Agent Sandbox is an open-source Kubernetes SIG Apps project that provides a CRD-based, declarative API for managing isolated, stateful, singleton workloads with stable identities, aimed particularly at AI agent runtimes and development environments. It is runtime-agnostic and can use backends such as gVisor or Kata Containers when stronger isolation is required for untrusted or LLM-generated code. The supplied material describes its architecture and intended benefits, but does not independently establish reproducibility, production reliability, or broad adoption as shared Kubernetes infrastructure.
No intersection was found in the supplied Scott wiki or radar hits. Although Agent Sandbox is topically adjacent to agent harnesses and runtime isolation, the evidence does not show that it bears on a position or project Scott already holds or tracks.
queries asked of Scott's wikis
- coding-agent sandbox isolation architecture
- agent harness and runtime boundary
- reproducible stateful agent environments
- Kubernetes-native agent infrastructure
- untrusted code execution threat model
- sandbox provisioning latency and warm pools
2026-08-11T20:42:46Z
Repeated checks have produced only adjacent ecosystem examples, with no independent deployment, reliability evaluation, or adopter of Kubernetes Agent Sandbox itself. With no project-specific catalyst expected, the episode has faded and can be rediscovered if substantive implementation evidence appears.
2026-08-09T20:27:34Z
The staleness check adds no project-specific evidence; adjacent sandbox and Kubernetes-agent activity still has not produced an independent Agent Sandbox deployment, reliability evaluation, or adopter. Keep the case open but reduce review frequency until one of those facts appears.
2026-08-07T19:37:25Z
The latest movement is only minor engagement on previously known adjacent examples. No independent deployment, reliability evaluation, or adopter of Kubernetes Agent Sandbox has appeared, so the project-specific hypothesis remains unadvanced.
2026-08-05T15:27:16Z
Curie adds independent evidence that coding-agent deployment on Kubernetes is becoming a real infrastructure pattern, strengthening the ecosystem-demand premise enough to keep the project under active watch. It still provides no deployment, reliability, or adoption evidence for Kubernetes Agent Sandbox itself.
2026-08-05T15:22:01Z
evidence attached: hn.story.49183972 — Curie is independent evidence that teams are deploying coding agents onto Kubernetes, contextualizing demand for shared agent runtime infrastructure.
2026-08-04T16:29:38Z
The newly attached Isolade example remains adjacent demand evidence rather than an independent deployment or evaluation of Kubernetes Agent Sandbox. No new signal changes the project’s unproven reliability or adoption outlook.
2026-08-04T13:25:45Z
Isolade independently validates demand for microVM-isolated agent workbenches, but it neither deploys nor evaluates Kubernetes Agent Sandbox. The case still lacks evidence of this project’s reproducibility, production reliability, or adoption.
2026-08-04T13:21:45Z
evidence attached: hn.story.49168002 — An independent local-first coding-agent workbench demonstrates a concrete microVM isolation pattern relevant to adoption of shared agent sandbox infrastructure.
2026-08-02T16:30:20Z
No independent deployment, reliability evidence, or consequential adopter has appeared; the tiny engagement change is repetitive amplification and does not advance the adoption hypothesis.
2026-07-31T16:24:05Z
grounded: novel/low — No intersection was found in the supplied Scott wiki or radar hits. Although Agent Sandbox is topically adjacent to agent harnesses and runtime isolation, the e
2026-07-31T16:23:33Z
origin walked (codex/luna, conf 0.96): anchor hn.story.49125061 -> echo.github.f22e235b95 by Kubernetes SIG Apps (Kubernetes community)
2026-07-31T16:22:17Z
case created — A first-party Kubernetes SIG implementation creates a bounded adoption and reliability episode around isolated agent execution.