Fapuxiaoxiang · Legal AI Assessment: H5 Generator for Legal Risk Assessment
Fapu Xiaoxiang · Legal AI Intelligent Assessment is a Skill that transforms legal liability analysis into interactive H5 content. It is located at `@user_be98013b/fapuxiaoxiang-legal-analyzer`, with version 2.2.0, sourced from the community, and categorized as professional. It offers three core capabilities: questionnaire-based risk assessment, citation of real judicial precedents, and long-image sharing to social media. It covers legal knowledge scenarios such as alcohol-related disputes, contract disputes, and tort liability. It can be triggered by keywords such as "Fapu Xiaoxiang," "Pufa H5," or "Responsibility Analysis Tool." Distributed based on the standard SKILL.md format, its metadata includes a `requires_api_key=true` tag, meaning an API Key must be prepared before activation. However, official pages do not specify how to obtain the Key or where to input it, nor is documentation available regarding integration differences across hosts. The typical usage involves invoking the tool via trigger words in a conversation, providing scenario details, and exporting a long image after the H5 is generated. It is suitable for developers or content creators who need to rapidly produce interactive legal knowledge pages. Please note that this is a community release with zero installations; legal accuracy regarding precedents and liability conclusions should be independently verified before external publication, and this tool does not constitute legal advice.
Read MoreClawHub Daily: Automated Daily ClawHub Skill Recommendation Briefing
ClawHub Daily (Coordinate: @indiv-ai/clawhub-daily-ai, Version 2.0.9, Category: ai-agent, Maintained by Community Publisher) solves the pain point of high costs associated with manually browsing leaderboards to filter from thousands of Skills on the ClawHub platform. It daily scrapes the Top 500 Skills (based on the Convex API with 0 token consumption) and generates comprehensive recommendations across six dimensions: trending, quality, newcomers, panorama, actively_maintained, and verified, producing briefings of 8 valuable and non-duplicate Skills. When candidates are insufficient, it automatically relaxes thresholds; if there are no candidates for the 'verified' dimension, it skips it without consuming quota. It incorporates a library of 7 major scenarios for pain-point weighting and performs cross-dimensional deduplication over 7 days, ensuring a Skill is displayed only in the first scenario it matches. Entries with changelog summaries display "Recent Changes." Briefings consist of a one-sentence summary in Chinese with the original English text collapsed. Multi-channel delivery is supported via Feishu Cloud Documents and Card Messages, Tencent IMA Knowledge Base (optional), Obsidian vault, and local data/recommended/*.md files, with independent exception handling ensuring that failures in one channel do not block others. Credential priority is CLI arguments > environment variables > references/config.json; if no credentials are configured, it only generates local files without performing external pushes. Permission declarations include network and filesystem access, explicitly stating it does not read unrelated environment variables such as GH_TOKEN or GITHUB_TOKEN. It is suitable for users who want to receive daily scheduled briefings, find Skills based on specific pain-point scenarios, and track ecosystem trends. Please note that recommendations are fixed at 8, IMA Method B requires the pre-installed ima-skill CLI, and the default Obsidian path follows Windows formatting.
Read Moredsh-auto-continue-community: Auto-continuation for local model max-token truncation
Introducing the community plugin @quekenao/dsh-auto-continue-community (v0.3.2). When the DSH Web client works with a local llama.cpp model and responses are truncated due to max-tokens, this plugin transparently issues follow-up model calls, gated by provider and governed by a continuation budget. The default loop mode performs real-time continuation via agent.steer(); the stream mode serves as a fallback that buffers and merges outputs. Includes official installation commands, a cordis config example, and a clarification of the supported scope validated only for DSH 0.2.1-rc.5 with the Web stack.
Read MoreIntelligent Structural Parsing of Drug Instructions: Transforming Drug Instructions into Four-Tier Traceable Standardized Data
This article introduces the **Intelligent Structured Parsing Skill** for drug labels available on SkillHub, a tool that parses PDF, Word, and image-based drug instruction sheets into a four-layer data model consisting of Document → Product → Sections → Evidence. It supports three major regulatory templates—NMPA, FDA, and EMA—and performs deep structuring on six key fields: indications, dosage and administration, contraindications, adverse reactions, drug interactions, and special populations. Each field is bound to a provenance chain containing an evidence_id, page number, and paragraph-level location. The system incorporates a three-tier quality control mechanism (PASS, WARNING, MANUAL_REVIEW, FAILED), 14 error code categories, and 10 hard-rule validations implemented in code, with output available in JSON, Excel, and Markdown formats. The article outlines the tool's capability boundaries, typical data flow, and applicable scenarios. Official URL: https://api.skillhub.cn/baitongai/drug-label-structured-parser.
Read Moredsh-wide-focus: Widescreen, Task Panel, and Context Manager for DSH Desktop
dsh-wide-focus is a DSH desktop client plugin maintained by quanwenbo, providing widescreen reading, a draggable task panel (Pending Review/In Progress/On Hold/Approved), context usage badges, handover documentation, and reasoning tier assistance on the local machine. To install, use the git+tag command in the README; if dsh-context-guard was previously installed, it must be uninstalled first. MIT licensed, suitable for long-session development and multi-round tool call scenarios.
Read Moredsh-boss-mode: Boss Mode Preset where the Main Agent Commands and Sub-Agents Execute Tasks
dsh-boss-mode is a DeepSeek Harness workflow plugin maintained by qsheeeeen, introducing a new "Boss Mode" (id: boss) in the mode selector. The primary agent plans, delegates, tracks progress, and validates outcomes using tools such as subagent, send_message, todo_write, and goal, without directly reading or writing files; sub-agents execute using the Standard preset and support resumable sessions. To install, point the install_bundle of plugin_manager to the plugin directory and restart DSH; it depends on dsh-base and dsh-web-app. For customizing the persona.prefix, refer to cordis.patch.yml. Licensed under MIT 1.5.2.
Read MorePinduoduo's Huang Zheng Thinking Framework: Embedding Huang Zheng's Mindset into an Agent
The Pinduoduo Huang Zheng Thinking Framework is an Agent Skill that locks in Huang Zheng's thinking framework and communication style—pre-committing his mental models and patterns of expression deeply anchors into the agent—released by SkillHub community author user_231765ee. Version 1.0.0, falls under the knowledge-management category, and requires no API Key. Content is based on insights distilled from 12 firsthand articles and 7 complete interviews, yielding 7 Core Mental Models, 10 Decision Heuristics, and a Complete Expression DNA. As a thinking advisor, it is used for analyzing business problems, reviewing decisions, and reflecting on life. It can be triggered by SKILL.md parameter usage when users say phrases like "From Huang Zheng's perspective", "How would Huang Zheng view this", "Huang Zheng mode", "Pinduoduo mindset", or "Switch to Huang Zheng". As of the time of capture, the page shows a rating of 100000 and 802 downloads. The article covers its positioning, core capabilities, trigger methods, general installation instructions, and typical usage, and usefully adds an important note on community-curated content.
Read Moredsh-image-size-guard: Preserving Historical Images in Long Conversations Within Gateway Request Body Limits
dsh-image-size-guard is a DeepSeek Harness community plugin designed to handle request body size limits (e.g., 16 MiB) for gateways such as agent2api. It measures the actual request body before it exceeds the budget. For batch requests, it replaces image bytes with placeholder text from the oldest historical images and aborts the current attempt for the agent to retry, ensuring subsequent requests remain below a low watermark. Additionally, it reduces repeated prefix cache invalidation through parameters like `targetTurns` and `protectedTurns`. The plugin supports per-provider configuration of limits and toggles via the Settings panel or YAML, and can work in coordination with `dsh-compaction-image-offload`. Installation: `dsh plugin --profile web add dsh-image-size-guard`.
Read Moreai-brain-learning-memory-pro: Integrating Learning and Memory Mechanisms into Agents
An advanced skill designed for AI developers and agent architects, implementing agent learning and memory as explicit mechanisms rather than context patches. Due to the dynamic rendering of the official page and the failure to scrape the body of SKILL.md, the following text contains only verifiable information: SkillHub rating of 99,877, 793 downloads, and a classification of ai-agent. The installation command, configuration fields, and typical usage examples are marked as to be verified, with a reminder to consult the original text before selection.
Read Moredsh-web-preview: Preview Local Web Services in DSH GUI
dsh-web-preview is a DSH client plugin maintained by qoqpqpq that embeds loopback web services on the DSH host into the GUI via a same-origin HTTPS proxy, solving the problem of localhost and mixed content unable to be previewed in iframes within remote browsers. It provides a Web Preview tab, a PREVIEW floating window, and Console/Network/Errors diagnostics. Installation: `dsh plugin --profile web add dsh-web-preview`, with optional configurations such as defaultUrl, defaultUpstream, extraAliases, etc. MIT 1.1.2.
Read MoreChinese Workplace: A Knowledge Manual Compiling Approximately 25 Classic Chinese Workplace Books into a Skill
"Chinese-Style Workplace" is a knowledge-based Skill tailored for Chinese workplace scenarios, published on SkillHub under the ID @user_b5022f34/zhongguoshi-zhichang. It is at version 1.0.3, sourced from the "community" category, and classified under "business-ops." Integrating approximately 25 classic Chinese workplace texts into a comprehensive manual, its core references include *Canglang Zhi Shui* (The Water of Canglang), *On Being a Staff Officer*, *Dynasty 1566*, *Zeng Guofan*, *County Officials*, *From the Soil*, and *Hidden Rules*. The Skill organizes content across five layers—novelistic insights, historical depth, social DNA, academic frameworks, and practical skills—and covers six dimensions: cognitive awakening, managing up, horizontal relationships (peer relations), self-conduct, long-term strategy, and practical skills. With a specific focus on government agencies and state-owned enterprises (SOEs), it addresses the issue of generic, unstructured responses when AI Agents handle questions within a Chinese workplace context. Currently, the Skill has a rating of 100,000, 807 downloads, and 0 installations, with the tag *requires_api_key=false*. This article introduces its method of content organization, typical usage scenarios, and applicable contexts.
Read MoreFrom Print to Digital: Techniques for Converting In-depth Articles from Internal Newsletters into Engaging WeChat Official Account Posts
This article introduces the Skill "Print-to-WeChat" (@user_9db251a1/print-wechat, Version 1.0.0, community-sourced, content-creation category). It adapts long-form articles from print media and internal publications into posts suitable for WeChat Official Accounts. This process involves rewriting titles, adding opening hooks, replacing subheadings, condensing the length, and converting formal bureaucratic style into colloquial language. By default, it outputs .docx files for the editorial department. Activation is intent-based; it triggers when users mention phrases like "adapt into a post," "this draft is too long," or "internal publication needs to go online." Uploading corporate interviews, work records, project summaries, or leadership speech transcripts also triggers it, even if "WeChat Official Account" is not explicitly mentioned. The article also clarifies its scope boundaries (outputting submission files rather than final layouts) and verifiable metadata (requires_api_key=false). However, it notes parts that could not be verified: the official page lacks installation commands and configuration fields, and enablement methods for Cursor, Codex CLI, and Claude Code were unavailable during this capture, so unverified details are omitted. Additionally, the article flags data conflicts: leads claimed a rating of 99,877 and 813 downloads, whereas the page displays a rating of 100,000, 876 downloads, and 0 installations, noting that official real-time data should prevail. It also reminds users of the community source, lack of fact-checking capabilities, and the necessity of manual verification of data and proper names before publication.
Read MoreCC Skill: Continual Learning—A General Skill Metadata Page Labeled as "Continual Learning"
"CC Skill: Continuous Learning" (@org-02qudk26/cc-skill-continuous-learning) is a universal SKILL.md-format skill registered on SkillHub. Its publisher is Zhiyu Technology, sourced from enterprise, classified under ai-agent, and versioned at 1.0.0. It requires no API key. The official description is only one sentence: "A development skill from everything-claude-code." The article consolidates the verified metadata and points out several inconsistencies worth noting: the name includes "continuous learning," yet the description does not explain the specific mechanism; the source is marked as enterprise, which conflicts with the attribution to everything-claude-code in the description; and the install count is 0, with no publicly available records of usage. Because the body of the official page could not be retrieved, this article does not provide installation commands or usage examples, and only provides the official URL for readers to verify on their own.
Read MoreSidebar Balance: View DeepSeek Account Balance at the Bottom of the Sidebar
Sidebar Balance (dsh-sidebar-balance) is a DeepSeek Harness client web plugin maintained by Qliphoth11. It displays the Platform balance, consistent with the Account page, within the sidebar.footer.action area via the Remote account feature. It supports click-to-refresh, Chinese and English localization, and theme adaptation, with no telemetry. It can be installed via a local bundle absolute path (install_bundle) or using the pnpm spec `github:Qliphoth11/sidebar-balance`. A logged-in DeepSeek account is required, and it has been verified on version 0.2.1-rc.5.
Read MoreDe-AI-ifying Public Accounts: Removing AI Tone and Adding Human Touch to Long-form Content
"WeChat Official Account AI De-flavoring" is a community-sourced, general-purpose Skill in the SKILL.md format, published by user_140a98d6. Currently at version 1.1.0, it falls under the content-creation category. Designed specifically for WeChat Official Account long-form articles, it performs two core functions: removing AI-like characteristics and adding a human touch. Its goal is to make revised manuscripts read like authentic human writing without altering facts or structure. The Skill is tagged with `requires_api_key=false`, indicating that no additional API key configuration is necessary. During verification, both the official page and the API address were inaccessible. As the documentation lacked installation commands, directory paths, and reproducible examples, this report does not provide unverified commands or output results. The data snapshot shows a score of 100,000, 903 downloads, and 0 installations, implying a very low rate of actual usage following download. The article outlines a basic information table, mounting logic, and typical usage workflows. It also includes applicable scenarios and four key precautions, suggesting that users first test run the process on a non-critical manuscript to confirm the desired expressive style before integrating it into formal workflows.
Read MorePractical Handbook for WeChat Mini Programs: A Collection of Inspection Scripts for Iterative Development and Debugging
Mini Program Practical Manual: An Agent Skill published by community developer user_71d62880, based on the general SKILL.md format, version 1.0.1, categorized under dev-programming, and requires no API key configuration. It covers five high-frequency scenarios in WeChat Mini Program iteration and troubleshooting: - Static review of interaction logic. - Screen failure (black or white). - Canvas chart rendering and headless self-checking. - Boundary on third-party integration. - Style transplantation from reference sites. With checklists provided for each scenario, developers can run locally for self-assessment. This article covers the positioning, core capabilities, installation entry, and typical usage practices of this Skill, translating into applicable scenarios and limitations.
Read Moredsh-panel-control: Sorting, Hiding, and Grouping of Sidebar and Top Bar Entries
Maintained by QIUZAIYOU, `dsh-panel-control` manages the top sidebar global panel navigation for DeepSeek Harness, featuring drag-and-drop sorting, individual item hiding, named grouping, and collapsing. Configuration is persisted in `~/.dsh/panel-control/prefs.json`. The plugin also imposes a height limit and scrolling on the navigation list, preserving minimum height for the session/workspace below to alleviate layout issues caused by excessive entries. It supports optional automatic model categorization (2–4 groups). It is recommended to install via `npm install dsh-panel-control@0.7.2` in the profile directory and add it to `dsh.profile.bundles`. MIT License.
Read MoreThree-Province and Six-Ministry System (dsh-san-sheng-liu-bu): Automatically switching inference adjustment and tool selection according to task difficulty
The Three Departments and Six Ministries Process Optimization Plugin (dsh-san-sheng-liu-bu), maintained by qiushuiqianpingsha, is a workflow-class DSH plugin. It performs local heuristic assessments of user message difficulty, setting simple tasks to off/low and hiding approximately 50 heavyweight tools, medium tasks to low, and complex tasks to high; if uncertain, it prompts via a countdown. By restricting step-specific tools via inbox/claimed and rewriting reasoningEffort during requests, it mitigates latency caused by large contexts and high-intensity reasoning at each step. This article covers difficulty level mappings, configuration options, official installation and verification procedures, and known limitations.
Read MoreAnalysis of A-Share Shareholder Count and Average Holdings: Visualization of Quarterly Shareholder Data and Chip Concentration
This article introduces the "A-Share Shareholder Count & Average Holdings Analysis" Skill on SkillHub. Published by user_058048a4 (Version 1.0.0), it is categorized under "data-analysis" and requires no API key. Users can input a stock name or code to obtain quarterly shareholder count and top-ten shareholder holding data. It calculates the average holding value per account using the formula (Total Market Cap - Top Ten Shareholders' Market Cap) / Number of Shareholders, generating a composite chart featuring five-year stock price trends, shareholder count bar charts, and average holding value bar charts. The article outlines core features, trigger words, typical usage, applicable scenarios (chip concentration, major capital trends, and bull-bear cycle analysis), and important notes.
Read MoreIntroduction to the SQL Optimization Expert Skill
"SQL Optimization Expert" is an Agent Skill released by Kunlun Growth, located at @kunlungrowth/kg-sql-pro. It is version 1.0.0, sourced from enterprise, categorized under dev-programming, has a rating of 100000, 857 downloads, 0 installs, and does not require an API Key. It focuses on three areas: SQL statement optimization, database schema design, and slow query troubleshooting. It is suitable for scenarios where existing SQL needs to be accelerated, new modules require table structure design, or slow queries have appeared in production and need to be located. The official materials retrieved do not provide installation commands, configuration fields, or reproducible usage examples. The main text states this truthfully and defers to the official page, without fabricating installation steps or performance data. It should be noted that the optimization suggestions from this Skill still need to be verified in conjunction with specific data distributions and query patterns, and cannot directly replace DBA judgment.
Read MoreDSH English Break: Utilize fragmented time in DSH to practice daily and work English
dsh-english-break (English Break) is an MIT open-source DSH community plugin maintained by qiufengsir, compatible with DSH 0.2.1-rc.3. It enables daily conversational and workplace English practice in the sidebar or right panel, supporting sentence-building feedback, Chinese follow-up questions, favorites with spaced repetition, AI-generated random topics, and local JSON backup. Models follow the DSH system defaults with usage incurred for API calls; data is stored in storages/dsh-english-break and does not read work sessions. Installation can be performed using the tgz file from Releases or GitHub tags, with an optional custom marketplace index.json.
Read Morears-academic-reviewer: Multi-Perspective Paper Review Using a Five-Member Review Panel
The **ars-academic-reviewer** is a community-sourced professional Skill that operates a five-member role-separation review panel (comprising a Journal-Fit Reviewer, three peer review roles, and a Devil's Advocate). It supports six modes: full review, re-review, quick assessment, methodology focus, Socratic guided, and calibration. Trigger words cover English, Chinese, and Korean. The document explains its definition, core capabilities, installation and activation, typical usage, applicable scenarios, and caveats. It is important to note that role separation does not equate to independent verification; also, as the current crawl did not provide specific installation commands and configuration fields, users should verify details on the official page before use.
Read Moredsh-chatgpt-subscription: Using ChatGPT subscription as the DSH primary model (OAuth, no API key required)
This document describes the maintenance of the DSH model reasoning plugin, dsh-chatgpt-subscription (0.1.8), by qiuming. It leverages "Sign in with ChatGPT" OAuth to integrate a ChatGPT subscription into the primary model provider, openai-chatgpt, eliminating the need for an API Key and avoiding fallback to API billing. The plugin supports streaming responses, local tools, tool-history replay, image input, and controlled web_search within the same request. It requires Node.js 24+ and DSH 0.2.1-rc.2. Version 0.1.8 is not available for direct installation via GitHub; manual installation is required using source code commands (npm ci, npm run check, npm run install:local or integrate:web), followed by dsh-chatgpt login and either dsh web or dsh-chatgpt run. Users are advised to review the source code and pay attention to authorization storage and subscription limits before use.
Read MoreHigh School Entrance Exam Question Drilling Assistant: A Skill for Generating Practice Exercises by Chapter, Question Type, and Difficulty
Zhongkao Practice Assistant (@user_af163eaa/z-brush-zhongkao, v1.0.1, education, community) is a practice-assist Skill for the Chinese high school entrance exam. Questions are generated by AI based on a generic syllabus; it is not for question prediction and not official. It covers high-frequency exam points via a chapter map, and can generate exercises across three dimensions of chapter, question type, and difficulty. Answers and explanations are included; wrong answers are saved to a local mistake repository, which can be reviewed repeatedly for reinforcement. This article introduces its positioning, core capabilities, official address, typical usage, and important notes. The material does not provide specific installation commands; refer to the official page for details.
Read MoreSkill Doctor: Evaluating and Recursively Optimizing Agent Skills
Skill Doctor (@user_d0deb03a/skill-doctor-v1, v1.0.0, community-sourced, category: ai-agent, requires_api_key=false) is a skill designed for evaluating and optimizing other Agent Skills. It conducts multi-dimensional evaluations and scoring on any AI Agent Skill, generates diagnostic reports, and automatically optimizes them through a recursive self-improvement loop. It utilizes a keyword-driven trigger mechanism, activated by: evaluating skills, scoring skills, optimizing skills, skill doctor, detecting skill quality, conducting a skill health check, or evaluating this skill. As the documentation does not provide official installation commands or configuration file paths, this article only lists verifiable coordinates, versions, sources, and tags. The developers explicitly state that it is not suitable for natural language polishing, code debugging, or general Q&A. Real-time data from SkillHub indicates a rating of 100000, 854 downloads, and 0 installs. Due to the limited community verification sample, it is recommended to use it in a rollback-capable environment and verify it independently.
Read More