Li Mu's Paper Intensive Reading Skill: An Executable Pipeline for Paper Reading
The Li Mu Paper Reading Skill is an Agent Skill for academic paper reading and literature research, published by community developer user_d5d46a15. It is at version 1.2.0 and falls under the knowledge-management category. It formalizes paper reading into a three-pass pipeline: first, the "five Cs" screening method determines whether a paper is worth reading and how deeply to engage with it; second, a systematic close reading is carried out chapter by chapter; and third, the argumentation chain is dissected, experimental designs are reviewed, and reproduction risks are assessed. Input supports arXiv links, PDF files, DOIs, and paper titles, and it can process paper-list tasks. For output, the Skill automatically extracts figures and images from PDFs and generates local HTML reading notes with MathJax-rendered formulas, making it convenient to view image-and-text content alongside the original in a browser. The official description also covers extended uses such as paper comparison, reproduction planning, research topic discovery, and literature mapping. The tag requires_api_key=false means no API Key is required, and it is based on the general SKILL.md format. As of the capture time, the SkillHub rating was 100000, with 1102 downloads and 0 installs. No official installation command or changelog has been provided; it must be loaded in the client via the coordinate @user_d5d46a15/mli-paper-reading-penghel.
Read MoreMedia Sense: Understanding Local Audio and Video with Configurable Models in DSH
Media Sense (xhznAI/media-sense) provides local audio and video understanding for DeepSeek Harness. By configuring audio and video models along with FFmpeg in the settings, it passes workspace media to interfaces such as Gemini, Qwen Omni, and OpenAI Audio, with DSH compiling the text-based responses. It is compatible with DSH >=0.2.1-rc.3 <0.2.2-0 and can be installed from github:xhznAI/media-sense#v0.1.2. Based on the official README, this article covers key points including installation, configuration, model selection, and the 25 MB default limit.
Read MoreConsensus Expectation Analysis: Breaking Down Institutional Expectations into Verifiable Data
"Consensus Expectation Analysis" is a stock analysis skill powered by the Hundsun Juyuan MCP financial database, maintained and released by community developer user_cc9425b3. Operating in version 1.0.1 within the "professional" category, it systematically deconstructs four core datasets: institutional earnings forecasts, rating adjustments, analyst report insights, and corporate fundamentals. By detecting marginal shifts in market expectations, it identifies investment opportunities arising from expectation gaps. The skill activates through two primary triggers: users requesting a "Consensus Expectation Analysis Report" by providing stock codes/abbreviations, or direct inquiries about institutional forecasts (upward/downward revisions), expectation gaps, or rating changes. Official examples demonstrate usage with benchmark stocks like Kweichow Moutai, CATL, BYD, and China Merchants Bank. Note that trading suffixes should be included when referencing securities. Key considerations include: 1) Requires API key configuration (requires_api_key=true) prior to activation with Hundsun Juyuan MCP; 2) Installation details, MCP service endpoints, and field specifications require verification on the official page, as documentation lacks distinct installation procedures for Cursor, Codex CLI, and Claude Code; 3) Consensus metrics derive from aggregated institutional forecasts where sample size variations may induce mean volatility - requiring combined analysis with coverage breadth and rating trends; 4) Outputs serve informational purposes only and do not constitute investment advice. As a community-maintained tool (v1.0.1), interface fields may evolve with database updates. With zero recorded installations and no large-scale validation data available, users should test with single-stock scenarios before batch processing.
Read Moredsh-second-opinion: Get a second opinion via an independent advisor session when stuck
dsh-second-opinion is a DSH workflow plugin maintained by xfgil. When a session gets stuck, it uses ask_advisor to analyze the situation in an independent advisor session, with the answer returned synchronously as a tool result—without impersonating the user or requiring the user to relay information. Each help request automatically binds a dedicated advisor (titled "Advisor + first three characters"), and additional independent second or third advisors can be requested. On the advisor side, advisor_watch allows read-only observation of the main session, and advisor_note enables leaving notes. Configuration is written in cordis.patch.yml, and installation commands are documented in the official README. The plugin depends on DSH internal interfaces, so please read the source code and the MIT License before using it.
Read MoreAssistant for Diagnosis and Prevention of Common Quality Defects in All Disciplines of Building Construction: Transforming Quality Inspection from Experience-Based to Closed-Loop Management
This article introduces the "Building Construction Quality Defect Diagnosis and Prevention Assistant - Pengge Liaogongcheng AI" on SkillHub, posted by user_833fca90, categorized under business-ops, versioned 1.0.1, and sourced from the community. Targeting total process quality management in building construction, this Skill covers specialties including foundations, main structures, masonry, roofing, waterproofing, decoration, doors and windows, curtain walls, plumbing, electrical, HVAC, fire protection, intelligent systems, outdoor works, and finished product protection. It identifies defects such as cracks, hollowing, leakage, honeycombing, exposed reinforcement, rebar displacement, formwork expansion, and visual defects. Upon user upload of on-site photos, construction records, or text descriptions, the Skill follows a three-step process—problem localization, cause analysis, and rectification plan generation—to output standard templates such as quality diagnosis reports, prevention schemes, rectification notices, and inspection checklists. It classifies risks into three levels: general, relatively large, and major, forming a closed-loop rectification process with problem numbering and status clearing. The article also outlines applicable scenarios, six quality management principles, standards for handling common quality issues, and output boundaries regarding structural safety, material determination, and legal liability.
Read Moredsh-activity-bar: Collapse DSH sidebar into a VS Code-style icon bar, freeing up workspace width
dsh-activity-bar is a DSH client plugin maintained by XenoPhonLau. It compresses the left and right sidebars, which typically occupy nearly 40% of the screen, into 44px icon rails on each side. The grid effectively indents rather than overlapping, while icons such as Bot and Settings remain at the bottom of the rails. The right panel width can be adjusted from 300px to 60% via dragging, and this setting is retained. Users can restore the native DSH layout by disabling specific sides via Settings. Requires DSH 0.2.1-rc.3+ and Node >=22.13. Released under the MIT license. Installation: Run `dsh plugin --profile desktop add "dsh-activity-bar@git+https://github.com/XenoPhonLau/dsh-activity-bar.git"` and restart.
Read Moredsh-message-rewind: Edit sent messages in DSH and roll back subsequent rounds
dsh-message-rewind is a DSH client plugin maintained by xCube007 that provides editing and deletion on messages you have sent, truncates subsequent conversation turns, and offers the option to restore either only the message or both the message and text file changes. It requires DSH ≥0.2.1-rc.3 and Node ≥20. Recommended installation command: dsh plugin --profile web add dsh-message-rewind. It is suitable for correcting and retrying in long Web conversations. Note the rollback limitations regarding Git working tree, expired snapshots, and binary files. The plugin runs with the current process permissions. Please consult the source code and MIT license before installation.
Read MoreCognitive Restructuring Engine (Erlang Shen): Decomposing Complex Problems into Actionable Components
The Cognitive Restructuring Engine (Erlang Shen) is an Agent Skill providing a deep analysis and problem-solving methodology for general problems. Its coordinates are @indiv-r_stevens/erlang-shen-thinkings, it is categorized as 'knowledge-management', maintained by community publisher user_6fc68038, and is currently at version 2.8.9. It originates from the community and does not require an API Key. It integrates multiple thinking methodology documents, including the "Framework of Thinking" series, "Growth Mindset", "Deep Thinking", "Cognitive Awakening", critical thinking, long-termism, and the separation of tasks. By leveraging an external knowledge base for deep reasoning, its core capabilities are centered on deconstructing complex problems, clarifying the essence of an issue, establishing analytical frameworks, reasoning with cognitive models, distinguishing facts from opinions, and formulating systematic solutions. It is triggered by keywords such as "Erlang Shen thinking", "thinking methodology", "analyze problems with a framework", "how to think clearly", "deconstruct a problem", "problem analysis", and "cognitive restructuring". The article introduces its positioning, core capabilities, and trigger methods. It also clarifies that there is no directly verifiable installation command in the official documentation, and that the skill is not suitable for pure writing polishing, fact-checking, or professional numerical calculations within a single domain.
Read Moredsh-privacy-gateway: Outbound Request De-sensitization and Response Restoration
**dsh-privacy-gateway** is a DSH community plugin maintained by xbzbing. It intercepts all model HTTP requests made via `fetch` at the host's undici global dispatcher: outbound requests have API keys, cookies, passwords, private keys, and optional personal information replaced with placeholders, and inbound responses are restored transparently—both the agent and provider remain unaffected. It supports a plugin configuration page and `cordis.patch.yml`, a `/privacy` command and session privacy panel, as well as local JSONL statistics. To install, run `dsh plugin --profile web add dsh-privacy-gateway@latest`; do not install from the GitHub source tree. The plugin runs with the DSH process's permissions, so please review the MIT-licensed source code before installation.
Read MoreBazi Chart Casting and Character Reading Compass: A Pure Front-end Tool for Metaphysics and Human Insight
"Bazi Charting & People-Reading Compass" is a community-sourced Skill, located at @user_8ca2df90/tianjisuanm, published by user_8ca2df90, version 1.0.5, in the life-service category. It bundles seven major modules—Bazi chart casting, people-reading techniques, human weaknesses, legal hazards, social rules, and hidden trades—into an interactive "Life Compass" single-page application, featuring a dark gold metaphysical UI and scan QR code to use on mobile. Built entirely as a front-end application with no backend dependencies, it requires no API key and runs directly in the browser. It is designed for developers who want to quickly test metaphysical and people-reading content products without writing backend code. Please note that the installation count is zero, and the module list is subject to the official page. The metaphysical content should not be used as a basis for decision-making.
Read MoreTencent Cloud Harness Best Practices - Skills: Assembling Git Repositories into a TAPD + CNB Closed-Loop Workflow
This article introduces the community Skill on SkillHub, "Tencent Cloud harness best practices - skills" (Version 1.0.1, categorized under it-ops-security). It orchestrates existing Git repositories into a closed-loop workflow: TAPD story $\rightarrow$ feature branch $\rightarrow$ PR $\rightarrow$ CNB AI code review $\rightarrow$ merge to main $\rightarrow$ automatic commit write-back to TAPD story. The Skill covers capabilities including repository integration with CNB/TAPD, automatic association of commits and PRs with stories, enabling cnccool/ai-review, installing Git hooks to block direct pushes to main and enforce feature branch workflows, creating feature branches from TAPD stories, and updating the "Code" tab in TAPD stories with commits. The article lists seven official trigger scenarios, suitable teams, and considerations. It also notes that key information, such as installation commands, is missing from the captured data and should be verified on the official page.
Read MoreSession Rail: View Environment and Git Changes by Session in the DSH Right Sidebar
Session Rail is a DSH Web client plugin (v0.1.2, MIT-licensed) maintained by wzc520pyfm. In the right sidebar it presents, per session, the Environment, Context, Plugins, and Git Changes sections. The Context section reads from the token-meter projection. The Git area can request a live Agent to commit the workspace with a fixed message; clicking a file jumps to the Review page or shows an inline diff. For local installation, after running npm build in the package directory, execute: dsh plugin --profile web add "file:$PWD". The host requires Node.js ≥22.19.0. The plugin does not modify Git or files on its own; reviewing the source code and permissions before installation is recommended.
Read MoreOffice360: Offline Office File Reading, Writing, Creation, Conversion, and Editing Toolkit
office360 is an offline Office file toolbox published by community author user_d41e2baa, version 2.1.0, categorized under office-efficiency, requiring no API Key. It breaks the processing of .pdf/.docx/.pptx/.xlsx into four pipelines: READ / CREATE / CONVERT / EDIT. Reading has zero dependencies. Conversion supports dual engines: Microsoft Office/WPS COM and LibreOffice on Windows. XLSX creation supports formulas such as SUM, AVERAGE, IF, and 13 financial color schemes (blue=input, black=formula, green=cross-sheet), with calculated values truly recomputed via LibreOffice headless mode. PPTX supports visual QA, and PDF supports OCR and forms. The article introduces its positioning, verified capability boundaries, prerequisites for dual-engine usage, typical use cases and applicable limitations, and provides the official address.
Read Moredsh-session-title-regen: Regenerating Titles from Recent Conversations Using the Session's Own Model
The DSH plugin **dsh-session-title-regen** adds an "AI Regenerate Title" option to the session line's "…" menu. It reads the last 10 user prompts and the model routing information from the final request/header in the local session log, calls the LLM via the loopback API, and writes back to standard session/title events after preview confirmation. It does not rely on the global sessionTitle configuration and requires zero setup panel configuration. Web installation: `dsh plugin --profile web add github:wyewyewye/dsh-session-title-regen`; the host must be restarted after installation. Desktop users should use plugin_manager. MIT licensed, maintained by wyewyewye.
Read MoreAcademic Omni-Search: Unifying Chinese and English Literature Retrieval into a Single Skill
Academic Omnisearch is a community Skill designed for the universal SKILL.md format, published by user_b0b14db8. Currently at version 1.2.0 and categorized under knowledge-management, it consolidates five major data sources—arXiv, Semantic Scholar, PubMed, CrossRef, and Baidu Xueshu—into a single entry point. It supports concurrent searching, citation chain tracking, author search, DOI resolution, BibTeX export, and AI-powered intelligent summarization. Automatic routing sends Chinese keywords to Baidu Xueshu, while English keywords trigger concurrent searches across all sources. The tag `requires_api_key=true` indicates that API Key configuration is necessary for usage, although the official documentation lacks specifics regarding key fields and client-specific installation instructions. The article outlines core features, installation access, typical usage, applicable scenarios, and limitations.
Read MoreIntroduction to the AI E-commerce Video Ad Skill (LinkPix): Generating Video Ad Assets Based on Product Information
LinkPix is an ad video generation Skill published by Qinghu AI on SkillHub (Coordinate: @qinghu-ai/linkpix-video-ad-assets, Version 0.1.8, Category: content-creation). It is positioned to rapidly generate ad video assets based on product information, making it suitable for feed ads, brand promotion, and social media marketing. The tag `requires_api_key=true` indicates that an API Key is required for invocation. This article introduces its positioning and verifiable capabilities, and honestly notes that the retrieved materials do not provide official installation commands, configuration fields, or invocation examples. The official page should be consulted before integration. It is suitable for e-commerce sellers, ad delivery teams, and content operation agencies that need to batch-produce ad assets by SKU. Users should pay attention to billing, asset compliance, and copyright issues when using it.
Read Moredsh-tmon: Bringing the tmon Long Command Workbench into DeepSeek Harness
dsh-tmon, maintained by wr0ld, is a DSH client plugin that embeds the tmon web workbench into sessions as a "Command Terminal" tab. It registers tmon skills, system prompt sections, and the tmon_tasks tool, allowing the model to query the status, duration of silence, and output of long-running commands. Users must install @qiu745/tmon globally via npm prior to use; the plugin is installed via `dsh plugin add dsh-tmon` and requires no additional configuration. It is ideal for Harness users who need to distinguish between hung and slowly executing commands.
Read MoreLearn Companion: Record the entire learning process
Learn Companion is a community-produced learning companion Skill (located at @user_47a4634f/learn-companion, v2.3.3, category: education; SkillHub rating: 100,000, with downloads in the order of 1,000). Once activated by a passphrase, students provide a six-field profile (subject, teacher style, student level, course, student ID, and name), which is automatically remembered thereafter. The AI conducts guided conversational teaching based on the selected teacher's style. Every round is automatically timestamped and recorded in an append-only manner. Upon completion, the full learning process can be exported as a timeline-based HTML file with one click to share with teachers for verification, feedback, and retention. The entire process features local storage, is touch-screen friendly, supports image-free inline LaTeX rendering, and requires no API Key. It is suitable for individual students who require supervision and verification of their learning process. Please note that records are append-only, local data contains personal information, and LaTeX will display as source code in environments without rendering support.
Read Moredsh-experience: DSH Self-Learning Local Experience Base
`dsh-experience` (wqx11235) is a memory plugin v0.1.0 of DeepSeek Harness: a local Markdown experience library, weighted retrieval via error signatures, proactive injection of pitfall-avoidance guidance when tools fail, and a closed-loop ingestion pipeline where gaps are stored through `experience_promote`, accompanied by a Web-side sidebar panel. No model, no vector database, and no network access by default. Installation: `dsh plugin --profile web add github:wqx11235/dsh-experience`; requires `dsh >= 0.1.7-rc.2` and a host restart.
Read MoreLocal-First De-identification Skill for Chinese Legal Documents
Introducing "Legal Document Redactor" on SkillHub: a local-first, Chinese legal-document de-identification tool (handle: @indiv-lawyerzhu/legal-document-redactor, version 1.0.8, community source, category: professional, requires_api_key=false). Core capabilities include online pre-AI de-identification, structural redaction for external submissions or litigation materials, and user-selectable full-party obfuscation to block Party A or Party B (or both), covering names, contact details, signatures, and the entire company seal. It preserves the original file format, generates local audit records, performs residual checks, and requires manual review of results. Suitable for documents sent externally, filed for litigation, or processed before being uploaded to AI; not intended for contract drafting, legal research, or non-Chinese PII pipelines. The article honestly notes that the official installation command and configuration fields do not appear in the materials retrieved and should be confirmed on the SkillHub page.
Read MoreThe Text Interpretation: Synthesizing Methodologies from Ten Experts into a Single Agent
The "Text Interpretation Expert Team" is a general-purpose, community-driven open-source Skill adhering to the SKILL.md format. It distills the interpretive methodologies of ten renowned literary critics—Sun Shaozhen, Zhan Dan, Qian Liqun, Ye Jiaying, Jiang Xun, Chen Riliang, Rong Weidong, Li Zehou, Ye Lang, and Matsuura Tomohiro—into a structured reference document equipped with an intelligent triage system. Once a user specifies a text title or expresses a particular confusion, the triage system employs discriminative questioning (distinguishing between personal comprehension and teaching preparation, or difficulty in initiating analysis versus connecting details) to automatically route the query to the most suitable expert, providing a one-sentence rationale for the selection. Upon routing, the system loads the corresponding expert's `method.md` and specific volumes, strictly adhering to that expert's methodology to generate the interpretation. Designed for Chinese language teachers in primary and secondary schools, educational students, and teaching researchers, this tool covers scenarios such as textual interpretation, lesson preparation, instructional design, close reading, and whole-book reading.
Read Moredsh-fileguard: Blocking Dangerous Deletion and Modification with Soft Constraints in Full Authority Mode
dsh-fileguard is a DSH admin-security plugin maintained by Wldpm. Under full workspace-write permissions, it preemptively blocks dangerous deletions and modifications and requires the model to explain the impact in natural language. Users can obtain a one-time token via /guard-gettoken to allow the action. It provides four policy levels, hard restrictions on root and system directories, and an AI temporary-file ledger that bypasses review, and explains the boundaries of soft constraints and the link-based installation method. Version 0.2.1, MIT license.
Read MoreMaster Job Analyzer: Generate Complete Job Profiles by Inputting Job Title
The Job Analyzer is a community-based Skill based on the SKILL.md format, identified as @user_068c5cf6/job-analyzer, version 1.1.1, categorized under business-ops, and sourced from the community. It requires no API key. It addresses the problem of drafting job documents from scratch with inconsistent formats: by inputting a job title, it generates job responsibilities, standardized SOP processes, boundaries of authority and collaboration, compliance red lines, hard and soft skills, performance assessment dimensions, workplace pitfalls to avoid, and career growth paths. It supports supplementing industry, business model, company size, and job level for precise output, and also allows pasting enterprise recruitment JDs or job links to generate enterprise-customized content. Trigger words include job analysis, generate job SOP, job responsibilities, job design, workflow, job description, job SOP, job rules, job assessment, and job competencies. The SkillHub rating is 100000, with 1,124 downloads and 0 installations. The official address is the SkillHub page and API URL. When using it, note that the output is a structured skeleton, compliance red lines and assessment indicators need confirmation by the business side, the customization effect depends on the quality of JDs and constraints, and as it comes from the community with 0 installations, it is recommended to review the generated content first.
Read MoreEnterprise Dual Carbon Planning: Connecting Carbon Inventory to the Roadmap
Corporate dual-carbon target planning is an enterprise-level competency for planning pathways toward carbon peaking and carbon neutrality. Coordinate: @user_0c00f153/shuangtan-zrrr; Version 1.0.0; Published by user_0c00f153 from the community source; Category: professional; Tag: requires_api_key=false. It integrates organizational carbon emission inventory, carbon emission accounting, emission structure analysis, carbon reduction diagnostics (matching the National Green Technology Catalog), and the generation of carbon peaking and carbon neutrality roadmaps and action plan tables into a single skill. The methodology adopts GHG Protocol Scopes 1, 2, and 3 with the emission factor method, covering industries such as manufacturing, energy, construction, and transportation. This article introduces its core features, enabling conditions, typical use cases, and usage limitations.
Read Moredsh-clear-conversation: Clear model context with natural language to save tokens
Introducing dsh-clear-conversation, a dsh client plugin maintained by William20100221: in long chats, use natural language or /clear to stop old turns from entering the model surface; transcripts are retained, thereby immediately reducing input tokens. It covers install commands, example conversations, differences from /compact, token cards and update-check environment variables, plus Preview status and usage caveats.
Read More