故障智巡 v0.2.0

AI 线上服务故障排查与处置平台

Productized Incident Response MVP

• Evidence-driven investigation using read-only Logs, Metrics, Trace and Git Diff fixture tools
• Root cause diagnosis with evidence references and escalation when evidence is insufficient or conflicting
• Human-controlled remediation through the live Approval Gate
• SAFE / ASK / BLOCK pre-execution policy; current remediation executor is simulated
• AgentEvent Trace and read-only Replay
• Evaluation from the 12-case fixture catalog and actual pytest JUnit results when available
• Chinese incident response console with responsive dashboard, incident detail, evidence, approval, trace and evaluation views

Scope and attribution

The current MVP uses simulated incident fixtures and simulated remediation. It is not connected to production telemetry or an operations approval system. Agent Runtime is built on/adapted from MiniClaw. The original MiniClaw copyright and MIT license are preserved; see README, NOTICE and docs/project-ownership.md for attribution and ownership boundaries.