dsh-plugin-legion: One Goal, One Advancing Harness Team
The `dsh-plugin-legion` maintained by seanliii is a DSH workflow plugin (version 0.6.0, Technical preview). It allows users to select companions and share tools within a native session centered on a single goal, continuously revising the same deliverable. It can utilize Discover capabilities or connect to a self-managed Legion core to delegate members. To install, you must replace the placeholder with a Release-verified tag or commit using the following command: `dsh plugin --profile web -w add 'github:seanliii/dsh-plugin-legion#<verified-tag-or-full-commit>'` Requirements include Node >=24.9.0 and dsh 0.2.1-rc.2. The plugin operates with current dsh permissions; please review the source code and license before installation.
Read Moreemail-triage: A Skill for Generating HTML Dashboards to Classify Daily Emails into T0-T3 Levels
email-triage is a community-sourced office-efficiency Skill, published by user_8c9b6603, version 1.0.0. Every morning, it automatically fetches unread emails from all folders in the user's mailbox, passes them to an LLM for analysis, and classifies them into four priority levels (T0-T3). It generates an HTML dashboard saved to the "Email Daily Report" folder on the desktop, and pushes a briefing via WeCom/WeChat. The article introduces its positioning, verified core capabilities, location, and official address. It notes that the requires_api_key=true tag indicates that an LLM API key and a messaging push channel must be configured before running. It also clarifies that the official documentation does not provide the criteria for each of the T0-T3 levels, installation commands, or usage examples; therefore, users need to calibrate grading expectations themselves and conduct a small-scale trial run beforehand.
Read Moredsh-pi-ai-patch: Check the text/image modality declaration of the model in the settings.
dsh-pi-ai-patch is a plugin for the DSH client, maintained by sdwf4wgv5z-rgb. In Settings > Image recognition configuration, it lists llm-pi-ai models grouped by provider and uses checkboxes to add or remove the input: [text, image] declaration in cordis.patch.yml, avoiding manual YAML indentation errors. It only manages modality declarations on the DSH side and does not modify API capabilities. It supports YAML validation, line-level safe editing, and backups before saving. To install, link the local repository in the profile’s package.json and register the bundle; changes take effect after restarting DSH. The HTTP server listens only on 127.0.0.1:19877. Released under the MIT License; requires Node.js >= 20.
Read MoreData: A Data Processing Skill Covering the Full Workflow from Extraction and Cleaning to Analysis and Reporting
Data is an Agent Skill designed for data analysis tasks. Its coordinates are @user_053d27d0/user-053d27d0-data, version 1.0.1, maintained by community developer user_053d27d0, classified under data-analysis, and requires no API key. It breaks a data processing task into five stages—extraction, cleaning, analysis, visualization, and reporting—with the goal of enabling the Agent to complete the end-to-end workflow from data retrieval to delivery within a unified process. The article outlines its capabilities (it does not bundle function libraries; it relies on the host environment's Python, SQL, and plotting tools), the general installation method (download it into the skill directory and declare it in the host configuration), typical usage (specify the data source, cleaning rules, analysis objectives, and expected outputs in the prompt), as well as applicable scenarios and caveats: the data scale is suitable for single-machine use; cleaning rules need to be specified in detail in the prompt; SkillHub ratings and download counts vary by the time of retrieval and should be based on the current page.
Read MoreSecond Brain: Equip an Agent with a Cross-Session Memory and Governance Protocol
Second Brain is a portable, open-source second brain framework and Agent governance toolset (currently version 1.0.11). Released under the MIT license, it is local-first and requires no API keys. It upgrades session-based assistants into a continuously growing "Brain OS" that persists across sessions. The framework incorporates a universal five-layer management protocol template, honesty boundaries, coordination discipline, instantiation guards, MEA governance, and machine-readable GIR/CRL schemas. It also features a sanitization tool that enables the secure open-sourcing of a private brain after data masking. The framework supports self-learning, self-iteration, and automatic wake-up. After a secondary skill is invoked for the first time, the system automatically routes to it in subsequent sessions based on conversation content, making it friendly to beginners, ultra-long conversations, and large-scale projects. Memory is stored locally; when switching computers, continuity is maintained simply by copying the folder or allowing the Agent to generate a transplant package.
Read Moredsh-interject: Deliver queued messages at the next boundary without interrupting the current model call
dsh-interject is a DSH host-side workflow plugin, maintained by sdlhs. When the Agent is busy, "queue and send" messages default to entering next-turn and are only claimed after the entire turn completes. This plugin promotes such messages to next-step during policy matching, allowing them to enter the conversation history at the next step boundary without aborting the current llm.stream request. By default, onQueue is set to steer and onSteer to pass, without sending additional requests. Configuration in ~/.dsh/dsh-interject.json supports hot reload. It can be installed by searching for dsh-interject in the DSH plugin marketplace, or mounted via cordis.patch.yml.
Read MoreHigh Net Worth Planning: An Introductory Guide to Wealth Organization for Individuals and Families
High-Net-Worth Wealth Planning is a professional “Skill” published in a SkillHub community (version 1.0.3), maintained by user_daa9c75a. It is positioned as an entry-level wealth review for individuals and families, producing a "Family Wealth Foundation Review Report" that includes five parts: a family balance sheet; cash flow and liquidity; a risk flags checklist; starting points for succession planning; and an illustrative asset allocation framework. Trigger scenarios include reviewing household assets, calculating net worth, succession planning considerations, getting started with asset allocation, and calculating liquidity gaps. It explicitly does not apply to single-product recommendations or buy/sell advice on individual stocks, and does not constitute investment, tax, legal, trust, insurance, or immigration advice. It does not provide tax-evasion, nominee-holding, or asset-concealment solutions, does not include stock ticker codes, and does not promise returns; for in-depth solutions, use the professional-version skill: high-net-worth-wealth-planning. Declared as requires_api_key=false; the SkillHub page shows a rating of 100000, 791 downloads, and 0 installations.
Read More@dsh-cbw/bilibili: Monitor Bilibili, read subtitles, write daily reports, and @ you in DSH
@dsh-cbw/bilibili is a DSH web plugin maintained by Scorpioflz. It is a fork of dsh-bili-scroll v2.3.1 and bundles bilibili-mcp. Its duty-shift routines leverage peak/off-peak windows to search Bilibili, read subtitles, filter videos, and generate daily reports, and it supports private-message monitoring and 13 host tools. This fork adds a persistent session (shared by duty and DMs), masterUid and hobby profiling, half-life learning of interest keywords, storyboard frames, three-tier safety gates for comments, and viewing/inbox statistics. Install with `dsh plugin --profile web add @dsh-cbw/bilibili`; after restarting, go to Settings → “刷B站” to install dependencies and run scan-login. Pairing with @dsh-cbw/memory is recommended.
Read MoreAutomatic Skill: Turn "Find Tools Before Doing Work" into a Mandatory Pre-Check
Autoskill (@user_70b62916/autoskill-pro) is a Skill of the ai-agent category on the SkillHub store, published by user_70b62916, version 1.0.0, sourced from the community. It turns the practice of "after receiving any user request, first searching the SkillHub store for a matching skill" into a mandatory pre-execution check before starting work, covering all tasks such as programming, file processing, data analysis, web page generation, and search research, thereby implementing a zero-exception working mode of "find the tool first, then do the work." The store page lists requires_api_key=false, with a rating of 100000, 904 downloads, and 0 installations. This article introduces its positioning, core functions, official address, typical workflow, and applicable considerations.
Read Moredsh-notes-board: Integrating Engineering Decision Notes from All Workspaces into the Native DSH Board
dsh-notes-board is a read-only DSH plugin maintained by scccy that injects engineering decision Markdown files from `.agents/notes` across workspaces into the left panel of DeepSeek Harness. Its interface is based on the original `write-notes-like-deepseek` board, featuring four sections: Architecture Baseline, Evolution Timeline, Pitfall Repository, and Decision List. The drawer displays the main content across five sections. Dependencies are installed via `install_bundle` or profile links. The Host exposes only two GET endpoints—`workspaces` and `scan`—without parsing data or writing files.
Read MoreSelf-Driving Route Planner Pro: Multi-Day Self-Driving Route Planning Skill for Mainland China
Self-Driving Route Planning Pro is the official Skill from AI Self-Driving Meow, utilized for destination recommendations, interactive route planning, validation of real POIs and driving conditions, route modifications, and the final preservation of itineraries, with a planning range of up to 10 days. This article introduces its positioning, capability boundaries, installation and enablement precautions, and applicable scenarios.
Read MoreMultilingual Translation of Industry Drawings: Letting Agents Handle Mechanical Drawings
Industry Drawing Multilingual Translation is a paid Skill released by Ruan Ce Software on SkillHub, version 0.1.5, classified as Professional and located at @ruancyai/industry-drawing-translator. It targets text translation and multilingual delivery workflows for engineering drawings in industries such as mechanical, electrical, and chemical engineering. Based on the standard SKILL.md format, the Agent executes it according to the skill definition. Version 0.1.5 adds a preflight check for the WeChat Pay plugin: if the payment tool is missing, submission of paid tasks stops and prompts for configuration; translation features, delivery flow, and pricing remain unchanged. To use it, enable it in an Agent environment that supports SKILL.md, and prepare the WeChat Pay plugin along with a matching API Key. The Skill has a rating of 100000, 860 downloads, and 0 installs.
Read MoreJLPT Practice Drills: Questions and Explanations for Exam Preparation
Japanese JLPT Quiz is a community Skill designed for JLPT exam preparation scenarios, located at @user_af163eaa/jlpt-quiz-m01. It is published and maintained by user_af163eaa, currently at version 1.0.1. The source is categorized as "community" under the "education" classification. It is intended for use when users are preparing for the JLPT and require Japanese practice questions with explanations. Its core capability is providing Japanese questions accompanied by analyses. The tag indicates `requires_api_key=false`, meaning no API Key is needed for integration. Official documentation does not provide specific installation commands or differentiate between Cursor, Codex CLI, and Claude Code; therefore, it must be mounted according to the Skill loading method of the specific client being used. The article outlines the Skill's positioning, verified features, integration information, usage based on official descriptions, applicable scenarios, and precautions: notably, that the source is community-based, the coverage of levels and question types is not explicitly listed, and quality and scope must be verified through actual testing first.
Read Moredsh-rvc-cover: Local RVC Covers and Speech with the Same Model
dsh-rvc-cover is SB-TDM's DSH plugin for Windows, which registers the rvc-cover skill. It does not bundle its own Applio/RVC engine or .pth files; instead, the agent invokes PowerShell scripts in sequence — a health check, a 10-second preview, voice change, Edge TTS speech, and mixing. Installation uses `dsh plugin --profile web add <local-path>` or `github:SB-TDM/dsh-rvc-cover`, and requires the `DSH_RVC_HOME` and `DSH_RVC_MODEL` environment variables. The repository's README notes that the full-song workflow has not yet been validated on the target machine, and RVC 2.3 is WebUI-only. MIT licensed.
Read MoreEnglish Essay Audio Reading (MiMo TTS): Turning English Essays into Playable Audio
This community-sourced Skill (ID @user_c43ef728/english-essay-tts, version 1.1.0, category education) converts English essays into natural speech and returns the audio directly within the conversation. It supports pasting plain text or uploading photos of essays; in the case of photos, OCR is performed first. It then analyzes the sentiment of the essay to recommend matching English voices and speaking styles. Upon first use, it requests the user's own MiMo API key; finally, it calls Xiaomi MiMo-V2.5-TTS to synthesize the voice and returns playable audio directly to the chat. The metadata includes the `requires_api_key=true` tag, indicating that a MiMo API key is mandatory. Officially listed trigger phrases include "朗读这篇英语作文", "把英语作文读出来", "英语作文配音", "英语听力素材", as well as "generate audio for my English essay", "TTS for this English composition", "read my essay aloud", and "MiMo TTS". Usage notes: An API key is required; photo processing relies on OCR, and recognition errors will be directly read into the audio; the specific list of voice candidates is not listed in the metadata; this Skill is designed for English content. The SkillHub page displays a rating of 100000 and 860 downloads, sourced from the community.
Read Moredsh-session-delete: Permanently delete session in DSH Web Sidebar
The DSH community plugin dsh-session-delete (0.1.0, MIT), maintained by Sartherland, provides a "Permanently Delete..." option in the Web sidebar session menu. Upon confirmation, the Host performs a phased deletion of session logs, subagent sub-sessions, projection caches, workspace attributions, scheduled tasks, and spill files, with an optional trigger for full-text index reconciliation; forks and shared attachments are retained. It requires DSH >= 0.2.1-rc.3. Installation: Run `dsh plugin add github:Sartherland/dsh-session-delete` and enable the bundle on the plugin page. Currently open sessions cannot be deleted. As this is a destructive operation, installation should only occur after understanding the scope and reviewing the source code.
Read MoreSega Pixel Game Style Design | Jianshi AI: A Retro Arcade Style Interface Specification Skill
# Introduction When creating retro game-themed interfaces, the common practice is for designers to临时 source pixel fonts, determine corner radii and button styles based on feeling, and for developers to replicate the design based on screenshots. The result is often misaligned fonts, inconsistent corner radii, and buttons that do not look sufficiently "arcade-like." This Skill solidifies the visual specifications of the Sega brand arcade pixel style, allowing models to directly apply a unified style when generating interface designs. # What is this **Sega Pixel Game Style Design | Jianshi AI** is a Skill designed for interface design, maintained by community publisher user_fccf428b. Currently at version 1.0.2, it is categorized under design-media and originates from the community. It addresses the issue of retro game-style interfaces lacking unified standards and the need to piece together visual elements from scratch every time.
Read MoreThree-Stage AI Trace Elimination: Detection and Rewriting of AI Signals in Chinese Text
Three-Tier Humanizer (@cyinfo/humanize-cn-text, released by Chuyun Technology, categorized as content-creation, version 1.0.0) targets a common polishing pain point for Chinese AI-generated text: it can detect 35 AI writing patterns with one click, automatically select three levels of rewriting strength based on the target scenario (from business reports to social media posts), and self-score the rewritten result—re-editing automatically if the score falls short. The documentation confirms its tag is requires_api_key=false, meaning no additional API key is needed. However, it does not disclose an installation command, configuration fields, or reproducible invocation examples; therefore, no command has been constructed in this article, and it is recommended to verify against the official source before integration. It is suitable for Chinese content creation that requires a natural tone, and does not involve code generation tasks.
Read Moredsh-track: Making Task State Survive DSH Context Compression
**dsh-track** is a DSH memory plugin maintained by Saretheya. The model writes checkpoints via `checkpoint_write`; the plugin stores the data in its directory and re-injects it into the system prompt after every context compaction to prevent established conclusions from being lost during lossy summarization. The article covers four levels of injection strategies, context occupancy reminders, the installation command (`dsh plugin --profile desktop add dsh-track`), as well as usage and limitations regarding `topic/keywords` and `AGENTS.md` discipline.
Read MoreWeChat Channels Viral Content Pre-check Assistant: Multi-dimensional Pre-screening of Scripts Before Publishing
The WeChat Channels Viral Pre-check Assistant is a content creation skill hosted on the community SkillHub, maintained by user_da878c5a with capabilities powered by AIJJ. It addresses the pain point of lacking systematic checks before publishing works on WeChat Channels: after pasting the voiceover script, the AI collects key information through six rounds of Q&A, and then generates a pre-check report across 10 dimensions. It also provides targeted recommendations based on the platform's unique characteristics (WeChat ecosystem, social recommendations, and private-domain integration). Unlike post-publication data review, it shifts the evaluation process to the pre-publication phase. It requires an API key (requires_api_key=true). Since the official installation commands and configuration field names are not provided in the documentation, these details should be verified prior to use.
Read MoreVegas Tunnel Trading: A Universal Agent Skill for A-Share Decision Support
This article introduces the community Skill "Vegas Tunnel Trading," designed as an intelligent decision-making aid for A-share scenarios. It utilizes Exponential Moving Averages (EMA) to determine trends, uses Fibonacci levels to identify pullback zones, and employs a resonance score to summarize signals. A single call yields three types of results, with the official claim stating a threefold increase in scenario efficiency. Released by u_8ee34d77 on SkillHub, this Skill is at version 1.0.0, originating from the "community" source with a "professional" classification. It is located at @u_8ee34d77/vegas-tunnel-free, follows the universal SKILL.md format, and carries the tag `requires_api_key=true`. The article lists the combination of capabilities, the API key preparation required before enabling, and the activation steps organized according to general procedures. It also explains that there are differences in integration paths for Cursor, Codex CLI, and Claude Code. Since publicly available information does not provide verifiable installation commands or configuration fields, no itemized listing is provided. Finally, it provides the applicable scenarios, potential pitfalls (metadata volatility, API Key dependency, and the "3x efficiency" claim lacking verified data), and the official address.
Read Moredsh-flashback: Using Checkpoints to Let Task State Survive Context Compression
dsh-flashback is a memory plugin developed by Saretheya for DSH 0.2.x: models write checkpoints at milestones via checkpoint_write; the plugin injects a copy into the system prompt and re-injects it after every compaction, with data stored only in the plugin directory. This article introduces four injection strategies, the checkpoint_read/context_status/context_compact tools, context usage reminders across different context windows, and the install command dsh plugin --profile desktop add dsh-flashback; it also covers cross-session topic/keywords, known limitations, and security review points.
Read MoreInterview Assistant Skill: A General Assistant for Resume Optimization and Mock Interview
The Interview Assistant is a community-sourced Skill based on the universal SKILL.md format. Its coordinate is @user_db7424f5/udb7-interview-assistant. The current version is 1.0.0, categorized as business-ops, and tagged with requires_api_key=false. It provides five key capabilities: resume optimization suggestions, job description (JD) matching analysis, HR perspective review, interviewer perspective review, and mock interview question generation. It covers three core user groups: job seekers, HR professionals, and interviewers. The SkillHub page indicates a rating of 100000, 851 downloads, and 0 installations. It is sourced from the community, so users should verify its quality independently. This article introduces its positioning, functions, installation, and usage. Since official documentation lacks installation commands and examples, these sections are honestly marked as missing, without any fabricated information.
Read Moredsh-ctx: Leveraging Checkpoints to Preserve Task State During DSH Context Compression
dsh-ctx is a DSH memory-related plugin maintained by Saretheya: the model writes milestone checkpoints using checkpoint_write; the plugin stores the content in its own directory and re-injects it into the first paragraph of the system prompt after each context compression. It supports four injection strategies, context usage tier alerts, wake-up after compression, and cross-session summary cards. The installation command is `dsh plugin --profile desktop add dsh-ctx`; it requires DSH 0.2.x. Based on the official README, this article introduces its capabilities, usage, and limitations, and does not cover writing workspace files.
Read MoreCompetitor Monitoring: Automated Tracking of Public Activities and Briefing Generation
This article introduces a community-published, general-purpose Skill called “Competitor Watch” (@user_0129d78e/competitor-watch-v2, version 1.0.0, category: data-analysis). It can automatically monitor a specified competitor’s public activity, distinguish between no changes and key changes, and generate a brief when there are changes. It can be triggered with natural language commands such as “check the competitor’s latest moves,” “competitor daily report,” or “what new actions has the competitor taken.” The metadata indicates requires_api_key=false. The article outlines the capability boundaries, official URL and page, and honestly notes that among the retrieved materials there is no installation command or configuration fields, so it does not provide speculative installation steps. It also points out that although the Skill has a rating of 100,000, the install count is 0 and it is maintained by community sources, its usage scope is narrow, and it is recommended to run a small-scale trial first.
Read More