Centaur Secure Team Agent Runtime Model Card
这是可索引项目报告证据页：它保留 Centaur Secure Team Agent Runtime Model Card 的源材料入口、机制线索和限制提醒；正文仍需 reader/editor 与 academic public-copy review 后才能当作最终结论引用。
Centaur Secure Team Agent Runtime Model Card
| Field | Value |
|---|---|
| Repository | paradigmxyz/centaur |
| Category | Secure Team Agent Runtime |
| Stars / forks snapshot | 469 / 55 |
| Language | Python / TypeScript |
| License | View license / MIT signal from public site |
| Raw capture | raw-github/paradigmxyz_centaur.md |
| Updated by | hourly public metadata update, 2026-05-26 04:41 +0800 |
1. Role in Self Evolve
Centaur moves the self-evolution discussion from single-user local agents toward team runtime infrastructure: durable workflows, Slack-native collaboration, credential boundaries, harness-agnostic execution, and organization overlays for tools, skills, personas, prompts, and workflows.
2. Working Principle
Slack or API request -> durable control plane -> Kubernetes sandbox -> chosen agent harness -> approved tools / workflows -> replayable team-visible result
3. Evidence Path
Web-observed GitHub evidence showed 469 stars, 55 forks, 155 commits, Python/TypeScript stack, no published releases, Slack-native conversations, isolated Kubernetes execution, bring-your-own harness support, shared Python tool plugins, durable workflows, credential boundaries, replayable state, and organization overlays. Shell GitHub API freshness was not available in this run, so the metadata is web-observed rather than API-verified.
4. Teaching Use
Use this card to explain why mature self-evolving agents need governance infrastructure: durable state, sandboxing, credential boundaries, tool registries, workflow overlays, and audit trails can matter more than a clever prompt loop.
5. Limits
The repository was not cloned or deployed locally. No Kubernetes, Slack, credential-injection, tool, workflow, or harness path was executed in this iteration.