Audio Transcription Extraction Pipeline · Local Privacy First: Local Whisper Transcription and Structured Extraction
This section introduces the SkillHub community Skill "Recording Transcription & Extraction Pipeline (Local Privacy First)" (@user_65dc9f28/asr-pipeline, v1.1.0, classification: knowledge-management). It utilizes `faster-whisper` to locally transcribe WeChat voice messages, meeting recordings, and training audio into text. Featuring a built-in `scripts/transcribe.py` script, it uses PyAV for decoding (eliminating the need for `ffmpeg` installation) and requires zero API keys. The tool automatically categorizes recordings (meeting, interview, training, or other) and, upon one-click confirmation, applies templates to generate meeting minutes, interview insights, or knowledge extraction cards. The organization phase is handled by an Agent by default, supporting connection to local Ollama. It includes built-in environment self-check and repair, a machine transcription watermark, and flags sensitive information for verification only. Audio files never leave the local machine throughout the process, making it ideal for meeting minutes, interview summaries, and training organization scenarios where cloud uploads are inconvenient or restricted.
Read MoreCocos Preview Automated Testing: UI Inspection for Cocos Creator 3.x Previews Using BSK
Introducing the community Skill "Cocos Preview Automated Testing" (@user_4edac78e/cocos-preview-bsk-test, v1.0.3). It is based on browser-skill (bsk) to perform UI automated testing on the localhost:7456 browser preview of Cocos Creator 3.x projects: screenshot-based UI walkthroughs of each page, using evaluate to simulate real mouse events to click buttons within the canvas (bsk click cannot reach them), asserting cc.director node trees and Label text, and generating a visualized HTML report upon test completion. It is triggered when users say "test/verify Cocos game UI", "auto-screenshot to view pages", "interactive flow testing", "UI walkthrough", "click buttons to enter pages", "generate test report". No API Key configuration is required.
Read MoreCLI Configuration Assistant: Transforming Configuration Requirements into Paste-Ready CLI Commands
Command-Line Configuration Assistant (cli-config-assistant) is a community Skill that covers Linux, Windows Server, and Cisco/Huawei/H3C/MikroTik network devices. It generates complete CLI command blocks with line-by-line comments, parameter explanations, error-message analysis, and rollback commands based on natural-language requirements, spanning scenarios such as interface configuration, routing, firewall/ACL, VLAN/Trunk, SSH hardening, scheduled tasks, service management, user permission management, and log inspection. Published by @user_7a5540f4, version 1.0.1, in the dev-programming category, it requires no API key. The article provides an overview of its positioning and core capabilities, activation instructions, typical usage scenarios, and usage considerations.
Read Moredsh-BMAD-Method: Packaging BMAD-Method for Use in DSH
The `dsh-BMAD-Method` workflow plugin by `ph1996-cn` integrates the seven modules and approximately 90 skills of the BMAD-Method into DSH, eliminating the need for the `npx` installer. It supports toggling modules within profiles, configuring usernames and `_bmad-output` via the settings page, and provides tools such as `bmad_modules_list`. To install, run `dsh plugin add github:ph1996-cn/dsh-BMAD-Method`; a restart is required after installation. Use `dsh plugin update` for updates. Compatible with DSH 0.2.x and Node 22.19+/24+, licensed under MIT.
Read Moredsh-win-approval-notify: DSH Desktop Native Windows Toast Notification
dsh-win-approval-notify (npm package name) is a DSH desktop client plugin maintained by pengpenglele. On Windows 10/11, it pushes system notifications via PowerShell 5.1 WinRT Toast for scenarios such as approvals, questions, plan reviews, and task completion. Its graphical configuration takes effect immediately upon saving. Based on the GitHub README, this article describes environment requirements, three installation methods, configuration groups, self-test commands, and troubleshooting essentials. It supports only win32, performs read-only monitoring of session events, and does not participate in the approval workflow.
Read MoreAcademic Research Assistant: A Skill for Manuscript Polishing, Literature Review, and Innovation Extraction
# Introduction When writing papers, researchers often need to switch back and forth between different tools for three key tasks: polishing text, conducting literature reviews, and extracting innovation points. English polishing might be delegated to one model, while literature retrieval requires a separate search tool, and manually linking the conclusions into a comprehensive review. This Academic Research Assistant consolidates these three capabilities into a single Skill. Based on the universal SKILL.md format, it can be called directly within an Agent client, eliminating the need to piece together workflows manually. # What is this? **One-sentence definition:** A writing assistance Skill designed for academic researchers. It covers three core functions: paper polishing, literature review generation, and innovation point extraction. It supports both Chinese and English, allows for direct processing of uploaded PDF/DOC files, and can search for the latest literature when internet access is enabled. The source information is listed below and can be verified on the SkillHub page: - Name
Read Moreds-vision-skill: Augmenting Text-Only Reasoning Models with Visual Input
This article introduces the positioning and routing mechanism of ds-vision-skill: by default, `scripts/vision-router.ps1` distributes tasks by input type. Image understanding prioritizes the free race pool (GLM/Agnes), falling back to custom-1/custom-2/custom-3/local; document parsing utilizes MinerU; and pure text recognition employs Baidu OCR or Windows OCR. All underlying tools output unified standard JSON, which is then processed by the main model for inference and summarization. The article covers trigger scenarios, installation and enablement considerations, typical usage, and applicable boundaries.
Read MoreTool for Novelty and Inventiveness Search and Analysis of Mechanical Patents: Evaluation of Novelty and Inventiveness of Mechanical Field Disclosure Documents
Targeting inventions and utility models in the mechanical and structural fields, this Skill analyzes technical disclosures to automatically construct search formulas, perform patseek searches merged with Google Patents supplementary results, and conduct assessments of novelty and inventive step. Alternatively, if users directly upload comparison documents (docx/pdf/patent number lists), the Skill skips the search phase to proceed directly to analysis. The assessment is based on Article 22 of the *Patent Law* and the *Guidelines for Patent Examination* (2024). Evaluating strictly from an examiner's perspective, the Skill employs the five-element deconstruction method for mechanical engineering, one-by-one comparison for novelty, and the three-step method for inventive step, incorporating four examination standards and a five-dimensional judgment of common knowledge. It outputs a Word report with a fixed skeleton (containing a cover page, table of contents, and an abstract numbered as 0), while providing suggestions for improving the specification and annotations for items requiring manual review.
Read MorePersonal Knowledge Distillation: Turning Fragmented Conversations and Notes into Concise Knowledge Cards
Personal Knowledge Distillation (@user_fa14b5bc/knowledge-distiller-new, v1.0.0) is a knowledge-management Skill released by a community developer that addresses the issue of conversation records and notes remaining in fragmented states for long periods without being reusable. It follows a workflow of scanning, segmentation, model classification (into four categories: facts, opinions, todos, and inspirations), Jaccard deduplication, writing to the knowledge base in Markdown, and maintaining a tag index to converge existing content into structured cards. It supports incremental distillation via the `--since` flag and forced marking with `[!Distiller]`, and comes with an independent Python distillation engine featuring four commands (init, scan, commit, stats), allowing it to run independently outside of the conversation context. The article outlines the core capabilities, activation methods, and typical workflows, while noting limitations such as the classification model's dependency, lexical bias in deduplication, current installation count of 0, and community source.
Read Moredsh-computer-use-windows: Using Win32 primitives to enable Harness to control the Windows desktop
dsh-computer-use-windows is a DSH workflow plugin maintained by pdong15dth that provides 12 computer_* tools utilizing Win32 primitives such as SendInput, GDI+, and EnumWindows. These tools are executed by a native helper, compiled and cached upon first use, without relying on UI Automation or SendKeys. The article details the tool list, screenshot coordinate mapping, installation via install_bundle and the marketplace index, helper compilation, and environment variables, as well as security constraints regarding permissions, focus, and the clipboard.
Read MoreTomato Novel Creative Companion: An AI Skill Covering the Web Novel Journey from Zero to Signing
Fanqie Novel Creative Partner is a general-purpose Agent Skill in SKILL.md format designed for online fiction creation. Located at @user_78e5456d/novel-writing-partner, it is at version 1.0.0. Sourced from the community and categorized under content-creation, it is published by user_78e5456d with the tag requires_api_key=false. It positions itself as a full-process creative companion from inception to contract signing. Based on the verified introduction, it offers capabilities including market analysis and reader profiling, immersive scene assistance, viral work breakdown and style imitation, practical exercises with automatic correction, removal of "AI flavor" (de-AI detection), AI rate detection, cover design, contract assessment, and promotional copywriting. Trigger conditions include writing web novels, outlining, de-AI flavor, AI rate detection, cover design, contract evaluation, creating promotional copy, dissecting viral works, or imitating styles. The page metrics show a rating of 100,000, 1,139 downloads, and 0 installations. As the materials do not provide installation commands, directory conventions, or configuration fields, the text does not include unverified commands; instead, it provides the official address and SkillHub page, noting that differences regarding Cursor, Codex CLI, and Claude Code cannot be verified. Typical usage involves reproducible call examples focusing on outline conception, breakdown and imitation, de-AI flavor, covers, contract assessment, and promotional copy. Precautions include adherence to platform standards, sample dependency, AI rate reference only, contract assessment not being a guarantee, covers being descriptions or prompts, and the need to self-verify unspecified items such as update frequency.
Read MoreThe Design Media All-rounder: Integrating AI Image Generation, Editing, Subtitling, Voiceover, and Delivery into a Single Workflow
The Design Media All-Rounder is a design-media category Skill published by SkillHub community author user_ff29c4df, version 2.4.0, sourced from the community, with requires_api_key=false and pricing_type=alipay. It covers eight modules: AI image and 3D generation, video editing, audio, digital human broadcasting, UI design, material assets, and delivery, excluding film-grade production. The core consists of a five-element prompt formula and a platform size specification table, enabling users to obtain copyable prompts, editing workflows, and multi-platform size schemes. It also includes compliance guidance on copyright and portrait rights for commercial use, obligations for labeling AI-generated content, and the timeliness of platform regulations. The materials provide three typical input scenarios: creating multi-platform-sized posters, generating stable image prompts, and producing digital human broadcasting videos. This article describes the core functionality, acquisition methods, typical usage, and applicable boundaries.
Read MoreGlobal Persona (dsh-plugin-global-persona): Inject a shared editable AI persona into all sessions of the current Profile.
dsh-plugin-global-persona, maintained by Patrick126090380, injects the same persona setting into every conversation in the current DeepSeek Harness profile. The persona setting can be edited on the plugin page, takes effect in the next round after saving, and includes two fields: enabled and persona. Empty content is not injected. Supports adding plugins locally via ZIP, requires DSH 0.2.x, MIT licensed, purely local with no network connectivity required.
Read MoreGeneration of Gantt Chart Project Plans: Converting Project Nodes and Timelines into a Planning Framework
'Gantt Chart Project Plan Generator' (@kunlungrowth/gantt-chart-gen, v1.0.0, Kunlun Growth, business-ops, source enterprise, no API Key required) accepts project nodes and timelines, and outputs text-based Gantt charts, milestones, risk annotations, and resource allocation suggestions. Trigger words are Gantt chart, project plan, milestone, and project schedule. This article verifies the official introduction and meta-information. Due to the failure to retrieve the official page, the installation command, configuration fields, loading differences across clients, and official examples have not been verified. The main text describes this as-is and refers to the official address. Additionally, it includes a minimal reproducible input example. The retrieved meta-information shows a rating of 100000, downloads of 1243, and installations of 0. Citations should refer to the official page. This tool is intended for project managers who need to quickly build a planning framework during the project initiation phase. The output is text-based and still requires human judgment for risk and resource suggestions, which makes it unsuitable for direct use as formal reporting charts.
Read MoreImage Compression Skill: Unified NX API Compression for Local, Folder, and Remote URL Inputs
Image Compression (Image Compression) is a community-sourced Skill (@user_26f7b237/nx-image-compress, Version 1.3.1, Category: design-media) that consolidates three input types—local image paths, batch folders, and remote URLs—into a single conversational workflow. The NX MCP server handles the compression, returning CDN addresses and compression ratios, and supports options to overwrite original files or save as new ones. The article introduces core capabilities, trigger scenarios (image compression, image optimization, size reduction, TinyPNG, JPG/PNG/WebP compression), installation and enablement methods (requiring prior API Key configuration with `requires_api_key=true`), typical usage examples, as well as precautions regarding data transmission, remote URL fetch failures, and supported formats.
Read Moredsh-skill-experts: Manage sidebar skill experts and summon them in conversation
dsh-skill-experts is a community plugin for the DSH client maintained by ownduck (@woosau/dsh-skill-experts, MIT). It allows users to create, edit, enable/disable experts in the sidebar, configure descriptions, personas, and skills, and supports import/export. Experts can be summoned via the Composer's "Skill experts" or using "@". The "Continue" function defaults to continuing the context of the same expert, but users can also set specific session Agents. It comes with four built-in experts, including Frontend Developer. To install, use `dsh plugin --profile desktop/web add` (requires dsh >= 0.2.0); open the "Skill experts" sidebar after restarting. This is a non-official plugin that runs with the permissions of the current dsh process; please review the source code before installing.
Read MoreAI逍遥游 · Book List Viral Video Production: Step-by-Step Generation of Vertical Book List Videos from a Single Book
Introducing the community Skill "AI逍遥游 · 书单号爆款视频制作" on SkillHub. After entering a book title, it executes in SOP phases—planning → image generation → voiceover → compositing → delivery—confirming with the user at each phase before proceeding to the next, and produces vertical book list videos suitable for Douyin/Kuaishou/Video Account (WeChat Channels)/Xiaohongshu. Trigger words include 书单号, 书单视频, book video, 图书推荐视频, etc. Version 1.0.0, category content-creation, source community, requires API Key configuration (requires_api_key=true). The page displays a rating of 100000, 1223 downloads, and 0 installations. Since the scraped page did not provide installation commands or configuration fields, the article reports this as-is and defers to the official page.
Read Moredsh-plugin-novel-reader: DSH Status Bar Overview and Sidebar Novel Reading
dsh-plugin-novel-reader is a DSH client plugin maintained by overflow7782. It provides a book icon and a frosted-glass quick-view window in the conversation.composer.dock, and its right-hand sidebar “Novels” tab supports a bookshelf for txt/epub files and table-of-contents reading. Requires DSH 0.2.1+ and Node.js 22+; install via `dsh plugin add`. Data is stored locally in IndexedDB; the plugin does not connect to the internet. Includes a boss key and top-bar collapse, and integrates with dsh-better-sidebar and dsh-plugin-wallpaper-engine.
Read MoreOffice Automation: Transforming Highly Repetitive Tasks into Operational Automation Solutions
**Office Automation (@kunlungrowth/kunlun-office-auto)** is a skill for generating automation solutions tailored to enterprise office scenarios. Released by Kunlun Growth, it is version 1.0.0, sourced from enterprise, categorized under office-efficiency, and does not require an API Key. Its workflow comprises three steps: first, identifying high-repetition office tasks; second, designing a solution using a trigger-action-fallback structure; and third, outputting a runnable plan—such as scripts, integration configurations, or low-code setup steps. Unlike generic code-generation skills, it treats fallbacks and trigger conditions as first-class components of the solution. The article covers its metadata, core functions, installation items to confirm, typical usage organized around the three-step process, applicable scenarios, three key considerations (triggers are verifiable, fallbacks map to concrete actions, and permissions and data scope are defined), as well as its official address.
Read MoreTop Real Estate Sales Script Generation (6 Scenarios): Deconstructing Sales Pitches into 6 Reusable Steps
Top Sales Script Generator for Real Estate (6 Scenarios) is a Skill for real estate sales teams, located at @user_c0d139fd/sales-script, version 1.0.1, source: community, category: business-ops, tag: requires_api_key=false. It breaks down communication along the sales process into six scenarios: openers, needs discovery, selling points presentation, objection handling, closing, and referral. Needs discovery is built with the SPIN questioning method; selling points are organized according to the FABE framework; objection handling includes over 20 common objection scripts. The scripts are conversational and interactive, with each scenario kept within 800 Chinese characters. It supports customization for four customer segments—first-time essential buyers, upgrade buyers, investors, and high-net-worth clients—and fits three channels: phone, WeChat, and on-site conversations at the sales office. The article explains what it is, what it can generate, where the entry point is, and the issues that must be addressed when using it: fact checking, scenario coverage, and reviewing and proofing community-sourced content.
Read MoreDSH Home Sync: Syncing DSH Configuration and Mnemon Memory Data with Git
dsh-home-sync is a memory-related plugin for DSH maintained by OverDustD7. It synchronizes the web configurations in the DSH home directory and the syncable Mnemon files via Git bidirectionally. It offers Pull, push-to-remote, scheduled sync, conflict detection, a Chinese-language web UI, and creates backups during device initialization. This article, based on the official README, introduces the synchronization scope, installation commands (npm/GitHub/link/offline tarball), the meaning of the initialization merge mode, as well as push auditing and permission considerations, to facilitate multi-device use and migration scenarios.
Read MoreGlobal Legal Search and Compliance: Skill for Regulation Search and Compliance Queries Across 230+ Jurisdictions
Introducing the community Skill "Global Legal Search & Compliance" (@user_cd0fcd93/global-legal-search, v1.10.3). It is positioned to provide regulatory retrieval and compliance queries across 230+ jurisdictions, with trigger words covering overseas compliance, cross-border investment, data protection, GDPR, AI Act, HIPAA, CCPA, export controls, and security framework mapping (ISO 27001 ↔ NIST ↔ SOC 2). Core capabilities include one-click installation of dedicated libraries for 53 countries, with dual-source online verification for other jurisdictions. In hosted environments, it connects directly to installed MCPs (e.g., Meta-Juris, PKULaw), while in local CLI environments without MCPs, it automatically degrades to an offline statute library plus WebSearch for instant answers. Version 1.10.3 adds a unified FAQ summary, a five-step verification for niche jurisdictions, hybrid query routing, three-segment endnotes for offline snapshots, and a capability self-test with minimalist fill-in-the-blank templates. Official documentation provides no independent installation commands or configuration fields; verification confirms it uses the standard SKILL.md format and requires no API Key. Therefore, the main text only explains activation based on the environment and notes that this Skill is a retrieval entry point rather than legal advice.
Read Moredsh-experience-recall: Injecting cross-project experiences from memory into DSH at the right time
dsh-experience-recall is a DSH memory-type plugin maintained by OverDustD7: it monitors reasoning and tool events within sessions, matches against a memory bank and a document-derived vocabulary, and after gating uses a local Ollama small model to adjudicate whether to inject short experience cards before the next step. It requires DSH ≥ 0.1.5-rc.1 and Node ≥ 22.5, and is recommended to be used in conjunction with dsh-mnemon and the mnemon CLI. Install with: `dsh plugin --profile web add dsh-experience-recall`; verify via `/experience-recall` and `observe.ndjson`. Released under the MIT license; current npm version is 0.8.2.
Read MoreExperts in the Review of the "Double High" Initiative for Vocational Education: Transforming "Thick Dossiers with No Impression" into Actionable, Point-by-Point Feedback
This is a Skill designed for the review of application documents for the "Double High Plan" (High-level Vocational Schools and High-caliber Construction Programs) and similar high-quality development projects for vocational colleges. Maintained by community publisher user_63830f0e, it is categorized as professional, requires no API Key, and is currently at version 1.0.7. After uploading construction plans, application proposals, or research reports, the tool follows the review panel's approach to perform four key tasks: 1. **Error Detection and Annotation:** It identifies errors and marks them in red line-by-line. This includes inconsistencies across the three documents, internal contradictions, failure to convert research problems into actionable tasks, tasks lacking performance indicators or budgets, and outdated policy statements. 2. **Policy Alignment:** It anchors the content to the province's "15th Five-Year Plan" and creates an explicit mapping from policy provisions to specific measures in the documents. 3. **Differentiated Positioning:** It identifies a unique positioning from six perspectives: industry, talent supply and demand, job competencies, AI development, provincial competition, and national strategy. 4. **Quality Improvement:** It highlights strengths, removes "AI traces" from the writing, and simulates scoring by three types of experts (industry, vocational education, and finance). The tool outputs either a red-line comparison table or a revised DOCX file while preserving the original formatting. The report allows clickable navigation to highlight specific issues in the original text with corresponding page numbers. It features automatic degradation in case of file access errors or network failures. The package includes beginner guides, usage examples, and an FAQ, making it suitable for national/provincial Double High applications, mid-term inspections, and acceptance checks.
Read MoreLifestyle Radius Search Assistant: Transforming “Living Convenience” into Verifiable Search Criteria
"Living Circle Housing Assistant" was published by SkillHub user user_82787961. Sourced from the "community" category and classified under "life-service," it is currently at version 1.0.1. This tool treats daily needs—such as grocery shopping, dining, childcare activities, medical access, and commuting—as the core constraints for property searches. It first utilizes the Beike CLI to retrieve actual listings of second-hand homes and residential compounds. Subsequently, using the candidate property's coordinates as the center point, it invokes Tencent Maps to verify surrounding Points of Interest (POIs) and walking distances. Based on transparent weighting rules, it calculates a "Living Convenience Score" (0–100), explicitly marking whether each criterion is met or failed, and highlighting major deficiencies. Finally, it generates an HTML comparison report featuring maps and schematic diagrams of the living circle. This assistant is ideal for family home purchasing, multi-property comparisons, and pre-screening for real estate agents before property tours. Configuration of a Beike API Key is required for use, while a Map Key is utilized for the verification of surrounding amenities.
Read More