dsh-retry: Add a native "Restart DeepSeek Harness" entry to the DSH Desktop
**dsh-retry** is a DSH client plugin maintained by kouzerror that brings "Restart DeepSeek Harness" functionality to the Windows desktop. Tray and account menu items are implemented as native entries via **app.asar** patching, while the "Restart & Update" button in the settings page utilizes the official slot, ensuring the restart capability persists even after DSH upgrades. All three entry points are unified via a `POST /dsh-retry/restart` request, supporting graceful relaunches and automatic helper reinstallation if patches are missing. Installation is available via the plugin marketplace or by running `pnpm add file:` in the profile directory; after enabling, users must fully exit and restart DSH, allowing the plugin to self-heal and install patches within approximately three seconds. Based on the repository's README and package.json (v0.6.1), this article outlines installation procedures, status self-checks, and important considerations.
Read MoreEnabling Agent Support for MiniMax H3 Model: Introduction to the Hailuo-03 Video Generation Skill
This article introduces the community Skill "Enable Agent to support minimaxH3 model" (Coordinate: @user_b5df6666/minimax-h3-video, v1.0.0, community source, category: ai-agent). This Skill invokes the MiniMax H3 (Hailuo-03) video generation API to transform text, image, video, and audio inputs into videos with native audio. It supports two resolution tiers (768P/2K) and durations of 4-15 seconds. It covers text-to-video, image-to-video (first frame/last frame/first and last frames), multimodal reference-to-video, 768P to 2K regeneration, and H3-Context-IR prompt enhancement, while also providing API Key configuration, model querying, and price/cost estimation. The article provides a capability list, activation methods, typical invocation workflows, and usage notes, and is marked with requires_api_key=true.
Read MoreAladdin: Hit Web Novel Creation Workbench — An Offline Web Novel Pipeline from Topic Selection to the "Golden Three Chapters"
Aladdin Web Novel Hit Creation Workbench is a community-sourced universal Skill conforming to the SKILL.md format. It enables the offline, one-click development of genre directions, keywords, or setting drafts into web novel creation schemes ready for platform submission. It covers genre and tag selection (matching the tonality of Fanqie, Qidian, Feilu, and Jinjiang), worldbuilding setting cards, character cards (protagonist/antagonist/supporting cast + relationship networks + growth arcs + satisfaction point allocation), comprehensive book outlines (volume and chapter outlines + satisfaction point distribution + chapter-end hooks), 'Golden Three Chapters' drafts, pacing diagnostics (satisfaction density/hook density/emotional curve/padding detection), platform compliance scanning (prohibited words/bad ending risks/opening deterrence), and a quality scoring gate ranging from 0-100 (Grades A-D). Compared to solutions that only generate single chapters or provide mere character frameworks, it adds three core features: genre selection, global satisfaction distribution, and dual inspection for pacing and compliance. It operates entirely locally with zero API dependency and requires no keys; it is ready to use immediately upon installation.
Read Moredsh-skill-rag: Configurable DSH Skill Discovery and Retrieval
The Kosello DSH workflow plugin dsh-skill-rag offers three modes—catalog, local SQLite keyword, and hybrid retrieval—without modifying the native skill loader. The default catalog mode upon installation requires no API keys; the optional hybrid mode supports Jetson, Google, OpenRouter, or OpenAI-compatible embeddings. Models retrieve summaries via skill_search before loading the skill, while humans can maintain the index using /skill-search. To install, run: dsh plugin --profile web add github:Kosello/dsh-skill-rag.
Read More🎬 Video Audio to Text: Generating Clean Transcripts with the Qwen LLM
This article introduces the community Skill "🎬 Video Voice to Text" (@user_fb71c45d/shipin-yuyin-zhuanhuan-wenzi, Version 1.0.0, Category: content-creation). It leverages the Qwen large language model to perform deep semantic parsing on online video links and local video files. While transcribing speech, it removes invalid content such as filler words, slips of the tongue, and repetitive lines, while also supporting customized copywriting styles. It is suitable for scenarios including short video secondary creation, meeting minutes, course analysis, livestream reviews, and interview organization. Based on the standard SKILL.md format, the skill is marked with requires_api_key=true and necessitates the configuration of a Qwen large model API key. Due to a failed capture of the official page during the writing process, installation commands and invocation parameters remain unverified; no data has been fabricated.
Read Moredsh-self-evolve: Procedural Memory with Guardrails for DeepSeek Harness
Kosello's dsh-self-evolve is a DSH memory-class plugin, forked from jianghuife/dsh-self-evolve and targeting DSH 0.2.1-rc.2. The agent writes SKILL.md files into user-level and project-level skills directories through a protected skill_manage, along with read-before-write checks, duplicate rejection, credential scanning, and optional write approval. After an optional turn, a reviewer with Jev gating automatically patches or creates skills, while a Curator handles aging and archiving. Based on the README, this article introduces the installation command, the cordis.patch.yml configuration, the /skills command, and applicable boundaries.
Read MoreOrangeCRDE PPT Generator: Dark Tech Style PPT Generation Skill
OrangeCRDE PPT Generator is an Agent-oriented Skill in SKILL.md format, maintained by community publisher user_3c7278bf and categorized under office-efficiency without requiring an API key. It standardizes the OrangeCRDE brand visual guidelines, featuring a color palette of dark navy blue (#0A0F2E), orange (#FC5900), and cyan (#00C8FF), the Microsoft YaHei font, and a 10×5.625in 16:9 layout. The cover, closing, quote, and QR code pages utilize a dark tech aesthetic paired with official product hero images and gradient overlays, while a "Zhicheng Platform" bar with the official logo is embedded at the bottom of every page. It provides 16 page types (cover, section_title, content_overview, content_features, content_with_image, table, flow, chart, comparison, timeline, gallery, quote, team, qrcode, closing, text_content), supporting chart generation, content overflow detection, and speaker notes. Built upon python-pptx and PIL, its usage involves writing an outline as a list of page types, enabling the Agent to generate the presentation page by page with each page's type explicitly specified. The article also details applicable scenarios and limitations: the style is strictly locked to official color schemes, images rely solely on official hero visuals, overflow detection only identifies issues without automatic rearrangement, and installation volume data is based on a small sample size, necessitating prior verification.
Read MoreLingXi Financial Search (Guotai Haitong): A Financial Search Skill on SkillHub
This article introduces "Lingxi Financial Search (Guotai Haitong)" on SkillHub, an Agent Skill published by user_6399968a, sourced from the community, and categorized as professional. Operating at version 2.0.2, it requires users to provide their own API Key. Due to the failure to scrape the official page, the article only lists information verifiable from the snapshot (Score: 100,000; Downloads: 962; Installations: 0; Description: only one sentence). Installation commands and invocation examples are not provided, and users are advised to independently verify the Key, quotas, and return quality prior to integration.
Read MoreApple mandates an 18+ age rating for dating and AI companion apps containing romantic or sexual content.
On November 3, 2026, Apple announced App Store policy updates to developers: Apps offering dating or romantic relationship features, as well as AI companion capabilities involving sexual/romantic content, must use the 18+ age rating; starting November 16, new content descriptors must be added. The new rules are advancing alongside the global rollout of age assurance and child protection features, directly affecting the distribution and metadata compliance of companion apps like Character.AI and Replika on iOS.
Read Moredsh-offwork-countdown: A Draggable Off-Work Countdown Card for the DSH Web Interface
dsh-offwork-countdown is a DSH client (web profile) plugin maintained by konwait12, displaying a draggable and configurable off-work countdown card in the top-right corner of the DeepSeek Harness Web GUI. It features an HH:MM:SS countdown, a today's progress bar, and inline clock-in/out settings; supports collapsing into a small capsule, right-side panel integration, and stores data only in local localStorage. The installation command is `dsh plugin --profile web add github:konwait12/dsh-offwork-countdown`. Verified on DSH 0.2.1-rc.2 web profile, MIT licensed.
Read MoreGoogle defaults to off for Gemini and Search on K-12 student accounts, preventing schools from retaining Search access when AI is disabled district-wide.
On November 3, 2026, Google's Education VP Chris Phillips announced that, after consulting with schools, researchers, and policymakers, K-12 student accounts will have access to Gemini, Gemini Notebook, Gemini in Classroom, and Google Search disabled by default. District admins must explicitly enable them. At the same time, Google is promoting Guided Learning in Search and Classroom. TechCrunch, citing Phillips' remarks to the NYT, reported that Search is deeply integrated with AI; if schools block AI across the board, students will be unable to use Search. Officials said additional rollout details will be shared in the coming months.
Read MoreAI Assistant for Preparation, Review, and Control of Steel Structure Safety Plans - "Peng Ge Liao Gongcheng": Turning Experience into Structured Review
A Q&A and review assistant Skill for whole-process management of dangerous and major steel structure engineering projects, released by the SkillHub community, covering five capability categories: hazard identification, plan preparation, plan review, on-site hazard detection, and quality common defect prevention. Based on official documentation, the article verifies the standard input format, typical question patterns, three output structures, and six usage boundaries, and clarifies that it cannot replace expert review conclusions.
Read Moredsh-connect-two: Unified Check-in and Points Dashboard for Trae and WorkBuddy
**dsh-connect-two** is a DSH networking tool plugin maintained by **konwait12** (version **0.1.1**, MIT license). It does **not** replace **dsh-connect-trae** and **dsh-connect-workbuddy**; instead, it **centrally displays** the login, credits, and check-in status of the **four sections** — Trae/WorkBuddy **Domestic** and **Global** — within **Settings → Check-in & Credits**. It also supports **optional daily automatic check-in** and **check-in logs**. - **Domestic side**: supports both **one-click** and **single-line** check-in. - **Global side**: supports **credits only**. **Installation**: - Via **plugin_manager**: use `install_bundle`. - After npm publication: use `dsh plugin add dsh-connect-two`. **Note**: the **Host** side requires a restart of **dsh web**. State is saved in `$DSH_HOME/dsh-connect-two/state.json`.
Read MoreSpaced Repetition Flashcard Generation: Converting Any Content into Reviewable Cards
# Introduction A common pain point when organizing learning materials is: after reading a document, a note, or a slide, the key concepts that truly need to be remembered are scattered throughout the original text. Making flashcards manually is time-consuming; after completion, they need to be imported into spaced repetition tools like Anki, and the format needs to be checked again. The approach of this Skill is to hand over the step of "extracting key concepts from content" to the model to complete, directly producing reviewable flashcards instead of just providing a summary. # What is this Interval Memory Flashcard Generation | Jianshi AI, coordinates @user_fccf428b/flashcard-generator, is a community Skill based on the general SKILL.md format. Published by user_fccf428b, source marked as community, current version
Read MoreGrant Application Skill: Organizing Application Guidelines into an Evaluatable Application Plan
The Research Grant Application Skill is published by SkillHub user user_b4da0893, located at @user_b4da0893/research-grant-cn, version 1.0.0, category professional, source community, and no API Key configuration is required. It targets national, local, university, hospital, foundation, and enterprise research projects; it first freezes the current call guidelines and eligibility requirements, and then establishes a scientific question–objectives–tasks–evidence–milestones chain. It covers topic-to-call matching, proposal planning, research objective breakdown, technical roadmap, feasibility, budget justification, compliance checklists, mock review and revision, and identifies four common issues: overly broad scope, empty innovation, insufficient preliminary work, and budget mismatch. The article describes its positioning, verified capabilities, location and configuration information, typical use organized according to official processes, and limitations such as not fabricating materials, not guaranteeing funding approval, and not replacing formal review, and provides the official SkillHub address.
Read MoreCoding Development: Transforming I11 Intent into a Complete Coding Workflow
Coding Development is a downstream execution skill within the tri-intent system. It reads Snapshot §III to process the I11 Coding Development intent and autonomously manages the complete pipeline of "Requirements → Design → Tasks → Execution → Implementation Report". Featuring dual approval gates and pre-execution confirmation, it supports standalone installation and provides upstream dependency detection with three-state logic (snapshot mode/guided installation/degraded mode), requiring no API key. This article outlines its positioning, core capabilities, installation prerequisites, usage conditions, and limitations, along with its official URL.
Read MoreOpenAI Showcases End-to-End Hardware Engineering for GPT-6 Astra and Collaboration with Mazda on Simulation Research
On November 3, 2026, OpenAI published a blog post explaining the model's usage in CAD, simulation, PCB, and interdisciplinary system design. It showcased Astra's prototype device capable of completing conversational Codex without direct human editing during testing, while acknowledging that mass production still lacks DFM and tolerance considerations. The hardware team's token usage has increased approximately 23 times since February. On BenchCAD 2.0, the newer internal model achieved 0.875, while GPT-6 Astra reached a maximum inference of 0.594. Simultaneously, the company announced research collaboration with Mazda on aerodynamic and crash safety simulations, and expanded its scanning-to-CAD workflow with Chip Ganassi Racing.
Read MoreNews Sentiment Alpha Research: Operationalizing News Sentiment into a Reproducible Cross-Sectional Stock Selection Factor
**# Preface For those involved in stock selection factor creation, news sentiment data is often already in hand. However, practical implementation typically stalls at a few key points: aligning sentiment values with trading days, generating recursive time-series factors, and efficiently executing and archiving IC analysis, bucketing, and drawdown calculations once factors are synthesized. This Skill consolidates these steps into a reproducible workflow.** **# What is this News Sentiment Alpha Research, located at `@zerone-iteration/news-sentiment-alpha`, is published by Lingyi Intelligent. It is categorized as an enterprise source type under the data-analysis classification, with a current version of 1.0.1. According to SkillHub page data, it has a score of 100,000 and 968 downloads, though specific figures are subject to real-time page updates.** **One-sentence Positioning**
Read Moredsh-cookie-guard: Periodic cleanup of dsh-auth cookies to avoid 431 authentication failures in DSH Desktop
The DSH desktop version (Windows) suffers from an accumulation of dsh-auth cookies under 127.0.0.1 due to random ports, where request headers exceeding 16 KB trigger a 431 error (Web authentication failed). kongbaiwin's dsh-cookie-guard schedules cleaner.mjs via scheduled tasks and login shims to clean entries by prefix or optionally replace the entire database when the database is not locked. It also provides a settings page, backup, and logging features. It only supports the Windows desktop profile and should be configured under "Settings → Cookie Cleanup" after installation.
Read Moredsh-isaac-widget: DSH Bottom Right Corner Isaac Transparent Animation Widget
dsh-isaac-widget is a universal bottom-right corner widget for DSH, maintained by komeiji-cobalt and based on a modified version of dsh-whale-widget. It features a 512×400 transparent Isaac GIF, random HurtGrunt sounds and Isaac Dies (with increasing probability) on click, and 8 Isaac-themed clickable bubbles (default rotation, including Lost, balance, crying bombs, etc.). It retains the balance display and per-round consumption tips. The documentation covers Web/desktop installation commands, menu usage, the behavior of writing to $DSH_HOME for the first time, as well as licensing and asset declarations.
Read MoreMarketing Plan: Systematic Product Marketing Planning with the 7P Framework
marketing-plan is an Agent Skill built on the 7P marketing mix framework. It breaks a product marketing plan into eight stages—executive summary, situation overview, goal setting, target market strategy, value proposition, 7P tactical mix, implementation plan, and control & evaluation—guiding users to produce a complete plan item by item rather than generating a long-form document in one go. This Skill was published by community publisher user_08e29a08 on SkillHub at the coordinate @user_08e29a08/yr0001. Version 1.0.0. Category: business-ops. Source: community. Tag: requires_api_key=false (no API key required). It is triggered when a user needs to write a marketing plan, marketing program, product launch plan, go-to-market strategy, marketing strategy plan, or mentions terms such as “营销计划,” “market plan,” “marketing plan,” “上市方案,” or “推广方案.” At the time of capture, the page showed a rating of 100,000 and 916 downloads. No command-line installation command is provided in the materials; obtain it by locating the Skill on SkillHub using the coordinate.
Read Moredsh-fx-arena: Local FX Simulated Battles within the DSH Client
dsh-fx-arena (FX Fighter Dafeiyu) is a DSH client plugin maintained by koishi-fumo-code that lets you practice FX simulation with a local algorithmic opponent, 鲸鲸, in a left-side panel. It supports five instruments, 10–500x leverage, timed races, or an unlimited mode where you try to turn 300,000 into 20 million. Features include Canvas-based candlestick charts, rules for liquidation and negative-balance exit, background continuation, and localStorage-backed performance records. The entire experience is offline and uses no real funds. Installation is handled via scripts/install.mjs, which registers a profile bundle; it requires Node 22+ and DSH >= 0.1.7-rc.1. Restart DSH after installing. Intended for local downtime practice and bundle verification; prices are simulated via a random walk and do not substitute for live trading.
Read MoreShort Video Topic Library | Guo Yiyi: Handing Off Topic Pool Layering, Hooks, and Validation Priorities to Skill
A community skill for short-video planning and account operations at @user_8d36cde0/guoyiyi-short-video-topic-bank, version 1.0.0, categorized under content-creation, sourced from the community, rating 100000, with 927 downloads. Its tag indicates requires_api_key=false, meaning no external API key is needed. It takes three types of inputs—audience problems, product selling points, and historical content performance—and produces a tiered topic bank, hook angles, and a validation priority. It aims to fix the issues of choosing topics by gut feeling and sequencing validation by intuition. The material does not provide official installation commands or differences among Cursor/Codex CLI/Claude Code; it explicitly states that unverified commands are not listed and recommends checking the official page first. The article presents a reproducible input–output structure, notes that it only handles topic selection and validation order (not scripts or production), and that output quality depends on how specific the inputs are. As a community release, it advises running a small-scope trial before going live.
Read MoreAI Preparation of Pharmaceutical Financial Accounting Vouchers: Streamlining Pharmaceutical Industry Vouchers into an Automated Pipeline
The accounting voucher challenges for pharmaceutical enterprises stem from the requirement to simultaneously comply with three sets of standards—accounting, industry regulatory, and tax—for the same batch of business operations. Chemical drugs, Traditional Chinese Medicine (TCM) decoction pieces, and biological products are subject to different tax rates; R&D expenditures require capitalization judgment; academic promotion fees must be traceable; yet general bookkeeping skills only ensure debit-credit balance. The "Pharmaceutical Financial Accounting Voucher AI Preparation" (@baitongai/pharmaceutical-financial-ai-accounting-voucher-preparation, Version 1.0.0, released by Baitong AI Service Platform) solidifies this workflow into a process comprising source data parsing, intelligent account mapping, journal entry generation, compliance verification, voucher lifecycle management, and formatted output. It features built-in GSP/GMP compliance checks, R&D expenditure capitalization judgment, price spread checks for the "Two-Invoice System," differentiated drug tax rates, academic promotion fee monitoring, and cold chain batch tracking, while supporting import formats for Kingdee, Yonyou, SAP, Oracle, and others. This article outlines the functional scope, verifiable identification information, pre-call data preparation, and precautions regarding compliance review, tax rate adoption, and missing installation information.
Read MoreVIP Essay Detection: Delegating the VIP Plagiarism Check Workflow to an Agent Skill
This article presents the Agent Skill “VIP Paper Check” (@org-w5ehy5pt/vip-paper-check). Released by VIP Zhunzhi, the Skill originates from enterprise sources, falls under the professional category, and is currently at version 1.0.4. It requires no API Key. The tool accepts papers in DOC, DOCX, and PDF formats. It can query currently supported detection services and their real-time pricing. Once the user confirms, the Skill assists in account login, order placement, WeChat Pay processing, and report retrieval. Typical applications include plagiarism detection, AIGC detection, format checking, AI-based intelligent review, quality evaluation, data and image credibility detection, and proofreading. The article outlines the core functionality and typical workflow (Price inquiry → Confirmation → Order placement and payment → Report retrieval), along with precautions: enterprise publishers must confirm their affiliation, transactions involving real payments require user confirmation, prices are subject to real-time fluctuations, and no speculation will be made regarding unverified content on official pages.
Read More