dsh-subagent-bridge: Building a Subagent Bridge between Codex and DeepSeek Harness
`dsh-subagent-bridge` is a DSH workflow plugin maintained by Dogtiti that exposes a local MCP endpoint within the host. It enables Codex or Claude Code to dispatch tasks such as `dsh_task`/`dsh_spawn` to the Harness in a subagent manner and retrieve final responses. DSH agents can also reverse-delegate via `codex_dispatch` and `claude_dispatch`. It supports specifying models and workspaces per call, queue/steer follow-ups, and unattended hypothetical policies. Requirements: DSH ^0.2.1-0; MIT license; the endpoint is written to `~/.dsh/bridge/endpoint.json`.
Read Moredsh-video-vision: Storyboards and Text Maps Enabling the Harness to Directly "See" Video
dsh-video-vision is a DSH client plugin maintained by Destinhue that uses ffmpeg for local video analysis, providing five tool chains: video_probe, video_outline (zero visual token text map), video_storyboard, video_zoom, and video_read. It supports file paths, direct links, and Bilibili links. To install, enter the plugin directory, execute `node bin/install.mjs`, and restart DSH. It is suitable for scenarios requiring Agent reasoning with reference to screen content or on-screen text. Before use, please review the source code and MIT license, and note that the plugin runs with DSH process permissions.
Read MoreShuoruan · Cross-border E-commerce Profit Calculation: Turning Selling Price and Procurement Cost into Dynamic Break-even Models
Input the selling price and procurement costs to generate a formula-linked Excel P&L model and a one-page Word decision report. The system covers Amazon FBA, Temu (Full-Service), Shein, TikTok Shop, and DTC sites, featuring automatic switching between commission-based and full-service reconciliation modes. It includes four inverse calculation indicators: break-even selling price, break-even ACOS, reverse-calculated pricing based on target net margin, and price reduction headroom. It offers a four-tier traffic light grading for net profit margins. Selling prices support multi-currency with editable exchange rates, while costs are in CNY. The solution is zero-dependency, requires no coding, uses no API keys, and works offline; industry defaults are applied for missing parameters and clearly marked as estimates.
Read MoreDSH Technique for Accelerating Large File Downloads in GitHub Releases
Maintained by chenjrastro-tech, the DSH plugin `github-release-download-acceleration` operates with zero dependencies under Windows PowerShell 5.1 and the system curl. It features self-checking of proxies and mirrors, testing across four types of routes, time-limited concurrent ladders for large assets, multi-segment range downloads, and SHA256 verification using release digests. It supports natural language skills and the `ghdl.ps1`/`ghdl.sh` runners. Installation via `dsh plugin add github:chenjrastro-tech/github-release-download-acceleration` is recommended.
Read MoreDating Coach: Orchestrating Emotional Methodologies and Networking Resources into a Skillset
Love Coach is the main entry point for a skill package published on SkillHub, located at @org-bzwfrdvh/love-coach, published by AIPDD, version 1.0.1, categorized under education. It contains 10 sub-skills: 4 on the methodology side, distilled from approximately 200,000 words of first-hand emotional materials, containing 5 mental models, 9 decision heuristics, and a complete expression DNA, covering scenarios such as relationship analysis, relationship judgment, chat responses, pickup openers, and more; and 6 on the networking side based on AIworkin MCP, covering contact sourcing, demand posting, AI communication preview, SNS expansion, service listing, order taking, and more. The main entry point is responsible for demand guidance, information validation, and sub-skill orchestration, claiming to have successfully tested 30 MCP tools. An API Key is required for configuration. Specific installation commands and configuration fields could not be obtained from the official page during verification, and this article has not fabricated any information. It is suitable for products that need to handle emotional counseling conversations in Agents or connect to networking resources. Platform compliance requirements should be confirmed before integrating with the networking side.
Read Moredsh-fish-river: The "Big Fat Fish" Where Personalities Can Only Grow from the Event Stream
dsh-fish-river is an MIT-integrated version of ssamurias/dsh-fish and Britneycode/dsh-fish-soul: It has no personality settings page; personality relies on events.jsonl and four-stage growth. Among the three-layer memory, only the growth layer defines "who she is". The same identity can speak in DSH sessions as well as via QQ, Bilibili private messages, and optionally Douyin. Channels are disabled by default with visibility and whitelist constraints. It introduces installation commands, memory models, social layer gates, and typical usage, and reminds that plugins run with DSH process permissions.
Read Moredsh-island: DeepSeek Harness Desktop Dynamic Island
dsh-island is a community-maintained desktop plugin for DeepSeek Harness (Windows, requires Harness 0.2.1-rc.5) that provides a task status bar similar to the Dynamic Island at screen edges. When expanded, it allows users to view plan progress and context usage, stop tasks, approve permissions, respond to agent questions, and continue messaging without opening the main window. It also includes usage statistics and cost estimation based on official pricing. To install, add dsh-island from the Harness plugin page, or run `dsh plugin --profile desktop add dsh-island` after exiting the system tray. Licensed under MIT, currently at version 0.6.1.
Read MoreExam Question Analysis Generator: Reverse-Engineering Answers into Two Sets of Exam Paper Files
The Exam Question Parser (@user_015022d9/exam-question-parser) is a community-sourced education category Skill, version 1.0.0, requiring no API Key. It targets scenarios where existing answers and format requirements need to be transformed into exam papers: first parsing answers (extracting correct answers, identifying knowledge domains and concepts, annotating difficulty levels, extracting solution approaches and common pitfalls), then assigning question types based on concept nature (multiple choice/multiple selection/fill-in-the-blank/experimental/essay) and composing academic Chinese question stems, constructing distractors based on common errors from the original analysis, and allocating scores according to format requirements. It subsequently calls scripts/generate_exam_docs.py to generate a question file (exam structure table + grouped questions, with no trace of answers) and an answer file (answer key table + detailed question-by-question analysis), then directly displays both files for review. Trigger phrases include "generate questions from answers", "generate exam questions", "create exam", "reverse-engineer questions from answers", "generate exam from answers", etc. On SkillHub, it has a rating of 100000, 1284 downloads, and 0 installations.
Read MoreFact-Checking: Installing a Provenance Firewall for Agents
Fact Check (@user_b8060655/cue-fact-check) is an Agent Skill released by community developer user_b8060655. Categorized under knowledge-management (Version 1.0.1), it requires a user-provided Cue backend API key. Designed for journalists, editors, and researchers, it systematically deconstructs factual claims found in press releases, social media posts, research reports, policy interpretations, and AI-generated content. It traces sources hierarchically according to a credibility structure: T0 official sources → T1 authoritative media → T2 industry media → T3 supplementary sources. By employing a firewall mechanism to eliminate circular reasoning, it provides deep links for every conclusion and applies a four-tier rating system: ✅ Fact accurate / ❌ Fact error / ⚠️ Bias / ⚪ Unverifiable. The tool focuses on identifying temporal displacement, detail discrepancies, recycled news, and manipulated content. It is not suitable for subjective opinion disputes, personal privacy investigations, verification of real-time breaking news, or legal evidence authentication.
Read MoreCinematic AI Short Drama & Comic Pipeline: The "Prompt Skill" Workflow for One-Stop Production
Released by Woda Technology on SkillHub (coordinates: @woda/ai-maju-duanju, version 1.0.0, category: content-creation, metrics: rating 100,000, 1,481 downloads), this Skill is positioned as an integrated workflow for AI short dramas. It accepts input in the form of novels, plot synopses, or story ideas to generate an end-to-end, ready-to-use production package. This includes a cinematic production document (`script.html` featuring a three-sentence storyline, "Save the Cat" beats, and character/scene/visual tone definitions), a film-style "STYLE LOCK" visual scheme, an asset package of image prompts for character face locks/setting sheets/storyboards/props, a complete storyboard set of video prompts (anchored second-by-second for character/environment/style/camera/audio/timeline, including English text and recommended duration), and a production manual covering the entire process from script to final cut. All terminology is derived from a built-in 10-chapter library of film and television prompt terms (film/master styles, lighting, color, shot types, camera movement, sound, acting, editing) without fabrication, ensuring stylistic consistency and professional reproducibility. It is suitable for AI short drama creators, novelists, and short-form video content teams. Official URL: https://api.skillhub.cn/woda/ai-maju-duanju.
Read Moredsh-telegram-relay: Relay DSH Q&A, Approvals, and Round Summaries to Telegram
dsh-telegram-relay is a DSH client plugin maintained by sflaer that launches a loopback control panel on the host side and forwards ask_user_question prompts, tool approvals, turn summaries, and sub-agent notifications via a Telegram Bot. Both the Web UI and Telegram can respond simultaneously, with the first response taking effect. It supports ru/en/zh and uses long polling, eliminating the need for a webhook. Based on the official README, this article covers installation for desktop and web profiles, bot binding, whitelist and sub-agent toggles, cordis.patch.yml configuration options, and security considerations.
Read MoreSimplifying Financial Writing: Making Bank Materials Read Like They're Written by Industry Veterans
"Say It in Plain Human (De-AI-flavored—Banking Writing Materials)" is a community Skill in SkillHub’s content-creation category, located at @user_0d2b33de/srh, version 1.0.0, and does not require an API key. It targets materials for banks, fintech, government collaboration, leadership reporting, implementation guidance, meetings, and business plans, providing rewriting, diagnostics, or drafting. Its goal is to transform mechanical, defensive, vague, slogan-like, or obviously AI-generated Chinese drafts into writing that leads with the conclusion, carries judgment, provides actionable levers, is executable, and uses organizationally appropriate phrasing, while preserving factual accuracy and risk boundaries. The article explains what it is, its core capabilities, how to enable it based on @user_0d2b33de/srh, example rewriting requests, applicable scenarios and cautions, and provides the official URL.
Read MoreBook Writing · Expert-level Professional Book Writing Engine: Turning Your Expertise into a Publishable Book
Released as version 1.0.0 on the SkillHub community, this AI-assisted writing Skill is tailored for academic, industry, and technical monographs as well as professional textbooks. It constrains outputs through five ironclad rules—meticulous terminology, rigorous argumentation, outline-first approach, elimination of AI-generated artifacts, and source discipline—and integrates a five-stage workflow comprising genre and register calibration, argumentation and outlining, terminology governance, chapter-by-chapter drafting, and comprehensive global review. The article outlines its positioning, core capabilities, installation and activation procedures, typical invocation examples, and applicable scenarios. It also clarifies that the Skill does not handle plagiarism detection or submission workflows, and that bibliographic data still requires manual verification.
Read MoreSupervision Log - DOCX: Documenting Daily Construction Activities in Standardized Tables
Jianli Rizhi-docx is a community-sourced general skill (@user_8c865e02/jianli-rizhi-docx, version 1.6.4, classified under office-efficiency, requires_api_key=false). It takes raw input regarding daily construction and supervision activities, uses AI to optimize the text into standard professional language, and generates a .docx file based on a fixed table format (Title, Project Name & ID, Date, Temperature, Weather, Key Supervisory Personnel Movements, Construction Environment, Construction Content, Supervision Work, Issue Rectification, Others, Recorder, Chief Supervisor's Signature). The template is embedded within the generator as Base64 (BUILTIN_TEMPLATE_B64). Placeholders are validated before generation to ensure consistent formatting regardless of changes in model, environment, or user; rebuilding the template is prohibited. Built-in rules remove "AI-flavored" language, modifying only clichés and jargon while preserving numbers, units, and technical terms. Chinese punctuation is enforced throughout, while standard formats for dates, pile numbers, and intervals are retained as-is. The "Issue Rectification" field defaults to "None" if no content is provided, and "Key Supervisory Personnel Movements" should contain names only. Weather data is queried via the keyless Open-Meteo API based on date and location, supporting historical data normalized to a controlled vocabulary. Optional photos are automatically cropped to a 4:3 ratio and inserted at the end of the document in a 2×4 grid. Project information provided by the user upon first use is retained permanently. The generation engine relies solely on the pure Python standard library with zero third-party dependencies (except for optional Pillow for photo compression). As the sourced materials did not include installation commands or loading differences for Cursor, Codex CLI, or Claude Code, these details are not supplemented here.
Read MoreMemWeave: A DSH Plugin for Weaving Memory into Personality
**English Translation:** "Memoria (dsh-mem-weave) is a memory-and-personality plugin developed by QuaonY for DeepSeek Harness. It uses local SQLite storage, and features configurable default token limits for per-turn injection (e.g., 4000/1200/900 characters), supporting layered memory, half-life-based forgetting, conflict freezing, and 24-hour retraction. It provides an eight-page management interface and weave_* tools; the AI can only write memories and suggestions, while personality and rules remain subject to human adjudication. Installation can be completed via the tgz package from Releases and dsh plugin add. It is licensed under MIT."
Read Moredsh-ccd: Transforming Reasoning into First-Person Short Notes, Retaining Full Chain of Thought in Session
dsh-ccd is a DSH model inference plugin maintained by mmyddd. It translates raw Chain-of-Thought (CoT) from the interface into first-person work notes of one or two sentences in real-time, supporting original text viewing and historical session restoration; the Session and model sides retain full reasoning capabilities. It can be installed via the plugin page by adding a GitHub repository or downloading a release tarball. It supports transcription via providers or local MiniCPM5-2B, with configuration options including language, segmentation, concurrency, and retention policies. Licensed under MIT, notes are stored in SQLite under the DSH_HOME directory.
Read MoreDSH Fujiang Unrestricted: Desktop One-Click Install "Work Assistant" Presets and 240+ Skills
DSH Fujiang Unrestricted is a DSH client installation package maintained by HanKuCha: on Windows, DSH-Fujiang-Installer.exe only modifies DSH_HOME, registers the "Work Assistant" (ws-helper) preset and 240+3 bundled skills, without touching the program directory, disabling the sandbox, or bypassing approval. The article describes the preset composition, installation and uninstallation steps, environment requirements (Win10/11, .NET 4.8, DSH Desktop 0.2.x), and usage boundaries, and provides the SkillHub and GitHub sources.
Read MoreInterpreting Dreams (A Practical Guide): First Determine the Type, Then Provide a Four-Dimensional Analysis
This Skill is a community knowledge-based dream-interpretation Skill, located at @user_746fc03f/zhougongjiemeng, version 1.0.0, category knowledge-management, with the tag requires_api_key=false. It does not directly classify dreams as auspicious or inauspicious. Instead, it first identifies different types such as past-life dreams, prophetic dreams, organ-related dreams, and natural dreams, and then interprets them across four dimensions: Duke of Zhou’s dream interpretation, TCM dream diagnosis, modern psychology, and Buddhist and Taoist spiritual practices. The knowledge base covers 27 major categories from the Complete Original Duke of Zhou Dream Interpretation, the theory of dream-inducing pathogenic influences in the Huangdi Neijing, Freud and Jung’s modern psychology, and Buddhist and Taoist spirituality. The article outlines the core features, how to enable the Skill, and typical usage examples, and warns that it does not replace medical diagnosis, psychological counseling, or legal judgment.
Read Moredsh-aio-agent: Unified Integration of Coding Agents like Cursor and Claude Code into DSH
The DSH plugin **dsh-aio-agent** (v0.0.1, MIT), maintained by cocofhu, integrates Cursor, Claude Code, Codex, OpenCode, CodeBuddy, and Kimi Code into the DeepSeek Harness mode menu. In each turn, it calls the corresponding CLI or Kimi ACP on the host machine running DSH. DSH is responsible for rendering, Skills, /resume, and session logging, without invoking any proprietary models. The article details official installation commands, installation and login instructions for each agent, configuration options, and known limitations.
Read MorePersonal Knowledge Management: PARA Aggregation, Atomic Notes, and the Review Loop
This article introduces the "Personal Knowledge Management" Skill on SkillHub, located at @kunlungrowth/kunlun-knowledge-pro. Released by Kunlun Growth as version 1.2.0 and categorized under knowledge-management, it requires no API Key. The Skill organizes the PARA four-quadrant system, atomic notes, bidirectional links, evergreen notes, spaced review, and output-driven input into a set of actionable rules. It also embeds built-in parameters such as atomic note granularity, link density thresholds, weekly and monthly review cadences, card reuse rates, and a checklist of common distortions. It covers four primary scenarios: reading notes, second brain, research materials, and team knowledge accumulation. The article details its core features, installation and activation methods, usage considerations, and provides the official URL.
Read MoreSAP Financial AI Assistant: Financial AI Capability Package for SAP FI/CO
SAP Finance AI Assistant (user_7f2d3808/sap-finance-ai) is a community-sourced skill categorized as “professional,” currently at version 1.0.0, used to build AI capabilities for the SAP FI/CO finance modules. It includes three built-in analysis engines and seven scripts, covering the setup of AI scenarios for AR/AP, General Ledger, Asset Accounting, Cost Ownership, and Profitability Analysis, as well as batch maintenance and validation of four master data types—materials, customers, vendors, and prices—and it provides three SAP system connection configuration methods: RFC, OData, and BTP. The documentation structure consists of 30 files, six documents, and six role-based usage guides. Key technical features include zero dependencies, data linkage, leakage detection, IQR-based anomaly detection, seasonality analysis, and OData support. The tag is requires_api_key=false (no external API key required). At the time of data retrieval, the page shows a rating of 100000, 1,280 downloads, and 0 installations. It is suitable for building foundational SAP AI capabilities, implementing AI in finance scenarios, routine master data maintenance, and producing technical documentation.
Read Moredsh-identity: Providing DSH peers with a set of verifiable local identity statuses
dsh-identity is a DSH memory plugin maintained by Anarrogantcat. It provides agents with an editable kernel, evidence-based traits, memory and consolidation with confidence levels, journaling and reflection, mood v2, autonomous proposals, and governance/runtime capability layers. The identity segment and the "Big Fat Fish" panel function in any DSH session, with QQ serving as an optional channel. Based on the official README, this article introduces the five-layer architecture, installation commands (github:Anarrogantcat/dsh-identity), panels, and common group commands, while explaining local permissions, code gates, and optional self-configuration items such as voice and Bilibili.
Read MoreDataMax Data Visualization Large Screen Assistant: Breaking Down Large Screen Design into Executable Steps
This article introduces the community Skill "DataMax Data Visualization Dashboard Assistant" (Location: @user_10b0a9e9/datamax-cli-suite, Source: community, Category: data-analysis, Version: 1.0.10). It targets tasks involving data visualization, data dashboards, BI dashboards, executive cockpits, operational boards, sales dashboards, and screen design. It is applicable when planning layouts and visual styles, searching for templates, components, or assets, or creating, modifying, beautifying, and validating DataMAX dashboards. The core mechanism is task routing: selecting either datamax-max-design or datamax-max-peiping based on the specific task. If the formal capabilities are not installed, it guides the user to install them via datamax-skill. This Skill carries the `requires_api_key=true` tag, so credentials must be confirmed before use. Since installation commands and invocation parameters are not provided in the source material and have not been fabricated here, referring to the official page is recommended.
Read Moredsh-comfyui-workbench: Enabling the DSH Model to Invoke ComfyUI Workflows
dsh-comfyui-workbench is a DSH workflow plugin maintained by YuChenLiao. It exposes local ComfyUI workflows to the model for invocation via tools such as comfyui_run, and provides three UI panels: an asset library, workflow management, and a ComfyUI terminal. The plugin is split into seven entries—core, gallery, workflow, generate, edit, danbooru, and terminal—each of which can be toggled on or off independently. Configuring parameters such as baseUrl, workflowsDir, and outputDir is sufficient to connect to a local ComfyUI instance. Install it with the command `dsh plugin --profile desktop add github:YuChenLiao/dsh-comfyui-workbench`; a DSH restart is required after installation. It is designed for developers who need to list, inspect, and asynchronously execute arbitrary API-format workflows within agent conversations, and to centrally manage generated image outputs.
Read Moredsh-enhance: Adjust Brand Persona and Content Boundaries in System Prompts
dsh-enhance is a DeepSeek Harness plugin maintained by wufeng0105 that modifies system prompts through Cordis patches: it can retain, remove, or customize the opening branding sentence, append up to 8 KB of content boundary text at the end, and configure out-of-bounds approval with optional Enhanced permission levels. Configuration is stored in ~/.dsh/enhance/, synchronized between the settings page and files, and supports checking session logs to verify effectiveness. Applicable to DSH versions before 0.2.1-rc.1 to rc.6. MIT licensed.
Read More