dsh-context-relay: Keeping Lead in the Efficient Context Window via Phase Protocol and Bounded Handoff
dsh-context-relay is a DSH workflow plugin maintained by rezon-aki (v0.2.11, MIT), which registers a Relay mode in the Web profile: a phase state machine, budget sentinel, controlled compression, and bounded handoff to open new sessions. Together with tools such as relay_status, relay_phase, relay_organize, and relay_handoff, it makes decisions on compression or handoff at phase boundaries. The installation command is `dsh plugin --profile web add dsh-context-relay`, requiring DSH ≥ 0.2.1-rc.2 and Node.js ≥ 20.
Read MoreInvestment Team Skill: A Framework for Parallel Analysis by Four Roles in Investment Research
"Investment Team" is a community Skill on SkillHub located at @user_3c6cb52e/investment-team-ccey. It is Version 1.0.0, categorized under "professional," and published by user_3c6cb52e. It is positioned as an "AI Berkshire skill: Investment research team: Four-role parallel analysis framework," with the source file located at skills/investment-team.md. Starting with the pain point of using a single model to digest materials, the article clarifies that it confirms only one capability: a four-role parallel analysis framework. However, the specific four roles and the aggregation method are not explained on the page. The page indicates requires_api_key=false. The installation command is not publicly disclosed; the article only lists the coordinates and two official addresses without providing unverified commands, advising users to check the installation instructions for their respective clients. The "Typical Usage" section is explicitly left blank due to the absence of official examples, with a suggestion to read the Source file directly. Precautions include: it is community-published and not officially maintained; the parallel scheduling mechanism is unspecified; parallel operation imposes requirements on context and scheduling; and with zero installations currently, it lacks verification. The article concludes by providing the official addresses.
Read MoreGenerating Academic Posters from Research Papers: Converting PDF Papers into Bilingual Academic Posters
This is an Agent Skill designed for developers and researchers (@user_f7a184a2/pdf-to-academic-poster, v1.0.0, community-sourced, categorized under knowledge-management). It is triggered when a user provides a PDF academic paper and requests translation and/or poster generation. The workflow first utilizes PyMuPDF to extract text from the PDF, then analyzes the article content, and finally generates a professional bilingual (Chinese-English) academic poster. The output is a single self-contained HTML file that can be opened directly in a browser or saved as a PDF for printing. The official metadata lists requires_api_key=false. Trigger phrases include "翻译文章并生成学术海报" (translate article and generate academic poster), "生成学术海报" (generate academic poster), "PDF转海报" (PDF to poster), "pdf to poster", and "academic poster from pdf". Points to note: The version is low and installation count is currently zero. PyMuPDF cannot extract text from scanned documents without a text layer; OCR must be performed beforehand. The quality of the summary depends on the associated model, so manual proofreading is recommended for complex papers. Strict layout correspondence between Chinese and English is not guaranteed. As the official page failed to load during documentation writing, please refer to the official source for installation commands.
Read MoreAuditing Custom Agents: Structured Audit and Improvement
# Introduction Once a Large Language Model (LLM) is integrated into an Agent, the true challenge is not simply ensuring it runs, but enabling it to identify the source of its errors when it malfunctions. The common approach for most teams involves digging through logs upon encountering issues and tweaking prompts based on intuition, often without any certainty that the problem has actually been resolved. The "Audit Custom Agent" Skill offers an alternative pathway: it begins by conducting a structured audit of the existing Agent to generate a list of issues and an improvement plan, followed by the implementation of that plan. The following section outlines the positioning, verified metadata, capabilities, and limitations of this Skill. It should be noted that this article is based on page materials retrieved from SkillHub. As the official page failed to load during two separate attempts while writing, certain promotional details could not be verified and have been treated as such. # What is this? Auditing Custom Agents
Read Moredsh-model-manager: Unifying Multi-Vendor Models, Quota, and Actual Usage Into a Single Dashboard
dsh-model-manager is a DSH model inference plugin (v1.4.1, MIT) maintained by Rexray117. The sidebar provides full-page "Model Management": browse and search models by provider, mask or restore dropdown menu items, and configure secure defaults. It summarizes remaining channel credits, DeepSeek wallet balances, and opencode Go plan quotas. It also provides accurate token usage statistics based on session logs in ~/.dsh/sessions, categorized by hour, day, week, half-month, and month. Installation is performed by unpacking the Releases archive and executing `plugin_manager action=install_bundle`. All data is accessed locally for reading and writing, with no uploading.
Read Moredsh-qq: Enable DSH Agent to Send/Receive Messages and Auto-Reply in Group Chats on Local QQ
dsh-qq is a DSH client plugin maintained by ReWebaiyu (version 0.1.0, MIT license). By leveraging local NapCat's OneBot 11 HTTP API and event push, it provides Agents with six tools: qq_status, qq_send_private, qq_send_group, qq_history, qq_backend, and qq_bot. It also offers a web-based management panel and a DSH sidebar entry at 127.0.0.1:3014, supporting persona configuration, whitelist management, usage statistics, and automatic group chat responses. The installation command is `dsh plugin add github:ReWebaiyu/dsh-qq`. Requires Node.js >= 22, with NapCat and profile patch pre-configured.
Read MoreSeewo Whiteboard Mind Map Library Expansion: Adding Structure Templates and Subject-Specific Finished Diagrams to Seewo Whiteboard 5 Software
A paid Skill for Seewo Whiteboard 5, published by Jishang Network Technology, located at @jishangtech/js-97, version 1.0.0, categorized under education, sourced from enterprise. Core capabilities include 7 structural templates (logic diagram, mind map, tree diagram, bubble map, fishbone diagram, timeline, and bracket map) plus 16 ready-made subject diagrams for primary and secondary schools, supporting one-click generation of diagrams from outlines and exporting PPTX files that can be directly imported into Seewo Whiteboard 5. No API Key required; it is a one-time purchase model: ¥4.99 for unlimited use, with 1 free use per IP. The article explains its positioning, capability boundaries, verification of installation and activation, typical usage methods and applicable scenarios, and notes that the official page failed to be retrieved at the time of writing, with ratings and download counts based on real-time page display.
Read Moredsh-p3r-theme: Applying Persona 3 Reload Style Interface to DeepSeek Harness Desktop
dsh-p3r-theme (dsh-theme-p3r 0.4.0) injects a P3R-style UI into DeepSeek Harness Desktop: it overrides DSW design tokens, features an inline ripple background, chat bubbles in white-with-blue-text and blue-with-white-text, a beveled selection plate, and a HUD bottom-edge scale. For Desktop, install from a local directory within the app and restart. To change the background, use npm run media to sync to client.js. MIT licensed. Maintainer: renge1145141919810.
Read MoreQuote Tactician: Analysis of Engineering Bidding Pricing Strategies Based on Game Theory
Bid Tactician is a community-sourced skill for analyzing engineering bidding and quotation strategies, located at @user_adf1b7b3/bid-tactician, version 2.2.0, categorized under business-ops, and requires_api_key=false. It leverages Monte Carlo simulations and the NSGA-II genetic algorithm to solve for a Pareto-optimal bidding range with dual objectives of probability of winning and profit. It supports 5 evaluation approaches including the comprehensive evaluation method, the evaluated lowest bid price method, and the reasonable low-price method, among others. It integrates Sobol sensitivity analysis, opponent bidding probability distribution modeling, and a closed-loop calibration with feedback from historical data, and outputs bidding-score curves, tornado charts, and sensitivity heatmaps. The available materials do not include installation commands or reproducible usage examples; therefore, no installation instructions or reproducible usage examples will be fabricated here. The page reports a rating of 100000, 835 downloads, and 0 installations; it should be independently verified prior to adoption.
Read MoreJD.com Self-Operated Lowest Price: Skill for Querying Historical Low Prices of Self-Operated Products
The "JD Self-operated Historical Lowest Price" is a community Skill published on SkillHub (Coordinates: @user_b95ee7e5/jd-lowest-price-new, Version 1.2.1, Category: business-ops). It encapsulates the capability to query historical lowest-priced products on JD.com, returning self-operated products currently within a historical low-price range. It limits results to quality selections with a positive rating of ≥97%. It supports category search, intelligent category recommendations, price filtering, and sorting by discount magnitude. Results are returned in pages of 50 items per page, with the page number defaulting to 1. Data is relayed via Tencent Cloud, and no API Key is required. The article explains its positioning, verified core functions, installation and activation methods (official address and coordinates; specific commands should be based on the page), and typical usage based on page number parameters. It also outlines applicable scenarios and limitations: the output is a list of historically low-priced products rather than a price history curve; category search relies on built-in recommendations; users should be mindful of speed limits when batch paging; and for page scraping, ratings and download counts should be based on real-time data.
Read Moredsh-audio-tools: Processing Audio Tracks and Transcription in DSH with ffmpeg and Whisper
dsh-audio-tools is a DeepSeek Harness community plugin maintained by Ren-xing, providing three tools—audio_probe, audio_convert, and audio_transcribe. Leveraging local ffmpeg, it reads media information, converts formats, or extracts audio tracks from videos, and uses whisper.cpp or an OpenAI-compatible API to transcribe text and SRT subtitles. The plugin has zero npm dependencies and does not bundle ffmpeg. It must be installed via CLI as a tgz package under the desktop profile and configured through cordis.patch.yml. For local CLI transcription, the setup-whisper script can be run to download models. This article consolidates installation, configuration, usage, and security boundaries based on the GitHub README.
Read Moredsh-plugin-outside-workspace: Create "Not in Workspace" Sessions in DSH
Introducing **dsh-outside-workspace** (v0.1.0, MIT), a DSH community plugin maintained by **realhenrylan**. This plugin provides **"Outside Workspace"** and **"Ungrouped"** entry points for new sessions via the "hero" area in the Web/Desktop frontend. It sets the default current working directory (cwd) to `deepseek-harness/outside-workspace` within the docs directory and intercepts the first composer input for sessions located outside the workspace. The documentation covers official `plugin_manager` installation and uninstallation commands, typical usage scenarios, and known limitations for version 1.
Read Moredsh-netqos: Prioritizing the DSH Ecosystem over Gaming via Native Windows QoS
dsh-netqos is a Windows-exclusive plugin provided by Re5tart for DSH Desktop. When specified games or software are running, it throttles or nearly disconnects Coopanion, SnowLuma, and DSH via New-NetQosPolicy, resolving bandwidth contention caused by short-video traffic spikes during group play. No proxy installation or router modifications are required. Based on the official README, this article covers installation commands, config.json fields, assistant scheduled tasks, and known limitations.
Read MoreViral Campaign Long Image Builder: Convert Campaign Copy into Editable Long Images & Posters
This article introduces a community-sourced Agent Skill, "Hot Poster Generator" (@user_4a9e056e/hot-poster-generator, version 1.0.3, category: design-media). It turns a user-provided activity copy, images, and reference styles into long images for WeChat Official Accounts, event posters, single-page posters for WeChat Moments, Xiaohongshu image sets, and nine-grid images. It supports long-image layout, color extraction, revisions, and JPG/PNG export, with no API key required. The article outlines core features, typical use cases, applicable scenarios, and limitations, and provides the official link.
Read Moredsh-persona-framework: Universal Persona Framework for How to Think, Not Who You Are
The DSH workflow plugin dsh-persona-framework (v5.18.0) provides a general-purpose persona framework that sits alongside the persona file: four-layer gate markers, an 88-module body injected by number, nine framework tools and a framework_program assembly. By default it does not overwrite the persona file; its exit passes frameworkLanguageCheck and betrayCheck. Covers positioning, hard wiring, installation (session install or install_bundle), typical self-check and health-check usage, and notes on permission review.
Read MoreCode Simplification Refactoring: Reducing Complexity Without Changing Behavior
"Code Simplification Refactoring" is a developer skill released by Kunlun Growth under the dev-programming category. It is located at @kunlungrowth/kg-code-simplification and is at version 1.0.0. It streamlines code without changing its behavior, with the goal of reducing complexity and improving readability. Its content covers three areas: identifying signals worth simplifying, making incremental changes step by step, and example patterns for common languages. Official materials do not provide specific installation commands or configuration fields, and the skill’s tag indicates requires_api_key=false, meaning no additional API key is needed. In practice, follow the approach of "first identify signals, then make incremental modifications, and verify behavior remains unchanged at each step." The official page snapshot shows a score of 100000 and 861 downloads, but the installation count is 0; community case studies cannot be verified. Before adopting it, it is recommended to validate on a small set of code first.
Read MoreFive-Layer Production Standard for Industrialized AI Animation: Decomposing Character Consistency into Verifiable Levels
The "AI Animation Industrialization Five-Layer Production Standard" (@user_a1bd57c2/ai-animation-pipeline) is a community Skill on SkillHub (Version 1.0.1, Category: design-media) that requires no API Key. With a page rating of 100000 and 844 downloads, it divides AI animation and comic drama production into six layers: L0 Worldview Bible, L1 Character Profile, L2 Storyboard Control Chart, L3 Shot Control Table, L4 Video Generation Prompts, and L5 Final Shot Acceptance. It is designed for requests such as creating character bibles, designing storyboards, scripting shot lists, writing generation prompts for Veo, Kling, Sora, Jimeng, and Keling, and building animation IP production pipelines. Its core objective is to ensure cross-shot character consistency and prevent character drift. The article introduces the hierarchical structure, trigger scenarios, installation access, and usage limitations, while honestly noting that the official page did not return specific installation commands.
Read MoreHandDraw: Transforming Markdown / PDF / Word Documents into Hand-Drawn Style Cognitive Diagrams
handdraw is a community-sourced Skill located at @user_9e863bf7/handdraw (v1.0.2). Categorized under design-media, it requires no API key. It automatically converts Markdown, PDF, and Word documents into hand-drawn style visual cognitive diagrams. The output is a self-contained HTML file featuring hand-drawn rendering and export functions, supporting structures such as radial diagrams, flowcharts, knowledge trees, comparison charts, timelines, and matrix diagrams. It is suited for scenarios like study notes and document-to-diagram conversion. The output needs to be opened in a browser and is not suitable for precise typography or layout. Note that the official page fetch failed during this session, so installation commands and details require independent verification against the original source.
Read Moredsh-topbar-restart: One-Click Restart for the DSH Windows Desktop Title Bar
`dsh-topbar-restart` is a DSH client plugin maintained by Re-qi (npm 1.0.3, MIT) that adds a "Restart" button exclusively to the left of the minimize button in the Windows desktop title bar. A single click terminates the entire DSH process tree via a Host route and an external cmd helper, then relaunches the same executable in approximately 4–6 seconds. It can be installed via marketplace search or by running `dsh plugin --profile desktop add dsh-topbar-restart`. The button does not appear in the browser client. It performs a forced kill without replaying CLI arguments and includes TEMP logging and a dry-run probe. Tested and verified on DSH 0.2.1-rc.3 desktop.
Read Morehtml-to-editable-pptx: Converting HTML Slides to Editable .pptx
html-to-editable-pptx is a community Skill released by user_6cf958a6, categorized under office-efficiency (version 1.0.0). It converts HTML/CSS slides into editable PowerPoint elements rather than screenshots, ensuring that text, layouts, and visual components remain fully modifiable within the .pptx file. The tool automatically handles font embedding, CJK fonts, complex CSS, layout measurements, structural validation, and final visual audits, falls back to screenshots for highly complex decorations, and operates without relying on frameworks such as reveal.js or custom-built decks. It is ideal for quickly converting web presentations, product pages, and data visualizations into deliverable .pptx files, and requires no API key.
Read MoreAnti-Templated Front-end Design Taste: AI Agents Generate Non-clichéd Interfaces
Anti-template Frontend Design Aesthetics is an AI Agent Skill published by community developer user_3c6cb52e, version 1.0.0, categorized under ai-agent, sourced from the community, and requires no API Key. It is positioned as a design aesthetics training guide that helps AI Agents follow professional design standards and best practices when generating frontend interfaces, reducing templated outputs such as purple-gradient titles and homogenized card layouts. This article introduces its core capabilities (design specification guidance based on the SKILL.md format), official URL and installation/activation methods (no installation commands are provided in public materials; please refer to the official page for details), general usage paths, as well as applicable scenarios and limitations: the material only contains introductory-level information, and the original SKILL.md should be evaluated before adoption; it has 883 community downloads and 0 installations, so it is recommended to pilot within a limited scope.
Read Moredsh-feishu-session-notify: Create a Bidirectional Bridge Between DSH Sessions and Feishu
dsh-feishu-session-notify (Re-qi, a workflow-type Host plugin) pushes DSH completion outputs, failure reasons, approvals, and ask_user_question to Feishu (Lark) in Markdown format via lark-cli, with topic mixing enabled by default. Feishu replies and /dsh commands are routed back to the corresponding conversation or trigger new workspace tasks. Notifications do not invoke the model. Installation requires linking a local clone in the profile and restarting DSH, and depends on lark-cli and im.message.receive_v1.
Read MoreChatting and Match Analysis: Conducting deep research on European odds for Sports Lottery and Beijing Single Match betting markets.
Welfare Plumbing Watch European Core Football Analysis is a football match research Skill aimed at official Chinese sports lottery competitive matches and Beijing single-match betting. It is published by user_3652bc1f from a community source, with version 1.0.3, classified under data-analysis. It requires no API Key and is settled via Alipay. The Skill encapsulates data retrieval, five-dimensional European odds analysis (fundamentals, distribution assessment, European odds structure, Asian odds, and Betfair trading data), and report formatting into a fixed workflow: for a single match, it outputs a seven-step research report including evidence and counter-evidence, primary and secondary leanings, confidence level, data timing, and limitations; for batch scenarios, it generates a single most credible plan, with a price of 1 yuan per single match or a total of 1 yuan for the entire batch report, without cumulative charging based on the number of matches in the batch. The program is responsible for providing evidence, while the final assessment is completed by the WorkBuddy host model, making it closer to a structured research input rather than a bet-following signal. The article introduces the positioning, core features, installation and activation methods, official trigger words and supporting functions, as well as applicable scenarios and precautions.
Read MoreDSH-MLIR-Toolchain: Invoking the MLIR Toolchain with argv in DSH
dsh-mlir-toolchain is a DSH plugin maintained by Rcyc-1025 that reads toolchain.json and provides four tools—mlir_probe, mlir_run, mlir_tblgen, and mlir_parse_check—via subprocess. It operates without shell intervention or PATH lookups and has zero runtime dependencies. It is suitable for scenarios involving local LLVM/MLIR installations where IR transformation, TableGen, and parsing verification are performed within the Harness. To install, use `dsh plugin --profile desktop add ./dsh-mlir-toolchain`. Requirements include Node ^22.19.0 or >=24, and DSH >=0.2.1-rc.5 <0.3.0-0.
Read Moredsh-build-driver: Encapsulating the CMake + Ninja + MSVC Build Process for Agents
dsh-build-driver is a DSH client plugin maintained by Rcyc-1025. It provides four tools: build_configure, build_run, build_status, and build_stop. It features automatic MSVC environment injection, parsing of compilation diagnostics, and support for transitioning long builds to the background. This article describes the tool semantics, buildDir rules, optional cordis.patch.yml configuration, Chinese GBK decoding strategy, dependency versions, and the official README installation method. It also outlines applicable scenarios (Windows CMake + Ninja + MSVC) and notes the precaution of reviewing the source code before installation.
Read More