dsh-local-library: Local repository for cross-session use by Agents
dsh-local-library is a DSH Desktop memory plugin maintained by Fucaplos. It stores files in the local `local-library` directory. Its settings page supports uploading, tagging, searching, and packaged downloads, and it registers `library_list`, `search`, `read`, and `save` with the Agent. It can extract text from PDF and Office documents for retrieval and optionally uses visual models to describe images. On Windows, `install.ps1` is used to install the junction and `cordis.patch`. It is licensed under MIT.
Read MoreMr. Review: Turning Transactional Recaps into Next Steps
Mr. Retrospective is a post-mortem Skill with the persona of a traditional accountant, posting the books for projects, cycles, and activities. The core proposition is that the value of a retrospective lies not in summarizing the past, but in rewriting the next course of action. Its core craft includes the five-step retrospective method (quantify the objective, record only facts about the outcomes, compute the deltas, attribute causes by separating controllables from uncontrollables, and define the next step as a single change with a deadline), the Three Whys of attribution, the Three Abstinences, three retrospective templates (Five-Step Project Review, Three-Question Cycle Review, Activity Funnel Review), and a one-page, five-cell retrospective. It also has a built-in machine check to flag bogus moves such as making a ledger-style narrative, offering empty next steps, and blame-shifting attribution. Red lines: audit events, not people; where data is missing, mark it as (no data) and do not fabricate; change only one thing at a time; never turn retrospectives into blame sessions. Note: the official page failed to fetch this time; the installation command could not be verified. No fabrication has been made.
Read MoreInternal Audit of Management Systems: A Four-Phase Guide to Completing Audit Documentation
# Introduction: Internal audits of a Quality Management System require the production of four types of documents: audit plans, implementation schedules, departmental checklists, and audit reports. These documents have fixed formats and contain numerous clauses; creating them from scratch often leads to missed items and difficulties in aligning with standards such as ISO 9001 and IATF 16949. This article introduces the "Internal Management System Audit" Skill on SkillHub. It breaks down the internal audit into four stages, utilizes a Q&A method to collect information, and generates ready-to-print TXT and MD documents. It is ideal for those who need to perform system internal audits but wish to avoid manually organizing templates. ## What is this? The "Internal Management System Audit" is a full-process interactive internal audit assistant located at `@engicool/internal-management-audit`, developed by "Engineering...
Read More**Comparison of International Logistics Quotes and Checking for Hidden Costs: Standardizing Multiple Freight Forwarder Quotes**
This is a generic SKILL.md format Skill from the SkillHub community, located at @user_a8e9072b/compare-international-freight-quotes. It is Version 1.0.0, categorized under business-ops, published by user_a8e9072b, and tagged with requires_api_key=false. It performs three main functions: normalizing quotes from multiple sea, air, rail, or express carriers into a comparable format; summarizing known fees; and checking for omissions against a checklist covering items such as Port of Loading, Port of Discharge, local charges, fuel surcharges, peak season surcharges, documentation fees, customs inspection fees, detention and demurrage, delivery, and taxes/duties. It is suitable for selecting freight forwarders, reviewing logistics costs, and explaining quote discrepancies to clients. It does not provide real-time rates, nor does it guarantee that unquoted fees will not occur. The scraped page did not provide installation commands, directories, or configuration fields; none have been fabricated here, so please refer to the official page for installation instructions. The rating of 100000, 1028 downloads, and 0 installs reflect data at the time of scraping.
Read Moredsh-sidebar-home: Sidebar Workspace Home and Focus Mode
dsh-sidebar-home is a DSH client plugin maintained by fu-silent (version 0.1.0, MIT license). It adds a Home entry in the sidebar: after overviewing workspaces, users can enter Focus mode, which displays only the current workspace session and collapses other workspaces. Upon exiting, the default sidebar is restored. It supports both Chinese and English, with layout and focus state persistence via localStorage, and has zero dependencies with no telemetry. Installation can be done by adding the path or github:fu-silent/dsh-sidebar-home via the Desktop Plugins page, or by adding the absolute path under the dsh web profile. Requires DSH version >=0.2.1-rc.5 and <0.3.0-0, and has been tested on macOS Desktop 0.2.1-rc.5.
Read MoreOne-Page Brand Plan: Integration Skills for Annual Pharmaceutical Brand Planning
Developed by the Baitong AI Service Platform, the "Brand Plan on a Page" synthesis skill for the pharmaceutical industry (Coordinates: @baitongai/brand-plan-one-pager) is designed based on the STEPS framework. Adopting the perspective of a senior pharmaceutical brand planning integration consultant, it consolidates outputs from market insights, strategy, positioning, content, execution, finance, and monitoring into a logically coherent, actionable, and presentable annual brand plan. The skill embeds the Uptake six-part standard structure, ten major brand planning pitfalls, strategic task writing methodologies, four-level objective breakdowns (Business/Brand/KPI/Initiative), a three-column information architecture for the one-pager, and a global-local cascade (Single Source of Truth) mechanism. It also defines interfaces with upstream skills such as review, SWOT analysis, positioning, budgeting, and KPI setting. This skill is suitable for five scenarios: finalizing annual plans, cross-regional and cross-brand standardization, extracting one-page summaries for executive reporting, adapting global plans for local markets, and performing plan quality self-checks. No API Key is required.
Read MoreInvestment Promotion Script Design | Guo Yiyi: Turning Investment Materials into Presentation Scripts, Q&A, and Screening Questions
This article introduces the community source Skill "招商话术设计｜郭亿易" (Investment Recruitment Script Design | Guo Yiyi) on SkillHub, located at @user_8d36cde0/guoyiyi-partner-recruitment-script (Version 1.0.0). It accepts four types of inputs: project mode, partnership benefits, thresholds, and risks, and generates three deliverables: an investment pitch script, a FAQ list, and qualification screening questions. It is suitable for channel recruitment and BD teams. Official documentation does not provide installation commands; however, the tag `requires_api_key=false` has been verified, indicating that no additional API Key is needed. Based on the standard SKILL.md format, it can be recognized by placing the SKILL.md file in an Agent environment that supports this format. Official specific steps for Cursor, Codex CLI, and Claude Code are not provided, and this article makes no speculations regarding them. The typical usage involves organizing materials according to the four inputs and submitting them to the Skill to generate the three types of outputs. Since no verbatim examples are provided by the official source, the article presents the input structure rather than output samples. Key points to note: input quality determines output quality; the risk section must not be omitted; and qualification screening questions serve only for preliminary screening and cannot replace formal qualification reviews. The article also points out a discrepancy between the task clues (Rating: 99253, Downloads: 930) and the retrieved data (Rating: 100000, Downloads: 1026, Installs: 0), suggesting that the real-time display on the SkillHub page should be taken as the standard.
Read MoreQCC Report Generation Skill: Organizing QCC Results Report with Ten Major Steps and Seven Major Tools
The QCC Report Generation Skill (@user_2241deda/qcc-report-generator, v1.0.0, community source, office-efficiency) is based on the "Ten Steps of QCC" and "Seven QC Tools." It organizes user-inputted information—such as topic background, current status, objectives, root causes, countermeasures, and results—into a QCC achievement report suitable for presentation and review. It applies to scenarios such as QCC activity summaries, competition reports, and achievement demonstrations. No API key is required, making it ideal for groups that possess activity records but lack a unified reporting structure.
Read Moredsh-sidebar-entries: Make the DeepSeek Harness left sidebar behave like a proper long list
dsh-sidebar-entries is a DeepSeek Harness client plugin (v0.7.16, MIT) maintained by FragrantLotus. It enhances the left sidebar via DOM and localStorage: long-press drag-to-reorder for panels with one-click collapse, a 30% cap on full-row height, pinned conversations grouped together mirroring the running/pending-reply status, drag-and-drop for conversation view tabs, archived conversations sinking to the bottom, footer entries promoted onto the same track, and forked child conversations following their pinned parent conversation. Installation is recommended via `dsh plugin --profile web add dsh-sidebar-entries`; restart the Web app and then refresh the page. The plugin runs with the permissions of the current DSH process, so it is advisable to review the source code and license before installing.
Read Moredsh-session-delete: Permanently delete sessions in the DSH Web sidebar
dsh-session-delete, a DSH memory-class plugin (MIT, v1.0.0) maintained by Fragmented-Chaos, provides a "Delete Session" entry for the Web sidebar. On the host side, it first rejects running sessions, then archives and deletes directories under $DSH_HOME/sessions, and broadcasts an api-session/removed event. To install, execute the plugin by passing the identifier Fragmented-Chaos/dsh-session-delete to DSH via SkillHub. Please review the source code before use, as deletion is irreversible.
Read MoreWeChat Mini Program Skyline Style Guard: Check Against a Checklist Before Writing WXSS
Introducing the SkillHub community Skill "WeChat Mini Program Skyline Style Guard" (@user_4edac78e/taro-skyline-wxss, version 1.0.1). Designed for writing and modifying WXSS in Taro + WeChat Mini Program Skyline rendering projects, it organizes content into four categories: Skyline style support blacklists (radial-gradient supports only 'circle', sizes support only 'px', background-image accepts a maximum of 2 values, box-shadow does not support stacking, backdrop-filter/filter supports only single functions, no inline-block, no attribute selectors); default value differences (border-width defaults to 3, box-sizing defaults to border-box, display defaults to flex); layout and scrolling traps (misalignment of the first line with flex-wrap+gap, scroll-view offset reset by setData, horizontal scrolling scroll-into-view failing and requiring controlled scroll-left with alternating 1px offsets, scroll-view clipping box-shadow of full-height children, scroll-into-view remaining permanently active); and an SCSS compilation validation script independent of devtools (including automatic assertions for gradient violations). The article provides trigger words, typical reproduction scenarios, and applicable boundaries, noting that the installation command could not be verified and the official page should be treated as the source of truth.
Read MoreSocial Subtext Analyzer: Decoding the Speaker's True Meaning
The Subtext Decoder (coordinate @user_3a752629/subtext-decoder) is maintained by SkillHub community publisher user_3a752629. Classified under "life-service" with a community source, version 1.0.0, and tagged with requires_api_key=false. Designed for various social relationships—including family, superiors, colleagues, friends, and partners—it translates spoken words into their true meanings and underlying demands, while providing direction for responses. Trigger scenarios cover workplace rhetoric interpretation, family relationship discourse analysis, couple communication decoding, subtext analysis for social media posts and WeChat messages, as well as judgment requests such as "Is this statement emotionally intelligent?" or "Is there an issue with this person's speech?". The documentation does not provide installation commands, directory conventions, or configuration fields, nor are there official reproducible examples; the text explicitly notes this and recommends enabling the skill via the entry points of the host environment (Cursor / Codex CLI / Claude Code) using the provided coordinate. It is important to note that users must provide relationship background information themselves, and the tool offers common interpretation directions rather than definitive conclusions. Furthermore, with an installation count of 0 in the site data, its actual effectiveness lacks corroboration from user feedback.
Read MoreOne-Click Text Recognition for PDF Scans/Images: A Dual-Engine Solution for Local Offline OCR
"One-click PDF/Image to Text" is a local, offline OCR skill (ID: @user_b698fb2a/local-doc-ocr). It is a community-sourced tool (Version 1.0.1) categorized under office-efficiency. It offers two engines: Engine A utilizes RapidOCR for one-click conversion of PDFs/images to Markdown locally, featuring pure offline processing, no network connection, zero token consumption, and batch processing capabilities. Engine B employs a multimodal image-reading Agent to process pages individually, offering higher accuracy in table reconstruction, handwritten signatures, stamps, and blurry images. Both engines incorporate built-in preprocessing such as EXIF orientation correction, deskewing, grayscale enhancement, and sharpening. The output is saved as Markdown files, organized by page and image with paragraphs separated, including page number markers and manual verification prompts. The entire process is handled locally without external API calls or cloud uploads, making it suitable for classified and private materials. No API key is required. The skill is activated via trigger words. Note: This skill is strictly for text extraction; requests regarding "recognition hazards" or "security risks" belong to the security inspection category.
Read MoreAI文生图：把复杂信息画成手绘涂鸦知识卡片
# 前言 把一段财经知识、产业链分析或读书笔记变成一张能直接发出去的图，通常要经历「整理要点 → 选排版 → 找插画 → 调尺寸」四步，最后拿到手的还常常是模板感很重的成品。AI文生图 Skill 把这件事压缩成一次调用：你给出内容，它输出 3:4 竖版的手绘涂鸦风格（Sketchnote）知识卡片图片，并按内容量自己决定出几张、用哪套构图和视觉风格。下面介绍它的定位、能力和使用方式。 ## 这是什么 AI文生图是一个社区来源的通用 SKILL.md 格式 Skill，坐标 @user_33408c69/981028-2，由发布者 user_33408c
Read MoreFrontend Engineer Skill: Delegating Frontend Implementation Discipline to Agents
This Community Edition Skill addresses the issue of inconsistent frontend implementation quality. It triggers in four scenarios: interface implementation, responsive layout design, performance optimization, and cross-browser compatibility. It requires avoiding shortcuts, adhering to best practices, and designating accessibility and performance as non-negotiable baselines. It explicitly rules out three types of excuses: time pressure, assumptions about modern browsers, and "good enough" compromises. Version 1.0.0, sourced from the community, does not require an API Key. This article explains its positioning, trigger scope and constraints, and verified coordinate information. It also points out that official providers do not offer installation commands or reproducible examples, and recommends using it in conjunction with project specifications.
Read MoreC Drive Space Cleanup: Securely Assess and Reclaim Local Storage
C Drive Space Cleanup is a community Skill designed for Windows 10/11 and macOS, located at @user_6113fd9f/safe-storage-cleaner. It is categorized under it-ops-security, is Version 1.1.0, and requires no API Key. It features a dual-engine design: the PowerShell engine audits reclaimable space on the Windows system drive based on whitelist categories (user/Windows temporary files, browser cache, Recycle Bin, Windows Update and Delivery Optimization cache, error reports, thumbnail and shader caches, diagnostic dumps, pip/npm/uv and NVIDIA caches, Squirrel temporary files, DISM component store, and project artifact manifests); the Python engine performs read-only scanning of file system metadata to generate a Green/Yellow/Red plan, identifying data from AI tools such as ChatGPT, Codex, Claude Code, WorkBuddy, Cursor, Ollama, Hugging Face, and LM Studio for isolated handling. The security model defaults to scanning and reporting only; cleanup operations require explicit item-by-item approval from the user, with a mandatory -ConfirmClean double confirmation required on the Windows side. Trigger words cover Chinese and English disk space expressions, but it is not applicable to requirements involving RAM, antivirus, SMART, data recovery, servers/NAS, permanent deletion, or simple application uninstallation.
Read MoreDeep Dive into UI: The Skill to Block "Good Enough" Before Delivery
UI Deep Exploration (@user_3c47cf70/impeccable-pro, v1.0.0, community-sourced, dev-programming category) is an execution-driven Skill that demands zero-defect standards and rejects "good enough" compromises. Its core capability lies in executing rigorous quality checks, ensuring standards do not drift under time pressure and resource constraints. This article provides its verified metadata, an actionable three-step inspection workflow, coordinate-referenced activation methods, typical use cases, and applicable scenarios. It also highlights cautionary notes, including its community origin, low installation volume, and inconsistencies in rating and download data.
Read Moredsh-plugin-ai-video: Implementing the "Outline to MP4" Video Pipeline in DeepSeek Harness
flynow-harness's dsh-plugin-ai-video provides a workflow plugin for DeepSeek Harness 0.2: the Agent generates the outline and storyboard, while the plugin handles TTS, image generation, subtitles, preview.html, and MP4 export via ffmpeg. It supports image/html modes, three visual styles, aspect ratios, and incremental editing; built-in title cards and system voices are available when no API is accessible. Based on the official README, this article details installation commands, configuration, a tool table, and usage examples, while outlining plugin permissions, dependencies, and known limitations.
Read Moredsh-skin-collection: Provides multiple switchable skins for DSH Web GUI
dsh-skin-collection provides 9 independent skins (wabi-sabi, terminal, cyberpunk, etc.) for DeepSeek Harness Web GUI, each containing scoped decorative CSS and --dsw-alias-* tokens. Switch via the sidebar "Skins" or Settings → Appearance, with the default style restoring the native UI. The plugin persists the current skin using body[data-dsh-skin-*], separate from the official light/dark/system preferences. Installation: dsh plugin --profile web add dsh-skin-collection, then restart dsh web. MIT licensed, pre-built distribution with no build step during installation.
Read MoreSlack (@steipete/slack): Enable Agents to Read and Write Slack
@steipete/slack is an instant messaging integration Skill based on the SKILL.md format, published by steipete (Peter Steinberger) on clawhub. It is currently at version 1.0.0, licensed under MIT-0, and categorized as an ai-agent. It enables Clawdbot to control Slack via the slack tool, covering five default enabled action groups: reactions, messages, pins, memberInfo, and emojiList. It allows for replying to messages in channels or direct messages, pinning/unpinning items, reading, editing, and deleting messages, and querying member information. The tool utilizes the host's already configured bot token; input only requires channelId, messageId (Slack message timestamp format), emoji, or the 'to' target. The installation method is via `npx skills` or `openclaw skills install @steipete/slack`. This article provides examples of all action calls as described in the official documentation and points out notes regarding credential dependencies, messageId format, and the syntax for the 'to' target. The SkillHub mirror page shows a score of 100000, 73009 downloads, and 4231 installations; the upstream ClawHub page shows v1.0.0 and a Security audit Pass. Statistics are based on real-time data from the official pages.
Read MoreMachine Learning MCP Server: Breaking Down the Classic Modeling Workflow into 30 MCP Tools
A community-sourced MCP skill located at @user_053d27d0/mcp-ml-toolkit-zx, version 1.0.1, categorized under dev-programming, sourced from the community, published by user_053d27d0, with a declaration of requires_api_key=false. It consolidates 30 MCP tools covering classic algorithms and evaluation/tuning workflows such as logistic regression, KNN, SVM, decision trees, random forests, naive Bayes, linear regression, Ridge, Lasso, K-Means, DBSCAN, PCA, t-SNE, cross-validation, ROC, and grid search, targeting structured tabular data. The article follows a fixed structure to introduce its positioning, core features, installation and activation, typical usage, and applicable scenarios. It also notes that the currently scraped materials lack installation commands and tool input parameter definitions, so the official page should be treated as authoritative.
Read MoreThe Dianping Restaurant Planning Master: Converting restaurant planning into an executable checklist Skill
This is a strategic planning Skill tailored to restaurant store scenarios on Dianping . It is located at @user_620ecec4/dianping-restaurant-marketing, version 1.0.0, sourced from the community, categorized as business-ops, and tagged as requires_api_key=false . The article covers its positioning, verifiable core metadata, and known versus unverified details regarding installation and activation . The article cautions readers, however, that the official page could not be accessed for this review; consequently, the full body of SKILL.md, the feature list, installation commands, and typical examples remain unavailable, some numerical metrics are contradictory, and readers should verify the official page before implementing it for production use.
Read MoreFirst-Class Constructor AI Assistant: A Multi-Role Skill for Exam Preparation and Work
The First-Class Constructor AI Assistant is an Agent Skill released by Guizhou Hongnan Network Technology Co., Ltd. Based on the SKILL.md format, it covers scenarios related to exam preparation and professional work for construction builders. It features six built-in expert personas that can be switched and coordinated via instructions to complete tasks, without requiring an API Key. This article introduces its positioning, core features, installation and activation methods, and usage notes. The official link is https://api.skillhub.cn/hongnan/zaojizhe-builder-team, where ratings and details can be found on the SkillHub page.
Read MoreThe Best Companion for MCGS: Configuration Generation Skill for MCGS Configuration Engineers
Introduce the community skill on SkillHub: “MCGS Best Partner” (located at @user_51a923b7/mcgs, version 1.0.10, category dev-programming, rating 100000, downloads 1096, installs 0, tag requires_api_key=false). This skill is aimed at PLC programming and configuration engineers, and targets common MCGS workflow pain points—cumbersome device/variable configuration and unfamiliarity with screens and animations—by providing directly actionable configuration generation, template outputs, and troubleshooting checklists. The article explains the source attribution, verifies the core capabilities, and provides the official access entry point. It also notes that the retrieved materials do not include installation commands or reproducible examples, so no commands are fabricated. Finally, it outlines the target audience and limitations: generated results must be manually reviewed; the applicability of templates depends on the MCGS version; as it is a community-sourced skill with zero installs currently, it is recommended to validate it first with a small project.
Read Moredsh-session-overview: Cross-Workspace Recent Sessions Overview
DSH client community plugin dsh-session-overview from Flyeasy provides Recent sessions and All sessions panels in the sidebar, aggregating recent sessions across all workspaces from sessions.list and workspaces.list and distinguishing sources with ten-color workspace badges derived from path hashes. It supports search, pagination, workspace filtering, and a native-consistent context menu. Installation: dsh plugin --profile <your-profile> add dsh-session-overview. Requires the web client. The README states it is developed on Harness 0.2.1-rc.5 and is MIT licensed.
Read More