small 2 (24B, Apache 2.0) hits 68.0%. Both carry 256K context. Mistral claims 7x cost effic...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.