作者 · PerryLink

Web 作曲器终端风格输入历史：边缘优先的方向键召回并精确还原草稿与光标、浏览器本地持久化历史、Ctrl+R 反向搜索，以及滑动上下文感知；0.5.0 再加智能输入层——跨会话片段库（/save、/load）、带变量的提示词模板、复用洞察与压缩摘要高亮。

四合一迁移向导：把 Claude Code、Codex、OpenCode、Hermes 的会话、记忆、技能、指令与斜杠命令迁入 DSH（/move 向导，审批门 + 幂等 move.json，会话可续聊）。

把会话与工作区置顶到 Web 侧边栏顶部：行级图钉与换色、会话头开关、已置顶面板；0.4.0 再加导航组织器——Pin 分组（boards）、标签与保存视图、会话健康摘要与 /goto。

有界、分层、带审批门、可审计的跨会话记忆：`ctx.memory` 服务 + 零依赖 SQLite 存储 + `memory` 工具与冻结快照注入，并预演 dsh-memory-protocol v1——适配器注册表与可分发的一致性套件。

DeepSeek Harness 的 Claude Code /rewind 等价能力：每次变更型工具执行前捕获 git 优先的工作区文件快照，轮次边界 fork 会话，一条 /rewind 命令恢复文件并把会话回退到检查点。

DSH 的 LSP 动作面：诊断、格式化、补全、代码动作、符号、签名提示、inlay 提示与重命名，全部由真实语言服务器驱动。

DSH 插件开发知识库，作为按需加载的 agent 技能随 bundle 安装：官方约束、任务工作流、API 参考与社区踩坑，写插件时让 DSH 自己查。

Security-audit skill pack + plugin_vet supply-chain gate for DeepSeek Harness (dsh): 8 bilingual agent skills (secret scan, dependency audit, supply-chain review, prompt-injection review, audit orchestration, threat modeling, vuln intel, in

审批链上的第二模型自动审查：只读审查子代理返回带理由的 allow/deny 结构化裁决，默认 fail-closed。

Interactive long-session background agents for DeepSeek Harness: start a durable continuable child agent, watch its progress in the Web UI sidebar, message it any time, and interrupt it - all through the official subagent seam.

工程纪律守门：动笔前审讯需求，红绿测试证据门，交付后对抗评审，并汇总交付报告与逐维度核对。

官方级 GitHub CI 集成：composite action.yml、轮询 PR 评审机器人（幂等行内评论 + status-check 门禁）以及 PR/issue 工具，所有写入走人工审批门。

MCP management console for the official DeepSeek Harness MCP client: /mcp command with health diagnostics and pipeline trial calls, a Settings MCP tab with server CRUD (approval-gated writes, automatic backups) and a tool trial console over

运行时切换模型输出风格（对标 Claude Code outputStyles），另加 output.render.* 呈现协议：/style 命令、按会话持久化、systemPrompt 注入、六个内置风格、Web 选择器，以及带按会话/按工具规则与 /export 的渲染器注册表。

Claude Code 风格的声明式权限规则：按序 allow/deny/ask 的 YAML 规则，在 tools/pre-execute 瀑布上匹配工具名、参数、工作区路径与 agent 身份，带完整会话日志审计、干跑模式与热重载。

OpenTelemetry and Langfuse observability exporter for DeepSeek Harness: turn/step/tool/LLM spans, token and cost metrics, and sanitized LLM prompt/completion capture from the session/event stream, with async batching, bounded offline buffer

Cross-device DeepSeek Harness session sync: a dedicated git mirror with append-only keep-both conflict resolution (fork files, never silently overwritten), /sync command and sync_* tools

Prompt-injection, jailbreak, and secret-leak detection with allow/ask/block interception for DeepSeek Harness: an Aho-Corasick pattern engine and heuristics ported from the Prompt-Injection-Payloads, Jailbreak-Detector, and Secret-Key-Leake

Local document knowledge base for DeepSeek Harness: library_add/remove/list, hybrid semantic+keyword library_search with diversity re-ranking, relevance filtering and lost-in-the-middle avoidance, citation-aware injection, library_cite_chec

Cost governance for DeepSeek Harness: aggregated token/cost metering per model, session and day, budget caps with threshold alerts and over-limit policies, carbon footprint estimation, per-model latency benchmarks, a Settings budget tab, an

Cross-platform native desktop control tools for DeepSeek Harness (Windows first): screen_shot, screen_read (accessibility tree + pixel description for text-only models), click/type/scroll/key, and app_list/app_launch — every mutating action

Vendor parameter translation and deterministic JSON repair for DeepSeek Harness: the /translate command maps temperature/top_p/max_tokens/stop/system across 11 vendors, and the tools/post-execute repair layer (plus the fix_json tool) fixes

PII masking middleware for DeepSeek Harness: anonymize names, phones, emails, ID cards, bank cards, keys, and addresses to placeholders before they reach the model, restore them at the display layer, keep the restore table only in memory an

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 ??surfac

Isolated install-and-smoke test drives for DeepSeek Harness plugins: installs a repo or npm package into a throwaway DSH_HOME profile, verifies the bundle patch layer and boot logs, records a structured pass/fail result matrix (JSON/Markdow

Unified static-image generation router for DeepSeek Harness: one image_generate tool with standard parameters, config-driven engine routing (OpenAI Images, Zhipu CogView, and any compatible endpoint) with health-aware fallback, durable work

Multi-dimensional quality scoring for DeepSeek Harness plugins: scores one repo or npm package across install success (consuming dsh-test-drive results), maintenance activity, documentation completeness, security scan, and protocol complian

Local-model (Ollama) integration for DeepSeek Harness: discover, pull, remove, and inspect local models, route requests to them by task type or keyword with automatic fallback to the cloud, and get a one-shot status overview via /ollama.

Voice-first session loop for DeepSeek Harness: a composer microphone button with browser/local speech-to-text (Web Speech, FunASR, whisper.cpp), a speak tool for text-to-speech replies (browser, edge-tts, piper), event announcements with mu