DeepSeek Harness Web UI Status Desktop Pet Plugin for dsh-deepseek-girl-pet
dsh-deepseek-girl-pet is an open-source Cordis desktop pet plugin based on DSH's "everything is a plugin" philosophy, specifically designed for the DeepSeek Harness Web UI. The plugin intuitively displays the current Session status through an animated character in the lower-right corner of the interface: it shows a standby animation holding a board when idle, switches to a working animation when executing tasks, and switches to a waiting animation when awaiting interaction. Additionally, it supports 16-direction mouse tracking, automatically returns to standby after the mouse stops for 1.1 seconds, allows size switching on click, and is compatible with prefers-reduced-motion. As an unofficial community project (MIT License), the plugin does not modify core source code, does not read or write conversation data, and does not rely on external CDNs, ensuring lightness and safety. Installation methods include installing via dsh.
Read Moredsh-outdoor-theme: A Wilderness Guide Skin and Trail Sign Panel for the DSH Web Interface
dsh-outdoor-theme is a DSH plugin maintained by Estellalee, positioned as an outdoor hiking-style reskin solution. Without altering the DSH interface logic, it achieves visual upgrades through pure CSS and JavaScript, primarily offering three features: reskinning the web interface into a mountain-wilderness style, providing a settings panel entry for switching skins, and adding a right-side "Trail Sign · Work Progress" panel. The plugin comes with two built-in color schemes—the light "Morning Meadow" and the dark "Dusk Forest"—and supports one-click switching back to the default style. Before installation, ensure that pnpm is available, then install via the `dsh plugin --profile web add` command from GitHub or a local directory. After installation, refresh the page for the changes to take effect, and users can switch in the settings
Read Moredsh-whale-pet: Whale Maiden Desktop Pet Plugin for DeepSeek Harness Web UI
dsh-whale-pet is a lightweight desktop pet plugin for the DeepSeek Harness Web UI, developed by Er1c0v0 and released under the MIT License. It injects the UI through the official shell.overlay slot without modifying core code or scraping text, providing users with non-intrusive status notifications. The plugin mirrors the five Harness states—idle, working, needs input, complete, and blocked—and enters sleep mode after 40 seconds of inactivity. It supports dragging, keyboard movement, collapse and restore, as well as bilingual display in Chinese and English. Its distinctive feature is reading privacy-safe activity summaries from metadata and displaying them in a bubble, avoiding occlusion of the view. Installation requires Node.js 22.19+ or 24+ and
Read Moredsh-moyan: Displaying a Quote in the DeepSeek Harness WebUI Sidebar
dsh-moyan is a lightweight plugin for the DeepSeek Harness WebUI, released under the MIT license, designed to display famous quotes, classical poetry, or iconic lines with minimal interference in the sidebar, positioned above the "Settings" button. The plugin does not alter the core workflow and primarily provides a maintainable quote slot with refresh controls. Core features include: automatic line wrapping to display sentences and their sources, support for refresh frequencies of "every open" or "daily" (with automatic quote changes across midnight); manual quote switching and the ability to roll back up to the last 5 entries, while the 10 most recently displayed quotes are not repeated in random selection. The corpus is stored externally in `corpus.txt`, supporting real-time editing that takes effect without requiring a service restart. The interface automatically adapts to light/dark modes by following the Harness theme variables, and it automatically hides when the sidebar is collapsed. Installation requires Node.js ≥ 18 and pnpm.
Read MoreEligahyu/dsh-sentinel-scanner: Read-Only Security Scanning for DSH Plugins
Eligahyu/dsh-sentinel-scanner is a read-only static security scanning tool for DeepSeek Harness (DSH) plugins, designed to perform supply chain and security risk assessment before code execution. The tool detects risks such as command execution, credential access, data exfiltration, and obfuscation through 51 AST-based heuristic rules, outputting a risk score from 0 to 100 along with a security verdict, but never executes the scanned code. It offers multiple integration methods: as a DSH Agent Tool plugin, a standalone CLI (`dsh-sentinel`), a GitHub Action, or a JavaScript API. Core features include three scanning modes—source code, package artifacts, and profiles—with support for pre-publishing npm packages.
Read MoreDSH-AgentMemory: A Local Memory Plugin for Integrating AgentMemory into DSH
dsh-agentmemory is a DSH memory plugin (MIT licensed) maintained by elementor-i, designed to integrate a locally running agentmemory REST service into the DSH framework. This plugin does not require an MCP bridge and directly depends on a local service (default localhost:3111), providing agents with a stable memory backend for reusing session records and injecting recalled context. Core capabilities include providing 54 `memory_*` tools (including 8 core tools) and a `memory_http` escape hatch; automatically capturing sessions, prompts, tool calls, and other activities in the background; and supporting optional context injection. The default configuration emphasizes security, including credential redaction, timeout control, and non-blocking observation, with destructive tools disabled by default. Installation can be performed via
Read MoreDSH-LLM-Vision-Bridge: Integrating Image Understanding into DeepSeek Text Models in the DSH Web GUI
dsh-llm-vision-bridge is a DSH plugin licensed under MIT, designed to address the issue of text models being unable to directly process images. This plugin registers the `deepseek-vision` provider in DSH's `LlmAdapter`. When a request includes images, it automatically routes them to a configured vision model (such as pi-ai or a local llama.cpp) to generate text descriptions, which are then used by the DeepSeek text model to continue the conversation based on the descriptions. If no images are present, the request is directly passed through to a fallback provider for zero-overhead operation. Key features include an LRU cache to avoid duplicate descriptions, automatic retries for 503/429 errors and timeouts, and configurable failure policies (placeholder or error). After installation, the web service must be restarted, and configuration is required in the settings.
Read Moredsh-mcp-bridge: Providing a Set of Validated MCP Servers for DeepSeek Harness
dsh-mcp-bridge is an MCP server plugin package designed for DeepSeek Harness (DSH), aiming to streamline the process of integrating agents with MCP tools. This plugin bundles multiple servers, including demo, memory, filesystem, GitHub, Playwright, and remote HTTP, and leverages built-in client support for stdio and streamable-http transports, automatic reconnection, and hot updates. Developers can ensure server status through connectivity verification scripts before delegating calls to models. Installation only requires configuring dsh and pnpm in the PATH, executing the `dsh plugin` command to add the plugin, and restarting. The plugin provides a default demo server that requires no API key, while other servers
Read Moredsh-go-rotator: OpenCode Go Subscription Key Rotation for DeepSeek Harness
dsh-go-rotator is an open-source plugin designed for DeepSeek Harness (dsh), aimed at addressing the management challenges of multiple OpenCode Go subscription keys. It provides load balancing, automatic rotation, and quota management features, resolving active keys at request time via `apiKeyEnv` to distribute load. When encountering rate limiting, quota exhaustion, or authentication errors, the plugin automatically cools down the current key and switches to other available keys to ensure service continuity. Additionally, the plugin supports cost estimation based on tokens, records rolling window and long-term ledger data, and allows monitoring of status through a web interface or local API. Its built-in quota watchdog can fetch real-time data from the Dashboard and automatically disable keys when thresholds are breached. The plugin stores self-contained data in a local `data/` directory, ensuring key security without writing to logs.
Read MoreDeepSeekEyes: Auditable Vision and Computer Use Plugin for DeepSeek Harness
DeepSeekEyes is a DeepSeek Harness (DSH) plugin developed by dttxorg under the MIT license, designed to address the issues that image pixels cannot be directly used as model input in agent tasks and the lack of evidence traceability. This plugin constructs an auditable visual and cross-platform Computer Use runtime, with core features including: supporting multimodal models to read raw pixels to provide unscaled, traceable visual evidence; offering native desktop control and browser automation operations on Windows/macOS; and integrating an MCP application layer to restrict, hash, and audit tool invocation results. Additionally, it provides health checks, failover, strict evidence validation, and token usage visualization capabilities. By default, automation and MCP features are disabled and require manual configuration to enable.
Read Moredsh-lineage: Lineage Evidence Plugin for DeepSeek Harness
dsh-lineage is an MIT-licensed plugin (v0.2.0, requiring Node.js 22+) developed by dongsheng123132, designed to build a locally verifiable object graph for artifacts, verified-fact records, actions, and reports in DSH workflows. It resolves references within the workspace and computes SHA-256 hashes to validate node states (verified/missing/stale), detect dangling references, cycles, and invalid relationships, ensuring upstream and downstream evidence integrity without silently promoting missing objects to facts. Core features include object graph validation and an append-only ledger, supporting idempotent replay and strict path boundaries, while rejecting
Read MoreDSH-Capability-Receipt: DeepSeek Harness Skill Loading Receipt
`dsh-capability-receipt` is an MIT-licensed DSH plugin (v0.3.0, requires Node.js 22+), designed to verify whether the skills actually loaded by the DeepSeek Harness (DSH) runtime are consistent with trusted sources by generating content-addressed receipts. The plugin only observes target capabilities without executing their logic; its core mechanism hashes the valid instruction body and compares it against trusted fixed hashes or evidence inputs from the pack-agent lock. The plugin provides three DSH tools and two MCP proof tools, supporting receipt generation from standard sources or the pack-agent lock. It is recommended to install it in an isolated DSH profile for specific compatibility.
Read MoreDSH-CAD-Review: ASCII DXF Inspection and Deterministic Rule Review for DeepSeek Harness
dsh-cad-review is an MIT-licensed plugin within the DeepSeek Harness ecosystem, designed specifically for entity-level deterministic review of ASCII DXF files. It abandons screenshot-based empirical inference and directly reads CAD entities, hashing source files and linking issues to entity handles, layers, source code lines, and geometric positions to ensure results are reproducible and evidence-backed. The plugin supports common geometry types such as LINE and CIRCLE, offering two core functions: `inspect` (evidence-only checking) and `review` (policy-based rule execution), with rules covering invalid values, zero-length lines, layer restrictions, and duplicate entities. It can be invoked via CLI, DSH tools, or the MCP interface, and supports configuring project-specific policies to override default rules. If the review identifies error-level issues, the system returns a specific exit
Read Moredsh-agent-plugin-market: Managing DSH Agent Skills and Hooks with Git Repositories
`dsh-agent-plugin-market` is a DSH plugin (MIT licensed) maintained by Diluka, designed to centrally manage Codex/Claude skills, plugins, and optional Hooks loaded from Git repositories. The plugin treats Git repositories as content markets, automatically discovering available resources by cloning the repository and identifying standard manifest files (such as `marketplace.json` and `plugin.json`). Core features include: supporting global default configuration with workspace-level sparse overrides, allowing independent enabling or disabling of specific plugins and standalone skills across different workspaces; enabling in-place plugin loading without copying files; providing agent tools such as `agent_market_info` for AI to query status or write workspace overrides, while restricting its execution scope to the global level.
Read MoreDSH-Office: Native Office Document Plugin for DeepSeek Harness
DSH-Office is a DSH plugin maintained by didclawapp-ai, designed to address issues in agent-based generation and editing of Office files, including unclear field structures, complex operations, and difficulties in result verification. The plugin, based on the local zagens-office engine, supports read and write operations on PPTX, DOCX, XLSX, and PDF files and is licensed under the MIT License. Its core workflow comprises four tools: first, `office_schema` retrieves a JSON contract to clarify the field structure; then, based on requirements, `office_write` is invoked to create new documents or `office_edit` to modify existing ones according to operation instructions; finally, `office_read` reads back the document for verification. This process ensures
Read MorePBR-Render: A PBR 3D Model Preview Plugin for DeepSeek Harness
pbr-render is a PBR 3D model preview plugin for DeepSeek Harness (DSH), designed to address the lack of intuitive visual feedback when agents handle game art assets. The plugin supports rendering GLB/GLTF models as interactive 3D previews, embedded in responses via the `pbr3d` fence. Key features include full PBR rendering, multiple material channel inspection modes (e.g., normal, roughness, metalness, etc.), five built-in environment lighting presets for switching, and interactive operations such as drag-to-rotate and zoom. The plugin also incorporates an active trigger mechanism and safety boundary protection, exposing only files with specific extensions within the workspace and limiting size to 512 MiB. Installation is performed via GitHub commands or local paths, requiring a restart of `dsh web` to take effect. The plugin is applicable for use with D
Read Moredsh-trajectory-governance: DSH Trajectory Governance and Anomaly Diagnosis Plugin
dsh-trajectory-governance is an observability layer plugin for DeepSeek Harness (DSH), designed to address the challenges of flat logs in long-running tasks, including difficulty in tracking branches, goal drift, and retry issues. The plugin operates with zero kernel modifications, reconstructing flat logs into a structured trajectory tree by subscribing to events, and saves observability layer snapshots containing node IDs and context hashes. Its core functionality includes three temporal anomaly diagnostic strategies: run-loop deadlocks, ineffective retries, and goal drift, with results mounted on tree nodes and displayed in a standalone GUI. The plugin supports desktop notifications, webhook alerts, and one-click interruption, and allows for creating new session branches at stable boundaries. Data is independently stored in a local SQLite database, with CLI tools provided for querying sessions, trajectory trees, and snapshots. Installation requires DSH version 0.1.0-rc.6 or later, available via B[uild?]
Read MoreDeepSeek Harness Plugin: dsh-llm-wiki
dsh-llm-wiki is a zero-runtime-dependency plugin based on DeepSeek Harness (DSH), built with a pure ESM architecture and requiring Node.js 18 or above. By invoking the LLM-Wiki engine CLI, it encapsulates seven core tools for agents: retrieval, page reading, statistics, validation, fixing, error log viewing, and data ingestion. This plugin aims to implement the philosophy of "everything is a plugin," supporting direct knowledge base queries within conversations, ingesting source texts into the knowledge base, and continuously maintaining the knowledge base structure. Among these, query-type tools require no additional LLM configuration, while data ingestion and the fix functionality with Finalize require configuring an OpenAI-compatible LLM API. The installation process includes installing the Python engine via pip, adding the plugin via the dsh command, and configuring it in the pr
Read Moredsh-openapi: Adding OpenAPI Discovery and Secure Invocation Tools to DeepSeek Harness
dsh-openapi is a plugin for DeepSeek Harness, designed to address issues such as inconsistent parameters, uncontrolled credentials, and excessive scope when models directly call APIs. This plugin indexes OpenAPI 3.x documents and provides three core tools: `openapi_list` (discovery and search operations), `openapi_describe` (viewing details), and `openapi_call` (restricted invocation). Its core features include strict security boundaries: by default, only GET/HEAD requests are allowed, access to local and private network IPs is blocked, response body sizes are limited, and sensitive response headers are shielded, while redirect checks are supported. Credential management is implemented through environment variable mapping, ensuring that keys are not directly exposed in configurations or outputs. Installation requires the use of the `dsh plugin` command.
Read Moredsh-web-open: Print URL after `dsh web` is ready and open the browser
`@dawsondx/dsh-web-open` is an automated tool designed for DSH plugins, aimed at resolving the tedious process of manually organizing addresses and opening a browser after `dsh web` starts. This plugin automatically prints the full GUI address and opens it in the default browser once the web service is ready, with cross-platform support for Windows, macOS, and Linux. Its core features include zero runtime dependencies, a fail-safe mechanism (only warns without errors when service retrieval fails), and support for system-assigned ports. Installation requires Node.js >=18 and specific peer dependencies, and can be added via npm or GitHub. Users can flexibly control whether to enable auto-opening through the environment variable `DSH_OPEN_BROWSER=0` or configuration files.
Read Moredsh-file-drop: Writing Non-Image File Paths into the DSH Input Field
`dsh-file-drop` is a persistent plugin for DeepSeek Harness (DSH), designed to address the inefficiency of non-image files (such as PDF, Word, Excel, etc.) in DSH, which currently can only be handled through manual path copying. The plugin offers three file input methods: dragging and dropping files anywhere in the window, clicking the paperclip button for multi-selection, and hover bubble hints. It supports directly writing file paths into the input box draft for the agent to read. Its path acquisition strategy attempts in the following priority order: first, directly obtaining the original Finder path through the desktop shell (zero upload, no size limit); second, using the `text/uri-list` path from the drag-and-drop operation; finally, uploading files to the session workspace via the `POST /api/dsh-file-drop` endpoint (with a single-file limit of 25MB). The plug
Read Moredsh-plugin-toggle: DSH Web Plugin Toggle Panel
dsh-plugin-toggle is a community-maintained DSH web plugin designed to simplify plugin state management. It adds a Plugin Switch interface to the Settings page, displaying loaded plugins as cards with descriptions, enabled states, and runtime stages. The plugin supports grouping by category, fuzzy search, and sorting, allowing users to start or stop non-Core plugins at runtime without modifying configuration files, ensuring reversible and safe operations. Core plugins are protected and cannot be stopped, while other plugin operations require confirmation. Each start/stop record is persisted to a local JSONL file, retaining the latest 1000 entries for querying. Installation methods include adding it via the dsh plugin command or as a profile dependency, requiring Node.js.
Read Moredsh-multi-chat: View Multiple Chat Instances Side by Side in DSH Web
dsh-multi-chat is a DSH client plugin with an MIT license, maintained by daetz-coder, designed to optimize the multi-instance management experience of DeepSeek Harness (DSH). The plugin introduces a multi-window grid view into the official DSH web interface, allowing users to display and monitor multiple conversation instances side by side within a single screen, and supports viewing progress on mobile devices through a LAN gateway. Core features include automatically scanning ports to discover running instances, launching new instances with one click within a window wall, flexibly controlling window layouts (such as adjusting column counts, maximizing, and refreshing), and enabling access from phones or tablets via a password-authenticated LAN gateway. Installation is straightforward: executing the specified command adds it to the web configuration, but the DSH web session must be restarted for the changes to take effect. Typical use cases include batch-launching instances, entering multi-window
Read MoreDSH-Bilibili: Adding Bilibili Video Transcript and Keyframe Analysis to DSH
dsh-bilibili is a tool plugin within the DSH ecosystem, designed to address the pain point that Agents cannot directly process Bilibili videos. This plugin decomposes videos into structured data through the `bilibili_extract` tool. Its core features include complete text extraction (metadata, transcripts, comments, and danmaku), with support for automatic transcription and multiple local ASR engines; as well as on-demand keyframe capture, featuring scene detection, sharpness optimization, and local or cloud-based visual description capabilities. A "two-pass workflow" is recommended: first extract text for the Agent to analyze and identify key time points, then retrieve corresponding frames to supplement visual information, ultimately generating a summary. The plugin also includes auxiliary tools such as login and environment checks, and features robust mechanisms like backoff retry for Bilibili API calls. Installation requires Node 18+ and
Read Moredsh-tool-chaos: A Reproducible Fault Injection and Resilience Evidence Plugin for DeepSeek Harness
dsh-tool-chaos is a DSH plugin maintained by cyanseek (MIT licensed, v0.1.0-next.0), designed to transform the verification of failure paths and recovery behaviors into reproducible experiments. Targeting the DeepSeek Harness tool pipeline, it provides five deterministic fault injection actions (deny, error, delay, abort, block), with no runtime randomness, and outputs structured reports in JSON and Markdown formats along with tri-state conclusions. The plugin is disabled by default and operates in a dry-run mode, with bounded injection behavior, and global wildcards must be explicitly enabled. Users can install it via a local tarball or a pinned commit, and the current npm prerelea
Read More