Fetching from the wire…
Infra2026-09-03 · source-backed
Released September 3, the fix keeps a failed iptables rule removal during steal-port teardown from taking down all clients rather than only the one tearing down, a blast-radius bug for teams using mirrord to give agents a pod-like Kubernetes environment. It also keeps remote file operations working after an agent reconnect, and makes user-data updates atomic so concurrent processes and interrupted writes can't corrupt stored data. NATS was added as a queue type in split_queues.
Each link below shares sources, entities, or timing with this story.
This is the most immediately useful thing in today's batch and it takes ten minutes to implement. paddo.dev measured the Go-rewrite compiler on real projects: a 9,140-line Astro blog went from 2.56s to 0.32s (8.0x), and a 159,320-line Next.js app went from 7.32s to 0.78s (9.8x...
The Apache-2.0 toolkit (9,345 stars) argues that while SWE-bench gives coding agents training signal, distributed production failures have no equivalent, so it ships "an open reinforcement learning environment for agentic infrastructure incident response, with end-to-end tests...
The CNCF Sandbox project shipped fixes on August 26 for command injection in the Kubernetes toolset plus slab, kubevela, inspektor_gadget and aks follow-ups, SSRF in fetch_webpage and connectivity_check, and hardening of kubectl-run to execute without a host shell (GitHub). Th...
Orchard Env is a Kubernetes environment service supplying reusable isolated components for data collection, RL rollouts, and evaluation without per-domain modification. The differentiating claim is harness-native training: a lightweight proxy records a real harness's own model...
mirrord lets a process on your machine, or inside an agent's environment, behave as if it were a pod in your Kubernetes cluster: real env vars, DNS, network, traffic. For agents writing and testing cloud services, it collapses the build-deploy-test loop so the agent validates...
Alibaba OpenSandbox (6.3K stars) — Isolated execution environments for AI agents. Docker, K8s, gVisor, Firecracker. The missing infrastructure for safely executing agent-generated code. GitHub Agency-Agents (6.1K stars, +1,469 today) — 55+ specialized agent personalities acros...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.