context-engineering 8

Gives DeepSeek Harness three automatic Jev judgments — prune oversized tool output, screen fetched pages for injected instructions, suggest which skill to use — plus the jev_ask and jev_gate tools, a persistent judgment ledger, and a /jev-status diagnosis

Read-only performance diagnostics for DeepSeek Harness: session load (open/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 the /fas

Lazy MCP gateway for DeepSeek Harness: one stable proxy tool instead of N tool schemas, servers connect on first use and idle out, metadata cached to disk.

Canonical project memory for DeepSeek Harness: one governing article per asset, an append-only journal beneath it.

Request Context Profiler for DeepSeek Harness — see what changed between model requests, and how cache reuse changed with it.

X-ray for your DeepSeek Harness — diagnostics for what's actually loaded, why, and what it costs: per-plugin context-tax attribution, per-request token ledger, skill catalog pricing, dependency cascades.

Multitask 减震器模式（Agent preset）：主对话只做需求理解、任务下发与结果汇总，读文件/跑命令/构建/搜索等污染性工作全部下放子代理。含协调者窄工具面、四类 worker、只读元数据工具 file_info 与子代理危险操作的审批闸门，零 DSH 源码改动。

Markdown folder long-term memory for DeepSeek Harness: one file per fact, an index line per memory, mounted into every request as a system prompt section