スキル · DeepSeek Harness プラグイン — 74 ページ

Unofficial DSH plugin — long-term, cross-project memory for coding agents: an embedded project-portfolio skill plus portfolio_status / portfolio_log tools over plain markdown files the user owns. Not affiliated with DeepSeek. 非官方社区插件，与 DeepSeek 无关联。

Project-scoped custom SKILL roots for DeepSeek Harness

DeepSeek Harness 的三合一能力台：提示词定制（按名称屏蔽 / 替换 / 注入 / 排序，按会话阶段与 agent 预设分层）+ 技能集合管理（导入 / 分组 / 开关）+ MCP Server 接入（分层遮蔽与工具级启停）。纯插件注入，不动官方源码。

提示词工程模式：结构化推理（CoT/ToT）、few-shot、模板变量、生产级提示词优化。受 wshobson/agents（38k★ MIT）启发。

DSH 提示词锻炉:零配置流式优化(复用会话模型)+ 轻量词库(JSON 文件)+ 一键存技能。Prompt Forge: zero-config streaming optimization + file-based prompt library + one-click skills.

SQL 优化模式：EXPLAIN 分析、索引策略、N+1 解决、查询改写。受 wshobson/agents（38k★ MIT）启发。

User-invocable XDSP frontend release skill for DeepSeek Harness

PyMOL headless scripting tools for DeepSeek Harness (skeleton: tools planned, see README)

SQL 优化模式：EXPLAIN 分析、索引策略、N+1 解决、查询改写。受 wshobson/agents（38k★ MIT）启发。

量化回测与风险指标：规避前视/幸存者偏差、考虑交易成本、walk-forward 分析、风险指标计算。受 wshobson/agents（38k★ MIT）启发。

`dsh-quick-invoke` 是一个面向 DSH Web 的独立 Cordis 插件，为 Skill、Agent preset 和 Plugin 提供统一的 `/` 快捷入口、候选选择和 Host 命令执行能力。 本插件不修改 DSH 核心，不提供任意 Tool 直调语法。Tool 仍由 Agent 根据自然语言选择，并继续经过 DSH 的 tools、approval、guard 和 permission 流程。

DSH bundle: Qwen multimodal bridge — vision (qwen3-vl), speech-to-text (qwen3-asr), text-to-image (qwen-image), via the deepseek-vision skill scripts

RAG 检索增强生成：向量数据库、嵌入、语义搜索、减少幻觉、来源引用。受 wshobson/agents（38k★ MIT）启发。

自适应 GitHub README 撰写与截图技能：先判断目标类型与截图能力——网页走浏览器截图，无网页的控制台/CLI 走终端截屏（System.Drawing / HTML 终端 / termtosvg），能则询问并自动配图，不能则走无截图模板；强调相对路径防丢图。作为 dsh 插件（cordis skill-bundle）安装后由 host 层注册 readme-writer 技能；SKILL.md 含可被插件商城索引的 name/description/keywords/category/version 元数据。

Realtime Avatar (realtimeavatar.ai) for the DeepSeek Harness: the developer's API key held by the harness, the public docs as on-demand skills, rta_* tools over the public REST API, and a /rta command that walks a developer from no key to a live avatar call.

Record & Replay for the dsh web GUI: (1) replay every automatically recorded session as a readable timeline, export/import shareable replay packs, re-run a recorded conversation; (2) Codex-style screen recording -> skill generation - record a computer-use session in the browser (getDisplayMedia -> webm + sampled frames), replay it, and have an agent analyze the frames and install a generated SKILL.md into ~/.dsh/skills. Hot-pluggable - mounted via the profile patch layer, no dsh source changes.

DSH refactor-insight skill plugin: turn codebase smells into an executable, priority-ordered refactoring plan (file-length, long-function, deep-nesting, god-object, high-coupling, and TODO-density heuristics plus a staged runbook).

DeepSeek Harness Web GUI remote HTTPS access bundle: Cloudflare tunnel + local password proxy guidance, legacy-browser compatibility taps (AbortSignal polyfill, phone bootstrap), browse directory picker, and the bundled dsh-remote-access skill. Install once, use everywhere.

DSH plugin bundling the remote-shell skill: secure SSH/SFTP/Telnet/WinRM remote operations with an encrypted credential vault.

Ships a skill that diagnoses and repairs a broken DSH deployment: boot failures before readiness, dead plugins, broken dependency bridges, and safe upgrades.

AI4Scholar research pack for DeepSeek Harness (dsh): research skills, literature-first workflow guidance and workspace conventions. Pairs with dsh-ai4scholar for search, full text, citations and figures.

DSH plugin: tidy messy pasted citations into uniformly formatted references (refs_parse → refs_verify → refs_dedup → refs_format + research-refs skill)

ResearchCraft as a DeepSeek Harness profile: DSH web UI and agent, scientific skills, lab notebook, specialist subagents.

Resume screening plugin for DeepSeek Harness (HR): ingest up to 100k Word/PDF/Excel resumes into one searchable library with skills / coarse categories / ingest time, auto-ingest & de-dupe new resumes, then screen by plain-language criteria — rule pre-filter in the DB plus LLM fine-judge on the shortlisted few, CSV export. 简历筛选插件:批量收简历建档(技能/粗分类/入库时间),大白话下指令,库内规则粗筛+LLM精判少数入围者,自动入库去重、可导出、可追溯。