
william-lu-stack/flawless
Deployment & DevOpsThis page was translated automatically. Technical details may differ from the original. View the original
AI SRE AgenticOps for Kubernetes and cloud infrastructure.
Overview
CISRE (Cloud Infrastructure Site Reliability Engine) connects infrastructure risk discovery, evidence collection, root cause diagnosis, manual approval, controlled changes, recovery validation, and experience accumulation into an auditable closed loop. The model is responsible for understanding, planning, and interpreting; Skill is responsible for domain handling knowledge; Plugins provide composability; Harness is responsible for state, permissions, and orchestration; Controlled executors are responsible for real changes; Verifiers are responsible for proving that the target has been restored. Execution API returns successful, the model claims success, or the old instance remains healthy—none of these are fault recovery. Only when new evidence of the real target meets the recovery contract does the task close the loop. - : Perform forensics, diagnostics, approvals, execution, and recovery validation around real resources. - : Discover risks at regular times or manually, and reuse the same operations kernel as SRE Run. - : Display resource dependencies, traffic, and explosion radius to assist with risk access control. - : Maintain issue trigger conditions, progressive forensics, actions, rollbacks, and success criteria. - : View plugin details, call conditions, service dependencies, security boundaries, and recent calls; Supports visualization or model-assisted creation, and allows one-click download of runnable Provider + Skill + Test + Deployment projects. - : Runtime overview, resource events, plugins and profiles, Agent Trace, and operational performance. - : Only counts issues that have passed recovery validation; Can expand root causes, skills, changes, and recovery evidence. "Contract ready" means interfaces, permissions, auditing, and closed-loop semantics are in place, but does not mean a specific product is connected. Pages must not forge resources or health data. CISRE has absorbed the combined concept of the official DeepSeek Harness but retains independent production execution boundaries: - Everything is a Plugin: Newly added domain capabilities are delivered first without modifying core orchestration. - Service Provider / Consumer: Plugins use provide/requires to declare capabilities and dependencies, which are parsed by runtime. - Profile / Bundle / Patch: Development, testing, and production can combine different Providers without being decided in the business code by the heap environment. - Event-driven: Supports observe, serial, parallel, and waterfall; High-risk access control can only be tightened and cannot be bypassed by plugins. - Reversible lifecycle: Loading, unloading, hot reloading, and resource release have a definite order. - Event Origination: Session events are added and written, desensitized, and hashed in tandem, supporting replay, fork, resume, and audit tombstone removal.
Overview (machine translated)
CISRE (Cloud Infrastructure Site Reliability Engine) connects infrastructure risk discovery, evidence collection, root cause diagnosis, manual approval, controlled changes, recovery validation, and experience accumulation into an auditable closed loop.
Current version: 5.6.0.
In a nutshell
The model is responsible for understanding, planning, and interpreting; Skill is responsible for domain handling knowledge; Plugins provide composability; Harness is responsible for state, permissions, and orchestration; Controlled executors are responsible for real changes; Verifiers are responsible for proving that the target has been restored.
Execution API returns successful, the model claims success, or the old instance remains healthy—none of these are fault recovery. Only when new evidence of the real target meets the recovery contract does the task close the loop.
Product Entry
- SRE Run: Conduct forensic, diagnostic, approve, enforce, and restore validation around real resources.
- AI Inspection: Identify risks at a scheduled or manual time and reuse the same operations core as SRE Run.
- Topology Impact: Displays resource dependence, traffic, and explosion radius to assist risk access control.
- Skill Library: Maintains problem trigger conditions, progressive forensics, actions, rollbacks, and success criteria.
- Plugin Center: View plugin details, call conditions, service dependencies, security boundaries, and recent calls; Supports visualization or model-assisted creation, and allows one-click download of runnable Provider + Skill + Testing + Deployment projects.
- Platform Capabilities: Runtime overview, resource events, plugins and profiles, Agent Trace, and operational performance.
- Operations Effectiveness: Only counts issues that have passed recovery verification; Allows for root causes, skills, changes, and restoration evidence.
Current Capability Boundaries
| Domain | Current Status | Access Method |
|---|---|---|
| Kubernetes | Complete closed-loop | Rancher, upload/paste kubeconfig, and ServiceAccount within the cluster |
| Database | Extended Contract Ready | Realm plugin + read-only Provider + Typed action executor |
| VM / Host | Extended Contract Ready | Domain plugin + read-only Provider + Enterprise execution platform |
| Storage | Extended Contract Ready | Domain Plugin + Array/CSI/Storage Platform Provider |
| Middleware / Cloud Resources | Contract Readiness for Extension | Access by Stable Adapter and Harness Service Contract |
| Network | Extended Contract Ready | Switching/routing, load balancing, DNS, ACL/Security Policies, and Link Provider |
“Contract ready” means interfaces, permissions, auditing, and closed-loop semantics are in place, but does not mean a specific product is connected. Pages must not forge resources or health data.
Harness and plugin model
CISRE has absorbed the combined concept of the official DeepSeek Harness but retains independent production execution boundaries:
- Everything is a Plugin: Priority delivery of new domain capabilities to plugins, without modifying core orchestration.
This page was translated automatically. Technical details may differ from the original.
Recommended Tools
Try a different keyword or remove a filter.
Install
npx skillfish add william-lu-stack/flawless