Qwen Image Generation Skill: Invoking Qwen-Image to Generate Images with Chinese Text
Introduces the community-sourced Skill “Qwen Image图片生成” (@user_0064faa7/moark-qwen-image, version 1.0.0, category design-media). It calls the Qwen-Image model on ai.gitee.com to generate images, supports Chinese text-to-image, text layout, negative prompts, size, seed, and LoRA, and is suitable for generating images with Chinese text such as couplets, posters, and illustrations. The tag requires_api_key=true means an API Key must be configured before use. The article explains the scope of capabilities, prerequisites for enabling, typical usage elements, and applicable limitations, and provides the official URL; because the official source does not provide installation commands or example parameters, the relevant sections state that these have not been verified, without speculation.
Read MoreBlender Geometry Nodes Script Generator: Procedurally Building Geometry Node Trees with bpy
This article introduces a community-sourced Blender Geometry Nodes Script Generator (@user_3c6cb52e, version 1.0.0, dev-programming). It generates Blender Python (bpy) scripts that programmatically build geometry node modifier trees via the node_groups API and the Geometry NodeTree interface, enabling parametric 3D asset generation without an API key. The article describes its positioning, verified capabilities, and enablement clues, and notes that official documentation provides neither installation commands nor reproducible examples; typical usage requires verifying the API against the target Blender version.
Read Moredsh-status-card: Persistent Status Card in the Top Right Corner of the DSH Web Content Area
dsh-status-card is a DSH web client plugin maintained by liuhusk. It provides a persistent collapsible status card in the top-right corner of the content area, aggregating current round timing (model/tool/waiting), official todos and history tasks, editable goals, submitted plans, session modified files, sub-agent lines, and background bash jobs. Data primarily comes from official projections and Remote, supplemented by Host routing scan logs. It supports four highlight color levels, narrow-screen push-aside, and auto-hide under the global panel. Requires DSH 0.2.1-rc.3+ for installation: `dsh plugin --profile web add github:liuhusk/dsh-status-card`. MIT licensed, current version v0.23.3.
Read MoreFrom Markdown to WeChat Official Accounts: A Technique for Handing Over Markdown Manuscripts to an AI Agent for Formatting
This article introduces the community Skill "Markdown to WeChat Official Account" (@user_3c6cb52e/md2wechat-skill-yt1g, version 1.0.0, source: community, category: content-creation). It addresses a specific gap: Agents excel at producing Markdown, but the WeChat Official Account editor does not natively support Markdown, making manual transfer and formatting time-consuming. The Skill is tagged with requires_api_key=false, and its official description states it is a "dedicated skill for converting Markdown to WeChat formatting, helping AI Agents efficiently complete related tasks." The SkillHub page shows a rating of 100000, 933 downloads, and 0 installations. At the time of verification, both the official address and the SkillHub page were inaccessible, so the installation command, configuration fields, and reproducible examples could not be verified. The article does not fabricate information; instead, it provides a general integration approach and a pre-adoption checklist, while cautioning that community-sourced skills require self-auditing of the source code and noting the absence of reference samples due to zero installations.
Read MoreAI安全与红队测试：面向 AI 系统的实战红队手册
This Skill serves as a practical red teaming playbook for AI systems, covering six major attack surfaces: prompt injection, unauthorized access and tool abuse, data and privacy leakage, hallucinations and quality defects, supply chain and model poisoning, and denial of service, mapping to OWASP LLM Top 10 risks. It provides a complete workflow from goal definition, attack surface modeling, and test case design to execution, reporting, and remediation retesting. It includes a library of test cases such as direct and indirect prompt injection, Agent privilege escalation and sandbox escape, training data leakage and memory attacks, and hallucination detection benchmarks, as well as tools for vulnerability stratification, remediation recommendations, risk checklists, and report templates. It is designed for AI engineering, security testing, and information security leads to facilitate pre-launch assessments and establish team risk baselines, forming a governance and technical closed loop to complement AI governance.
Read MoreBOPPPS Teaching Model_Qiuye Workplace: Generating Teaching Plans Based on BOPPPS
This article introduces the "BOPPPS Teaching Model_Auye Workplace" on SkillHub, a Skill version 1.0.0 released by Auye Group. Classified under "education," it is designed to generate teaching plans based on the BOPPPS model, with the trigger words "BOPPPS mode" and "generate teaching plan." The article explains its positioning, core functionality, installation and activation methods, typical usage, applicable scenarios, and limitations, and provides the official address.
Read Moredsh-balance: Real-Time Monitoring of DeepSeek Balance and Token Usage in the DSH Web Chat Interface
dsh-balance is a DeepSeek Harness plugin maintained by liudiw, targeting the web profile. It displays the DeepSeek account balance and local token usage in the top right corner of the chat window, featuring an expandable view for the platform's current monthly and historical cumulative totals (requires userToken configuration). Credentials can be saved to `~/.dsh/.credentials.yaml` directly via the interface and are not read back in plaintext. The recommended installation method is using `dsh-balance.tgz` from the releases via the `dsh plugin --profile web add` command; it takes effect after restarting dsh web. Requires Node.js ≥ 20 and is licensed under MIT.
Read Moredsh-assistant-suite: Checking Vendor Usage and Receiving System Notifications in DSH
dsh-assistant-suite is a DeepSeek Harness community plugin maintained by liuchaoxu. It merges the original dsh-model-usage and dsh-reminding plugins, allowing users to view multi-vendor balances and quotas, as well as Command Code balance consumption trends, directly in the session header. Notifications and panel display can be uniformly managed under Settings → Assistant Suite. It supports automatic/manual refresh, dark and light themes, and narrow-screen layouts. Requires DSH>=0.2.1-rc.2 and Node>=18. It can be installed via the plugin page or by running `dsh plugin add dsh-assistant-suite`; after enabling it, the two legacy plugins should be disabled to avoid duplication. Released under the MIT license as an independent community project.
Read MoreRound 1 Req 1：A Community Skill to be Verified
This article introduces Skill Round 1 Req 1, released by the community. At the time of writing, the official page was inaccessible, and neither the functionality nor the installation commands could be verified. The article documents only the information confirmed in the metadata: version 1.0.0, source: community, category: ai-agent, requires_api_key=false. It also clearly enumerates the missing information and provides instructions on how to verify it independently.
Read Morewechat-draft-push: One-click Push of HTML Articles to WeChat Official Account Drafts
wechat-draft-push is a community-produced Agent Skill maintained by user_6bee27e2. Currently at version 1.0.2 and categorized under office-efficiency, it enables one-click pushing of HTML articles to the WeChat Official Account draft box. It automatically handles three main tasks: uploading local images to the media library, stripping CSS features unsupported by WeChat (such as gradients, shadows, positioning, and animations), and creating or overwriting drafts. It is designed to facilitate an automated publishing workflow of "writing → direct draft submission" when used with writing tools like WorkBuddy. Prerequisites include a Verified Service Account or Subscription Account, configuration of WECHAT_APPID and WECHAT_SECRET environment variables, and adding the egress IP to the IP whitelist in the Official Account backend.
Read MoreText-to-Image for Interactive Fiction: A Skill for Directly Generating Character Art, Backgrounds, and UI
Direct generation of all-style text-to-image assets for interactive fiction/reading: character art, backgrounds, CGs, covers, and UI. Features accurate historical research on regulations from 14+ dynasties, generated by a high-performance text-to-image model, priced at ¥1 per image. Released by Pixel Tide Studio. Location: @org-pwohuje9/wenyou-prompt-gen. Version: 1.0.1. Category: design-media. Paid service; no API key required.
Read MoreHR Auto Resume Screener: Structure Initial Screening Judgments into a Five-Dimension Scoring Process
HR Resume Auto-Screener (@user_8d7aa34e/hr-resume-auto-screener, v1.0.0, community source, business-ops category, requires_api_key=false) is a resume pre-screening skill that encodes the decision-making models of senior HR professionals into SKILL.md. It begins with structured parsing to extract fields such as name, contact information, educational background, work experience, project experience, and skill stack. It then runs a five-dimensional, role-specific scoring system — Frontend Engineer scores on tech-stack match, project depth, academic credentials and learning ability, job stability, and engineering practices; New Media Operations scores on content creation, data sensitivity, platform experience, internet savvy, and overall competency — with scoring anchors and positive/negative signal detection in each dimension. Beyond scoring, a rigorous review layer checks for timeline contradictions, credential anomalies, skill stuffing, and insufficiently quantified achievements; any flag triggers a red-flag warning. Finally, based on a weighted total score, it outputs one of three verdicts — Recommend for Interview, Pending, or Not Recommended — and generates 3–5 interview probes targeting the resume's gaps, returning everything as pure JSON for easy integration into existing recruitment systems. Use cases: initial HR screening, headhunter evaluations, and self-assessment by job seekers. Limitations: the scoring system covers only Frontend Engineer and New Media Operations roles; without anchors for other roles, verdicts would be unreliable. As it is v1.0.0, community-sourced, with zero installs, it is recommended to first benchmark against resumes with known outcomes before going live.
Read Moredsh-qoder-link: Connect to DSH via Official Qoder CLI
dsh-qoder-link is a DSH client plugin maintained by liubaicai. It connects qodercli (international version) and qoderclicn (China version) via ACP stdio, reuses the CLI's local login state, and provides the qoder model provider and the qoder_ask delegation tool within DSH. The article covers installing the CLI, installing the plugin via a Release tgz package, selecting region and model through Web settings, and the security boundary under which tools are validated and executed by DSH. It also specifies the Node.js ≥22.18 requirement, configuration paths, and the /qoder command.
Read MoreExpert Panel on Smart Hardware Development: Deconstructing Hardware PRDs into Executable Schedules, BOMs, and Test Plans
This article introduces the SkillHub community Skill "Intelligent Hardware Development Expert Team" (@user_07354da8/intelligent-hardware-expert-team, version 1.2.1, category: professional, requires_api_key=false). This Skill targets the entire smart hardware development lifecycle, providing capabilities including PRD decomposition, BOM compilation, injection mold cost estimation, EVT/DVT/PVT/MP project scheduling, DFM review, reliability test planning, certification checklists, data desensitization, and multi-expert collaboration. For specific tasks, it recommends specialized assistants within its family (scheduling/BOM/issue tracking/risk management/procurement price comparison). The article addresses pain points in hardware development scheduling, BOM management, tooling, and testing, presenting core functions, coordinates and official pages, typical natural language usage patterns, applicable scenarios and limitations (not for non-hardware projects, does not replace engineer sign-off, cost and certification are estimates requiring verification, specific installation commands should refer to official pages), with official links provided.
Read MoreImage-to-PPT: Generating High-Fidelity Slides with gpt-image-2
Image-to-PPT (@user_afd1bce5/image-to-ppt, v1.0.0, community-source, office-efficiency) calls OpenAI gpt-image-2 to render each slide as an image, with styles drawn from built-in styles/<id>.md or mimicking a user-provided .pptx template, and outputs high-resolution slide PNGs and a 16:9 .pptx. It is suitable for scenarios requiring high visual finishing and accepting bitmap-based page content, such as pitch decks, investor PPTs, and magazine-style slides. A OpenAI API key is required (requires_api_key=true). This article describes its positioning, core capabilities, verifiable metadata, typical use, and limitations; the official page body text was not retrieved, installation commands and configuration fields were not verified, and no speculative approaches are provided in the text.
Read Moredsh-voice: Enabling Harness Agents to Speak and Sing with Emotion
dsh-voice is a DeepSeek Harness Host-only plugin maintained by LIU118117, registering voice_speak, voice_sing, and voice_status. It supports six emotional tones and various voice types; singing utilizes per-character pitch modulation via edge-tts and concatenation via ffmpeg. It defaults to local ChatTTS inference, falling back to edge-tts if unavailable. To install, use plugin_manager install_bundle pointing to the absolute path of the plugin directory; refer to cordis.patch.yml for configuration. It is licensed under MIT, with synthesis performed locally or via minimal text upload required by edge-tts.
Read MoreFull-Text Generation of Appeal Petitions: One-Stop Complete Output of Civil Second-Instance Appeal Documents
Appeal Document Full-Text Generation is a community skill on SkillHub, located at @user_b8af0005/appeal-docs-generator. Categorized as "professional," it is at version 1.0.5 with the tag requires_api_key=false, and is published and maintained by @user_b8af0005. Targeting the preparation phase for civil second-instance appeals, the workflow involves inputting a first-instance judgment and conducting a "six-dimensional diagnosis" based on the Civil Procedure Law. This covers six aspects: appeal deadlines, procedural subjects, comparison of claims and judgments, syllogistic logic, legal reasoning, and new elements. It then retrieves currently valid legal provisions and similar cases via the Meta-judgment connector. Before drafting, it engages in a fact-verification interaction with the client to confirm whether factual errors have been acknowledged item by item and to check for new evidence. Upon confirmation, it outputs a complete package in one go: the Civil Appeal Petition, the Second-Instance Evidence List, a Research Report on Legal Provisions and Similar Cases supporting the appeal, the Second-Instance Power of Attorney, the Letter of Introduction (Law Firm), the Second-Instance Interview Transcript, and a List of Facts Pending Confirmation. The appeal petition strictly adheres to the DeHeng format template—using SimHei 18pt for centered titles, FangSong 14pt for body text, and 1.5x line spacing—which the developers claim ensures 100% format fidelity. Additionally, a "scan" light-weight output is available, which solely performs appeal feasibility assessments and calculates appeal costs without generating the full set of documents; this is suitable for initially determining whether an appeal is worthwhile. Since the provided materials do not include CLI installation commands, directory conventions, configuration file fields, or command-line examples, this article lists reproducible steps based on the workflow without fabricating commands.
Read Moredsh-love: Distill Everyday Chats into Updatable Character Cards and Cross-Session Affective Modules
dsh-love is a DSH Host-only memory plugin that records QQ/session chat samples locally. It utilizes love_distill to distill speaking habits and key memories into card.json during desktop sessions, and maintains cross-session emotions and relationship intensity via emotion.json. The compressed content is injected into every assembly via the system prompt section and can be used in conjunction with the SnowLuma QQ Bridge. To install, use `plugin_manager install_bundle` pointing to the repository path; data is located in DSH_HOME/dsh-love by default.
Read MoreComprehensive Official Document Writing Skill (Lite): Consolidating Scattered Materials into Deliverable Formal Documents
The "Full-Spectrum Official Document Writing Skill (Lite Version)" is a paid writing tool released by Shangyan Information on SkillHub. Designed for office administrators, secretaries, and drafters, it covers 38 types of official documents, including requests for instructions, reports, summaries, speeches, and minutes. It supports drafting, rewriting, polishing, expanding, condensing, title optimization, structural adjustment, tone unification, and content review. The system profiles the target audience to adapt the writing style accordingly. Writing knowledge is fully retrieved via an online API, and missing data such as figures, document numbers, names, and dates are uniformly marked with "[To be filled]." The final output is delivered as a .docx file compliant with the GB/T 9704 standard. This article introduces the skill's core features, installation and activation methods, typical workflows, and usage considerations.
Read Moreima-skill: A Skill Unifying IMA Note Management and Knowledge Base Operations
ima-skill (Pro Edition) is an Agent Skill maintained by user_87f6522b, a publisher in the SkillHub community. Based on the universal SKILL.md format, it is positioned as a unified IMA OpenAPI skill that integrates note management (searching, browsing, creating, and editing) and knowledge base operations (uploading files, adding webpages, and searching content) into a single entry point. It addresses the issue where Agents fail to determine whether a user request—such as "note this down for me" or "search the knowledge base for XX"—falls under note management or knowledge base operations, and consequently cannot decide which interface to invoke. The skill solidifies intent recognition and operation dispatching, stipulating that a one-sentence conclusion must be provided before elaboration, structured result files must be delivered, and the basis for key judgments must be annotated. The documentation indicates `requires_api_key=true`, meaning an IMA OpenAPI key must be prepared before enabling the skill. However, as the pre-fetched data does not provide specific installation commands or configuration fields, the main text does not fabricate them. The documentation lists 77 downloads, 0 installations, and a rating of 100000; the small sample size indicates that boundary issues should be verified by consulting the interface documentation independently.
Read MorePublic Opinion Evolution Simulation: A Community-Sourced Data Analysis Skill
This article introduces the community-sourced Skill "Public Sentiment Evolution Deduction," located at @user_0ccfebe0/yuqing-tuiyan, version 1.0.0, categorized as data-analysis, and tagged with requires_api_key=false. It has a page rating of 100000, 888 downloads, and 0 installations. Since public information only provides metadata such as name, category, and version, without functional descriptions, installation commands, or examples, the article focuses solely on verifiable content: what it is, who produced it, verified capability boundaries, installation status, why typical usage scenarios are missing, and key verification points to consider before integrating community-sourced Skills. The article concludes by providing the official address and the SkillHub page link, without making any capability inferences.
Read Moresearch-school-sci-skill: Organizing Academic Research Output into Excel Visualizations
search-school-sci-skill is a DSH network tool plugin maintained by liu-Daniel7. By integrating OpenAlex with optional CNKI, it writes statistics such as publication attribution, quartile rankings, and student inference for specified university colleges into a 7-sheet visualization Excel workbook. The installation command is `dsh plugin add liu-Daniel7/search-school-sci-skill`; the typical workflow involves generating configuration via the wizard, fetching faculty data with faculty_fetch, conducting a census, and then generating reports with college_sci. The article outlines considerations including methodological scope limitations, permissions, and OpenAlex quota, and provides links to SkillHub and GitHub.
Read MoreCocos Project Scaffolding: Porting Pure Logic Code into Cocos Creator 3.8.x
Community Skill, ID @user_4edac78e/cocos-project-scaffold, version 1.0.0, category dev-programming, source community, no API key required. Intended for creating new or migrating Cocos Creator 3.8.x TypeScript projects, covering seven capabilities: project directory structure; BattleSim pure-logic decoupling; offline type checking with cc stubs; scene authoring by modifying the built-in scene-2d.scene template to avoid hand-writing uuid; mapping between top-left origin and center origin coordinate systems; designResolution configuration; and procedural vector rendering. To enable, follow the general SKILL.md format; refer to the official page for installation commands. Applicable scenarios and limitations are also listed.
Read Moredsh-quick-notes: Floating Notes on the DSH Web Interface
dsh-quick-notes is a floating note plugin (npm v1.0.0, MIT license) developed by LittleLemonJam for the DeepSeek Harness Web GUI. It supports multiple note directories, auto-save, a trash bin, basic formatting, and image pasting. All data is stored locally at ~/.dsh/quick-notes/notes.json and not uploaded to any server. Requires DSH ≥0.2.0. It can be installed by entering "dsh-quick-notes" on the Plugins page or via the command `dsh plugin --profile web add dsh-quick-notes`, and opened from the note icon in the left sidebar.
Read MoreAcademic-Radar: Transforming Multi-Source Academic Literature Monitoring into a Traceable Pipeline
Academic-Radar is a knowledge management Skill released by community author user_9e10d3ea (version 1.0.0), designed for clinicians, medical researchers, and graduate students. It periodically fetches new content from PubMed, Semantic Scholar, OpenAlex, bioRxiv, medRxiv, arXiv, X, and RSS based on research directions. After keyword filtering, cross-cycle deduplication, and LLM-based relevance scoring, it generates traceable reports. Core capabilities include six-layer keyword configuration, author tracking, retrieval frequency adjustment, and multi-channel push notifications. A key constraint is that bibliographic information is sourced from original APIs, while the model is solely responsible for scoring and summarizing. The Skill guides users through five sequential steps: direction decomposition, keyword calibration, model and channel configuration, initial execution, and setting up scheduled tasks. It carries a requires_api_key label, meaning users must prepare their own data source credentials. This summary is compiled based on data retrieved from the official page; specific installation commands are not provided here, so please refer to the official page instructions.
Read More