Read-only performance diagnostics for DeepSeek Harness: session load/restore timing, spill-hit counts, compaction count and trigger, context-injection volume (AGENTS.md/skills/tool-schema token share), and LLM cache hit rate — surfaced via /fast, persisted as reconstructable session events with async sampling off the model path.
DeepSeek Harness 实时 AI 自审插件：任何报错/警告第一时间连同时间、错误码、会话、用户原话、AI 动作与工作区写入专用日志目录，并主动通知 AI、内置 read_error_logs 工具随时读取。Real-time AI self-audit for DeepSeek Harness — captures every error/warning with timestamp, error code, session, user prompt, AI action and workspace into a dedicated log folder; instantly notifies the agent and ships a read_error_logs tool.
CI failure, diagnosed before you open the logs — DeepSeek Harness plugin that watches GitHub Actions for new failures and turns raw logs into structured diagnosis cards · CI 失败，打开日志前就完成诊断 —— DSH 插件：监视 GitHub Actions 新失败，原始日志转结构化诊断卡，签名账本识别复发问题
Proactive follow-ups for DeepSeek Harness: the model schedules host-level alarms (once/every/cron with jitter, quiet hours, daily budget) that resume cold sessions, with tombstone compaction and a live web panel.
