Fang Gao Pet: A Floating SVG Companion Dynamic with DeepSeek Harness Session State
fanggao-pet (@local/fanggao-pet) is a DSH local UI bundle maintained by gavinff1990. By subscribing to the main view's SessionSnapshot, it displays a tri-state square cake on the shell.overlay: running, sleeping (idle), and angry (error). It supports dragging and position memory via localStorage, and the overlay allows click-through without obstructing operations. Installation requires running `install_bundle` on the source directory via the plugin manager, rather than using a one-line remote `add` command. Based on the official README, this article explains its principles, usage, customization, and troubleshooting, while advising users to review the source code and MIT license prior to installation.
Read MoreDigital Human Avatar Generator: From Appearance Descriptions to Digital Profiles
Digital Human Image Generator (@org-x2fzhf9b/digital-human-creator, v1.1.3, design-media category, published by Guangxi Junting Tea Industry Co., Ltd., source: enterprise) is an Agent Skill designed for digital human image creation scenarios. Users input gender, industry, products, precise target audience, and style requirements (appearance, face, hairstyle, clothing, voice), and the tool generates exclusive digital human portraits and produces complete profiles. It covers roles such as digital humans, virtual avatars, digital twins, AI spokespersons, brand digital humans, livestreaming digital humans, and customer service digital humans, applicable to brand endorsement, livestream e-commerce, private domain operations, and social media. The page tag indicates requires_api_key=false. The article lists input elements required prior to invocation according to the Skill introduction. It also truthfully states that the scraped data does not contain installation commands, configuration fields, or reproducible examples, noting that installation methods for Cursor / Codex CLI / Claude Code should refer to their official pages. Additionally, it highlights capability boundaries (excluding voice synthesis, lip-syncing, and motion capture), notes that profile fields are undefined, indicates that page data represents snapshot values, and advises users to independently verify portrait compliance and material copyrights prior to commercial deployment.
Read MoreDomestic Model MCP Server: Providing a Unified Entry Point for Agent Frameworks
This article introduces a community-released MCP server for domestic AI models (Repository: @user_1a470ba8/cn-model-gateway, Version: 1.9.0). Operating via the standard JSON-RPC 2.0 protocol, it enables Agent frameworks such as Claude Code, Cursor, and Cline to uniformly access ten domestic model providers: DeepSeek, Tongyi Qianwen, Zhipu GLM, Kimi, Tencent Hunyuan, Volcano Doubao, MiniMax, 01.AI (Yi), Baichuan, and StepFun. The server offers 11 MCP tools, gateway status resources, 2 prompt templates, unified error mapping, streaming SSE, a daily health report, a batch task queue, automatic failover, usage statistics, and token price tracking. Activation requires only filling in the `api_key` in `config.json`, with no need for GPUs, fine-tuning, or private deployment. This article outlines its features, configuration, usage, applicable scenarios, and precautions, providing the official link at the end.
Read MoreZero-to-App: Orchestrating the Process from Requirements to App Store Release into Executable Steps
**zero-to-app** is a community skill (AI agent category, version 1.2.2) published by @user_aa360a38 on SkillHub. It is positioned as a full-process orchestrator for mobile app development, covering seven stages: requirements clarification, tech stack selection, project scaffolding, architecture implementation, the TDD development loop, quality gates, and store release (including domestic store compliance). It acts as a process orchestrator and does not replace platform-level skills such as android/skills, expo/skills, callstackincubator/agent-skills, or flutter-claude-code; it only instructs the AI on what to do at each step and when to reference which platform skill. Trigger conditions include wanting to create a new mobile application, advancing any stage of an existing app project, inquiring about listing qualifications or filing/registration, consulting on distribution without store listing (including HarmonyOS / HarmonyOS NEXT and sideloading), or posting app requirement documents or error reports. It is not applicable to pure websites or backend services, nor to single-stack deep-dive issues already handled by platform-level skills. No API key is required for use (requires_api_key=false), and it is based on the general SKILL.md format. During this verification, neither the official page nor the API could be accessed, and the materials provided did not include installation commands or the full text of SKILL.md; this has been truthfully noted in the text, with a recommendation to refer to the official page for authoritative information.
Read MoreNaval Ravikant Almanack Wisdom Assistant: Handing Over The Wealth and Happiness Framework from “The Almanack of Naval Ravikant” to an Agent
The Naval Wisdom Assistant (@user_c5bbd3ce/naval-wisdom, v1.0.0, community release, knowledge-management, no API key required) is based on "The Almanack of Naval Ravikant" compiled by Eric Jorgenson (CITIC Press, April 2022), which organizes Naval Ravikant's public views into a structured framework. The wealth section includes wealth accumulation (specialized knowledge, leverage, equity, judgment, luck, patience) and enhanced judgment (clear thinking, mental models, decision-making skills, passion for reading). The happiness section includes three parts: learning happiness, self-redemption, and philosophy, along with life formulas and personal principles. The text lists ten types of triggering intents and three prohibited areas (investment operations, psychotherapy diagnosis, legal and tax matters), explains integration methods, typical use cases, and precautions, and accurately notes the unavailable installation commands.
Read Moredsh-context-map: Mapping Conversational Context to Hierarchical Information Architecture Diagrams
dsh-context-map, a plugin developed by gavinff1990 for the DeepSeek Harness workflow, uses the context_map tool to distill long conversations into structured nodes categorized by objectives, decisions, and risks. It validates and applies deterministic layouts to generate outputs such as SVG, interactive HTML, Mermaid Markdown, and revisable JSON. This article covers the installation commands, output file paths, usage of the `session_facts` and `update` functions, and important usage notes.
Read MoreSurvey Report Review Assistant: Automating Internal Review Pre-checks and Review Sheet Generation with Skill
The Survey Report Review Assistant is a community Skill on SkillHub (@user_9a888c7c/survey-review, version 1.2.3, categorized as professional, tagged with requires_api_key=false), designed for first/second-round reviews, proofreading, and auditing of geotechnical engineering survey reports. It extracts text and chapter frameworks from docx/doc/pdf/txt files to perform reverse calculation verification of statistical tables, validation of specification numbers, verification of compilation depth, scanning for template remnants, checking for typos and combined errors, and cross-checking structure burial depth against actual borehole depth. Furthermore, it conducts targeted verification using online checks for seismic resistance and a three-way cross-checking methodology for workloads. It produces a pre-check report and a review workbench, ultimately generating a "Survey Document Review Form." It also supports comparison of initial and final versions of comments, review archiving, and inspection and repair of headers and logos. This article outlines reproducible usage methods organized into four steps: input, pre-check, targeted verification, and report generation, while listing applicable scenarios and limitations (e.g., scanned copies, external network availability, consistency of standards, and the fact that signing responsibility remains with human personnel). Installation commands should follow the official page; paths are not listed here as the information has not been verified.
Read Moredsh-super-wechat-bot: OCR Screen Reading + Standalone Bridge Process WeChat Automation DSH Plugin
dsh-super-wechat-bot is a DSH client plugin maintained by gautema8-hash that reads WeChat 4.x text bubbles on Windows through PrintWindow and Windows OCR. Detection, decision-making, and sending run in an independent bridge process, not consuming agent turns. It supports draft/auto mode, human-like delays, and T0–T3 safety policies, and provides wechat_* tools along with a local panel at 127.0.0.1:47821. Based on the official README, this article introduces environment requirements, installation commands, usage, and known limitations.
Read MoreSlow SQL ER Doctor: Paste EXPLAIN to get ready-to-run index DDLs and rewrite solutions
This article introduces the "Slow SQL ER Doctor" Skill (version 1.2.4, categorized under dev-programming, published by community developer user_8ee65e73) with the identifier @user_8ee65e73/sql-slowquery-doctor-en. The tool’s core functionality allows users to input slow SQL queries alongside EXPLAIN output to generate directly executable CREATE INDEX statements, query rewrites (addressing issues such as eliminating SELECT *, fixing implicit conversions, and implementing cursor-based pagination), and pre-deployment risk checklists. Additionally, it provides line-by-line execution plan interpretation, write amplification alerts, diagnosis of locks and long transactions, and troubleshooting for the Top 5 slow log entries. The article provides the official address, coordinates, and trigger methods for the Skill, while outlining applicable scenarios and verification precautions prior to implementation.
Read Moredsh-dialogue-mindmap: Transforming Each Dialogue Turn into a Branching Mind Map on a Side Canvas
dsh-dialogue-mindmap is a DSH Web UI workflow plugin maintained by gaor0605-kq: it provides a Canvas next to Chat, with one node per dialogue turn in a left-to-right tree layout. You can branch follow-up questions from any node within the same conversation (Ctrl+Enter), and it includes a details sidebar along with pan and zoom. Installation command: `dsh plugin add dsh-dialogue-mindmap`. The tree structure is stored in localStorage, it is pure client-side, licensed under MIT, and tested with dsh 0.2.1-rc.5.
Read MoreOutreach Manager · Influencer Outreach Management: Turning Creator Outreach into Reusable Sequences
Outreach Manager · 建联外联管理（@user_2c0f1e77/outreach-manager）is a general-purpose SKILL.md Skill in the SkillHub community, sourced from the business-ops category, version 19.0.0. It unifies outreach and networking for influencers, media, and partners into a single mechanism: personalized initial outreach, multi-touchpoint follow-up sequences, negotiation scripts including objection handling, and status pipeline tracking, covering target types such as creators, media/analysts, launch partners, and social-selling/advocate recruitment. It is triggered when users directly request actions like "write influencer outreach," "follow up with a creator," "pitch a journalist/hunter/launch partner," or "negotiate partnership terms." The materials do not provide official installation commands; you need to check the SkillHub page for confirmation. Note that it is not used for finalizing signed agreements (use contract-helper), nor for media list grading, embargo terms, or press release structure (use press-media-relations).
Read MoreExam Paper Analysis Visualization Dashboard: Turning Final Exam Analysis into a Visual Board
A Skill designed for final exam paper analysis, coordinate @user_c00f577c/shijuanyibiaopan, version 1.0.0, source: community, category: Data Analysis. It integrates data visualization analysis and WeChat Official Account article layout design into a single workflow; after inputting exam paper data, it generates over 20 types of charts, and provides deep analysis reports, outputting HTML suitable for WeChat Official Account display. This replaces the traditional process of manual Excel statistics combined with separate layout work. The tag requires_api_key=false indicates no additional API key is required. The official documentation does not provide installation commands, input formats, or parameter descriptions; this article only documents verifiable coordinates, addresses, capabilities, and scenarios, and recommends conducting trial runs with real data before integration to validate the layout effect within the WeChat Official Account editor.
Read Morecaipu Flexible Recipe Assistant: Let AI Make Recipe Decisions Based on Fridge and Dietary Data
The caipu Mix & Match Recipe Assistant is a Model Context Protocol (MCP) companion skill for the caipu personal cooking assistant, published by user_5005e69a. It is categorized under 'life-service' with version 1.3.0. The official description states it provides 22 tools covering 7 scenarios: fridge clearance, dish recommendations, recipe recreation, logging, analysis, recipe management, and weekly planning. Its goal is to enable AI to suggest meals, log diets, and manage refrigerators based on real data such as the fridge inventory, dietary habits, and recipes. SkillHub tags show requires_api_key=false; at the time of scraping, the rating was 100,000, with 961 downloads and 0 installs. This article outlines the capability boundaries, entry points for the seven scenarios, key points to verify in the SKILL.md before integration, as well as suitable and unsuitable use cases, and provides the official URL.
Read MoreFinancial Analysis Toolkit v1.0: Bridging Financial Data Acquisition and Analysis with Agent Skill
Financial Analysis Toolkit v1.0 is a financial analysis skill package designed for Agent invocation, published by RealQ, sourced from SkillHub enterprise, version 1.0.0, and classified as professional. It leverages OpenClaw built-in tools together with AKShare and pywencai to fetch and analyze financial data, with its methodology referencing the Anthropic financial-services skill package. Its goal is to provide Agents with a unified entry point for data retrieval and analysis when handling financial analysis tasks, rather than having each project implement its own separate data ingestion code. This article, based solely on information retrieved from the official page, covers its positioning and capability boundaries, its mounting mechanism based on the universal SKILL.md format, and how to validate its usage given that the public documentation does not provide reproducible commands or examples. It also outlines applicable scenarios and considerations: the data coverage is determined by AKShare and pywencai, the official documentation does not specify concrete markets or metrics, and the tag requires_api_key=false indicates that no separate API Key application is needed under the current configuration. The article further notes that the package's metadata shows a rating of 100000, 1017 downloads, and 0 installations, and recommends verifying data source connectivity and data definitions before integration. Finally, the official URL and the SkillHub page link are provided.
Read MoreUser/Expert Interview Assistant: Breaking Down Qualitative Interviews into Five Stages
This is a community skill that covers the entire user/expert interview process, located at @user_6d1dc249/interview-research, version 1.0.0, categorized under knowledge management, with requires_api_key set to false. It divides interview research into five sequential stages: interview guide design, recruitment brief writing, candidate screening, transcript-to-summary conversion, and multi-summary synthesis into a research report. It is designed for product managers, designers, and business stakeholders to obtain actionable insights from qualitative research, positioned as generating actionable insights rather than academic analysis. Trigger words include interview guides, recruitment plans, candidate screening, transcripts, interview summaries, research reports, user interviews, user research, interview outlines, recruitment briefs, and similar terms. The official page could not be accessed this time, and the provided materials lack specific installation commands or client differences; users should check the official address to confirm the loading method before use.
Read Moredsh-badge: Unread badges for tabs and taskbar, with an end-of-turn notification sound.
dsh-badge is a DSH client plugin maintained by gaolvsuan. It displays the number of unread turns on the tab favicon and the Windows taskbar button, and can prefix the title with "(N)". At the end of a turn, it plays notification sounds such as chime, beep, and rise via Web Audio. It can be installed via the SkillHub marketplace or with `dsh plugin --profile desktop add dsh-badge`, and must be enabled in the profile bundles. It can be configured through cordis.patch.yml and the Host plugin management form. macOS and Linux do not have taskbar badge support. The plugin runs with the permissions of the DSH process, and it is recommended to review the MIT-licensed source code before installation.
Read MoreManufacturing Optimization Consulting Skill: Empowering Agents with Lean, TPM, and Lighthouse Factory Methodologies
On SkillHub, the "Manufacturing Optimization Consulting (Lean/TPM/Lighthouse Factory/Biopharmaceuticals)" skill is maintained by community publisher user_313caaeb, is version 2.0.0, is categorized as professional, and requires no API key. It organizes four areas—Lean Six Sigma, Total Productive Maintenance (TPM), Lighthouse Factory, and biopharmaceuticals—into knowledge packs that Agents can call directly. It is suitable for four scenarios: improving OEE, running Six Sigma projects, advancing TPM, and preparing for Lighthouse certification. The article introduces its positioning, core capabilities, how to install and enable it, typical use cases, and applicable limitations: the Skill outputs structured, methodology-level recommendations, does not include real-time data interfaces, and its authority should be verified by users.
Read Moredsh-wayland-use: Raw Desktop Control for DSH under Wayland
dsh-wayland-use is a DeepSeek Harness workflow plugin maintained by Ganglith that enables screenshot capture, coordinated OCR, text-based clicking, and wayland_act batch actions under pure Wayland through Mutter RemoteDesktop and ScreenCast. A persistent Python backend enables millisecond-level subsequent operations. The installation command is `dsh plugin add` pointing to the GitHub repository URL. It is suitable for GNOME/Mutter environments; the plugin runs with dsh process privileges, and the source code and MIT license should be reviewed before installation.
Read MoreSales Script Generation Assistant breaking sales conversations into copyable modules
Introducing the "Sales Script Generation Assistant" on SkillHub, published by Kunlun Growth, version 1.1.2, categorized under business-ops, requiring no API Key. It breaks sales conversations into six stages—ice-breaking, pain point confirmation, value presentation, objection handling, closing, and follow-up—and embeds three methodologies: SPIN questioning, FAB transformation, and the objection sandwich. It outputs ready-to-deliver scripts, accompanied by bottleneck self-rescue tips and compliance guardrails. This article describes its positioning, core features, typical usage, applicable scenarios, and limitations, and provides the official URL; installation commands and enabling methods for various editors should follow the official page.
Read MoreWeChat Moments Deep Profile Report: Organizing Exported Text into Evidence-Based Multi-Year Life Narratives
Introducing the Skill "WeChat Moments Deep Profile Report": Starting from index.txt, index.md, or long-form Markdown text exported from WeChat Moments, this tool generates an in-depth analysis report based on evidence. The report covers multi-year life narratives, identity transitions, values, writing styles, personal branding, and business/IP insights. The article verified facts including official coordinates, version 1.0.1, community source, the data-analysis category, publisher user_1314aee0, requires_api_key=false, as well as the default output format of an A4 PDF (with HTML serving only as an optional editable source file) styled after the reference Zhimiao report. It explains the inputs, outputs, and typical usage, and notes missing information such as installation commands and differences between Cursor, Codex CLI, and Claude Code, advising users to refer to the official page for the latest details.
Read Moredsh-study-notes: Building a Markdown Study Library and Local Learning Portal from Scratch with DSH
The GamecialX DSH workflow plugin `dsh-study-notes` registers the `study-notes` skill to the profile. Following built-in guidelines, it generates linearly readable course-style Markdown note libraries, hands-on project tracks, and checklists for readers, accompanied by an offline single-file KaTeX web portal (supporting exercises and chapter-end scoring). Installation requires local absolute paths (via Web/CLI/plugin_manager), and it provides `check-library` and `run-course` validation. Zero-dependency, MIT licensed, Node>=20, version 2.1.2.
Read MoreClinical Trial Safety Signal Expert (ct-safety): Drug-Event Signal Detection Using Public FAERS Data
ct-safety is a professional-use Skill with coordinate @user_ff7413f5/ct-safety, published by user_ff7413f5, category professional, version 0.10.1. It performs drug–event disproportionality analyses based on publicly available FDA FAERS reports, outputting PRR / ROR / IC / EBGM and their 95% CIs, and provides a 0–100 safety signal score with T1–T4 grading. Data sources include openFDA drug/event.json, drug/label.json, NMPA pharmacovigilance bulletins (cdr-adr.org.cn), DailyMed product labeling, RxClass MED-RT, openFDA recall enforcement reports, and Hong Kong ADR alerts; official labeling states that only public data are used and no confidential inputs are involved. This article describes its scope, data sources, applicable scenarios, and the limitations of FAERS spontaneous reports; installation commands and differences across clients were not included in this write-up because the official page was not accessible at the time of writing. It has been verified that requires_api_key=false.
Read MoreSelf-Media Content Editor: Turning Raw Materials into Platform-Ready Drafts
Self-Media Content Chief Editor (@user_f0937bfb/creator-editor, v2.2.3, community, content-creation, requires API key) targets Chinese content production: given a topic, a piece of material, or a draft, it directly delivers usable text suitable for short-video voiceovers, Xiaohongshu (RED), or WeChat Official Accounts, and supports cross-platform rewriting and de-AI-ification. Its core is fact boundaries: for products, cases, experiences, data, and tasks marked "no additions," it may only use facts already present in the input, without supplementing operations, scenes, reasons, effects, or title promises; when material is scarce, it writes briefly; when only a topic is provided, it must not impersonate personal experience, group observations, data, or verification results, and general methods are written only as suggestions to try; rewriting must preserve the original meaning without adding new viewpoints; it refuses to produce false personal narratives involving weight loss, diseases, therapeutic efficacy, investment returns, legal outcomes, or public safety; it outputs only the components the user explicitly requests, with no rules, analysis, checks, self-evaluation, or service invitations. It is called silently and directly delivers the finished Chinese product, with no explanations about loading, reading, or using the skill. The article presents its core capabilities, activation methods, typical usage, applicable scenarios, and limitations.
Read MorePrecision Development Protocol: Constraining AI Code Modifications Within Boundaries
The Precision Development Protocol is a Skill designed for code modification scenarios, maintained by SkillHub community publisher user_8c3c5215, version 1.0.1, categorized under dev-programming. Rather than introducing a new toolchain, it establishes six hard rules: editing boundary sandbox (only modify what needs to be modified), platform pitfall lookup table (rely on lookup, not reasoning), tool failure stop-loss (switch strategy immediately after two consecutive failures), cost awareness (complete tasks with the minimum number of tool calls), compile verification loop (must compile after changes), and Git discipline (commit as soon as stability is reached). It also defines four trigger word groups—action, style, constraint, and platform—pointing to style, text, and layout modifications as well as precision control requirements within the mini-program and uni-app ecosystem. The article explains the meaning of each rule, how they are triggered, and their applicable scenarios. It notes that the protocol's value lies in reducing invalid modifications and invalid retries rather than improving code quality, and that the pitfall lookup table depends on previously accumulated data, so new issues it does not cover still require standard troubleshooting.
Read Moredsh-notify-toast: DSH Native Notifications and Taskbar Unread Badges on Windows
dsh-notify-toast is a DSH client plugin (MIT, v0.4.0) maintained by game25011, designed for DSH Desktop 0.2.1-rc.2 on Windows 10/11. When the window is unfocused, it utilizes native WinRT toast notifications to alert users regarding prompts, authorizations, and root session turn/end completions. Features include Goal summarization, 2.5-second notification merging, privacy mode, and an optional taskbar unread badge. If the Host is unavailable, the plugin automatically degrades to Web Notifications. It is installed via the official `plugin_manager install_bundle` command using a `.tgz` file or directory link, with settings located in the DSH "Notifications" page. Note that this package is not published on npm; do not use `npm install`. The plugin operates without accessing external networks or calling models.
Read More