Astro’s GitHub issue backlog is heading to zero for the first time in 5 years. Now Cloudflare is open-sourcing the tool that did it.
The New Stack Paul Sawers
Astro's GitHub backlog is about to hit zero for the first time in five years, thanks to an AI triage system Cloudflare just open-sourced. Agents now handle bug reproduction and diagnosis; humans still decide if fixes are good enough to merge.
Astro went from over 200 open GitHub issues in January to about 20 today, and the team expects to hit zero within a month. That would be a first in the framework's five-year history. The tool behind it, triagebot-action, runs as a GitHub Action and splits the work into four stages, each handled by its own agent: reproduce the bug, diagnose the cause, verify it's real, then attempt a fix. A state machine encoded in GitHub labels tracks where each issue sits, so the process stays visible rather than becoming a black box.
Fred Schott, Astro's co-founder and now a senior engineering manager at Cloudflare (which acquired Astro's team in January), says the fix stage is deliberately treated as the least essential part. Triaging an issue correctly and handing it to a human is, in his words, still a great outcome on its own. Getting agents to write mergeable code is a much higher bar, and Schott says that gap is why the team built triage first and let fixing arrive gradually, starting with zero fix capability, then a suggested fix on a branch, then eventually an agent-opened pull request.
The timing matters. AI coding tools have made it trivial to generate a bug report, a speculative vulnerability writeup, or an unreviewed patch in seconds, but verifying any of it still eats up the same human time it always did. Astro's approach flips that: agents do the grunt work of reproducing and diagnosing, and a maintainer only steps in once there's something concrete to evaluate. Schott traces the whole strategy to watching where agents were reliable, on triage, versus where they weren't, on code quality.
Cloudflare's acquisition of Astro raised the usual question about platform companies buying frontend frameworks: does the team get starved once its founders move on, the way Netlify's 2023 Gatsby purchase eventually played out. Six months in, Schott says Astro still runs as its own team with four full-time engineers, its own roadmap, and a release cadence, and shipped an Astro 6 beta with a redesigned dev server in January. Triagebot-action, and its underlying agent framework Flue, live under Astro's own organization rather than Cloudflare's, and neither depends on Cloudflare-specific infrastructure. Adoption elsewhere is early; Cloudflare's workers-sdk team is prototyping it, but Astro is the only production user so far.
This is the second AI-agent-focused open source release from Cloudflare in a week, following pvcli, a debugger built explicitly for agent use. The pattern is the same in both cases: take a task that assumed a human at a keyboard and repackage it as something an agent can run end to end. Schott is candid that full autonomy on the fix step isn't close, and that every automated fix stays subject to a maintainer's judgment call. But the direction is obvious enough, and Astro's backlog is the proof of concept.
My take
Handing agents the tedious verification work while keeping humans as the gatekeeper for actual code merges is the sane middle ground everyone should be building toward, not full autonomy hype. The real tell here is that Astro built triage first and fixing last, resisting the urge to let an agent open pull requests before it earned that trust. Cloudflare open-sourcing this instead of quietly using it as a competitive moat is the right call too, and a decent signal they're not planning to let Astro rot the way Gatsby did after Netlify bought it.
Read more about this at: The New Stack
Related stories
When it's the maintainer who's AI-pilled
TLDR · 1 month ago ·
10
Google says it fixed more Chrome bugs in June than over the past two years, thanks to AI
TechCrunch AI · 5 days ago ·
22
When vendor-supplied support matters: How AI is changing the open source security equation
The New Stack · 1 week ago ·
4