Productivity · DeepSeek Harness plugins — Page 58

DSH 的 fork 伴随插件:官方 fork 发生后,把原会话的直接子 agent 复制到 fork 出的新对话下(递归整棵子树,副本经官方 agent API 创建),每个副本成为可被官方恢复路径唤醒的冷会话,新对话保留子代理面板并能继续给它们发消息。Copy a forked DSH conversation's subagents onto it and keep them reachable.

Feishu/Lark bot bridge for DeepSeek Harness: bind a web-created session to a Feishu chat, messages flow both ways

Accurate cache-hit rate, token usage, and cost estimates for the DeepSeek Harness Web UI: one-decimal hit rate, bucket breakdown, peak/off-peak pricing with auto price-switch, and a session usage panel.

DeepSeek Harness (dsh) plugin: visualize Git status in the Web UI — current branch, HEAD, staged/modified/untracked counts, ahead/behind, recent commits and changed files.

DeepSeek Harness plugin: bind every conversation to its own isolated git worktree + branch in one click, so parallel agents work on one repo without interference — binding manager panel + 9 git tools

GitHub 集成插件 for DeepSeek Harness：一次认证（GITHUB_TOKEN 或 OAuth Device Flow）后模型可直接调用 GitHub REST API 完成建仓、推送、提 PR、管 issue、搜索等 45 个工具；常规读写直接放行，删除/force push/合并 PR/关 issue 等危险操作保留审批门；token 存于 DSH credentials 服务，不落日志。 / GitHub integration plugin for DeepSeek Harness: authenticate once (PAT or OAuth device flow), then the model drives 45 github_* tools over the GitHub REST API. Routine reads/writes pass through; destructive operations (delete, force push, merge PR, close issue) keep an approval gate. Tokens live in the DSH credentials service and never appear in logs.

Global task library with a persistent interactive panel: task CRUD tools, linked-subagent terminal-status sync, and a generative-UI surface renderer for the DeepSeek Harness

Enhanced goal surface for the DeepSeek Harness web UI: full lifecycle dock above the composer (create / multiline edit / pause / resume / complete / clear with live round progress), a composer tool-row entry with phase dot, a collapsible chip, and a General-settings switch to hide the entry.

DSH plugin: strip GPT-family tool-call sandbox_permissions that are not strictly wider than the current session.

DeepSeek Harness bundle that exposes the Grove CLI as agent tools.

Recall and re-run the current session's command history with ↑/↓ keys in the DSH Web composer.

QQ 官方机器人接入插件：让用户通过 QQ（私聊 / 群聊 / 频道 @）直接与 harness 完整 agent 对话（工具 / 记忆 / 子代理 / 文件系统 / 安全护栏，与 Web UI 同源）

Langfuse observability plugin for DeepSeek Harness: one generation per LLM call, one span per tool call, one trace per session turn.

DeepSeek Harness (DSH) bundle: a delete button on the last user message that soft-deletes that turn — user message plus all its model replies — keeping it out of future model context, stepping back to the previous user message until a compact boundary or none remain (最后一条用户消息上的删除按钮：软删除该用户消息及其全部模型回复，不再进入后续模型上下文；按钮随后移到上一条用户消息，直到 compact 边界或无消息)

DSH 跨会话长期记忆：命名空间隔离 + L1 索引注入（存在性编码）+ 行动验证写入 + 内容级近重复去重 + BM25 全文检索 + 跨命名空间提升 + 自动蒸馏候选 + 溯源/归档/回滚 + 自动维护。

DeepSeek Harness bundle for a Windows desktop UI context picker and MCP server for Codex, DeepSeek Harness, and AI agents

DSH web plugin: bridge to third-party web LLMs (Gemini, ChatGPT, Doubao, custom) — agent-driven consult_llm tool with review/code-review/adversary/verify roles, text-log primary UI with on-demand live view, named sessions, token metering, headed/headless browser daemon.

Popup-on-open + periodic reminder to save DSH session logs, with in-popup interval (10min-3h), backup folder & format (fish=raw .zstd / human=readable .txt) settings and one-click incremental copy. / 打开 Web 即弹窗 + 定时提醒保存 DSH 会话日志，弹窗内可设提醒间隔（10分钟-3小时）、备份文件夹与格式（鱼话版=原始压缩 / 人话版=可读聊天记录），一键增量复制。

AX-first computer use on macOS for DeepSeek Harness: accessibility-tree targeting instead of screenshot coordinate guessing, window-scoped input, action read-back verification, blank-frame-checked screenshots, and optional Jev (TypeSafe System One) semantic guards before irreversible actions — bridged to the macos-cu CLI.

DeepSeek Harness 插件:数学建模竞赛自动化论文撰写系统 v3(中控台双重质疑 + 五阶段流水线:扫描验证/评审决策、P0-P3 评级与阶段级回滚、教训持久化、索引导航、兜底定稿 Final_Paper.md、断点续跑、进度面板)

Codex-style per-session memory for DeepSeek Harness: qualified sessions auto-distill into per-session drafts (rollout_summaries); manual add/change/forget runs only on explicit user request; layered disclosure (summary + registry); idempotent integration watermark. / 一会话一草稿记忆插件：合格会话自动提炼草稿；手动增/改/忘仅在显式请求时执行；分层披露（总纲+注册表）；幂等整合水印。

Local, cross-session full-text search for DeepSeek Harness conversations: SQLite FTS5 indexing with Chinese word segmentation and accurate message-level jump targeting

DeepSeek Harness 通知插件：会话交互/任务完成/出错时通过系统通知、浏览器通知、飞书/企业微信/Discord/钉钉/Telegram 推送提醒

A model-native meta-agent plugin for DeepSeek Harness: the model synthesizes task-specific workflows at runtime from five agentic workflow patterns (prompt chaining, parallel workers, router, supervisor, evaluation loop) and coordinates tools and subagents — without fixed pipelines, templates, or hard-coded routing.