dsh-approval-policy
The DeepSeek Harness plugin addresses the issue of unattended approval requests hanging indefinitely. In dsh >= 0.2.0-rc.2, requests can stall if an interactive answerer is registered but never responds. The plugin introduces a pre-gate in the waterfall approval flow and provides a bounded response window (defaulting to 60 seconds) for unattended requests, after which the request is settled using the default decision. It is installed via npm, configured in `cordis.patch.yml`, and supports control by source (such as subagent, scheduled, or cron) or by session. It does not require additional on-disk state, relies only on host logs, and its default timeout decision supports only `rejected` or `unavailable`.
Read More@aiwayds/dsh-agent-dispatch: Pre-Dispatch Suggestion Plugin for DeepSeek Harness
The dsh-agent-dispatch plugin in the DeepSeek Harness (DSH) ecosystem is designed to provide “pre-dispatch suggestions” for intelligent agents. It does not perform actual delegation; instead, it uses TypeSafe Jev to determine, before a turn begins, whether a task should be handed to a registered agent, and appends a suggestion to the end of the message. The plugin supports three modes. In the default `auto` mode, it first filters requests through a local keyword gate, and after a match, invokes Jev to evaluate four atomic questions: whether to dispatch, to which agent, whether it is risky, and whether it is time-consuming. The plugin features secure key parsing, real-time Capability Gate validation, and dynamic Roster reading, strictly limits outbound data, and is suitable for users who need fine-grained control over agent dispatch.
Read Moredsh-plugin-usage-stats: Token Usage Statistics for All Sessions
dsh-plugin-usage-stats is a web plugin for DeepSeek Harness (DSH) that provides an intuitive dashboard for monitoring token consumption and caching. The plugin does not modify requests or inject prompts; it directly reads adapter data and is ready to use immediately after installation. The dashboard includes overview cards, a GitHub-style daily heatmap, stacked input/output trend charts, model usage donut charts, and detailed tables, and supports switching between 7-day, 30-day, and all-time ranges to help developers monitor costs and caching effectiveness with greater precision.
Read Moredsh-account-glance: DSH Web GUI Sidebar Account Balance and Billing Overview
`dsh-account-glance` is a client plugin for DeepSeek Harness (DSH), designed to compactly display key account information in the sidebar. Core features include real-time balance lookup, peak/off-peak billing countdown (with 50% discount support), and today's consumption statistics and call count. It is compatible with `@linxin666/dsh-usage` and can automatically hide the built-in card from that plugin. Installation is easy, the code is self-contained, data is transmitted via secure routes, and it adapts to both the Web GUI and collapsed sidebar states.
Read Moredsh-model-in-use: Model Menu Occupancy Flag and Gear Memory
The DSH Web Cordis plugin is designed to address the difficulty of identifying model occupancy status in the model menu during multi-session use and the issue of inference settings being reset. Core features include marking occupied models with green dots and group indicators, remembering and restoring each model's inference setting, preserving the collapsed state of provider groups, fixing visual spacing, enhancing the visual effect of sticky headers, and displaying the current model and setting in session floating cards. After installation, DSH must be fully restarted for the plugin to take effect. The plugin does not modify the host and only provides visual feedback and state persistence.
Read Moredsh-whale-inurqq: A QQ Client Integrated with DeepSeek Harness
`dsh-whale-inurqq` is a OneBot 11 protocol plugin that connects QQ with DeepSeek Harness. It handles message routing and session management, with core features including message normalization, persistent bridging sessions, a context handover mechanism, and turn-initiator determination based on `rpcId`. The plugin depends on protocol endpoints such as NapCat and supports inheriting DSH permission presets for QQ-triggered sessions. The author recommends tightening default permissions to mitigate risks. Persistent sessions are enabled by default, and the handover threshold can be adjusted through configuration. Users of the legacy `dsh-qq` plugin need to manually migrate their data directories. After installation, a reverse WebSocket client must be configured in NapCat to enable communication.
Read MoreDeepSeek Harness Plugin: dsh-stock-ticker (Real-Time A-Share Monitoring)
`dsh-stock-ticker` is an open-source plugin for DeepSeek Harness (DSH) designed to monitor A-share market conditions in real time through a Web GUI. It is implemented in pure JavaScript and requires no build step, directly rendering draggable and collapsible watch windows, with support for intraday curves using red for gains and green for losses, as well as hover-based detail displays. The plugin provides a settings page for managing favorite stocks and display parameters, and supports pinned positioning and accidental click/touch prevention. Data is sourced from East Money APIs and requires Node.js 20.3+ along with network access permissions. Users can install it using the `dsh plugin` command and utilize the HTTP API for configuration and data retrieval.
Read Moredsh-stop-kills-jobs
`dsh-stop-kills-jobs` is a client plugin for DeepSeek Harness designed to address the issue where background jobs continue running and trigger model wakes after a user clicks stop. The plugin listens to the UI's `turn/end` event, detects explicit stop actions initiated by the user, and cancels background bash jobs and subagents associated with the current session. After installation, DSH must be restarted; the master switch is enabled by default. The plugin supports configuring whether to terminate different types of background jobs and allows enabling verbose logging for troubleshooting.
Read Moredsh-request-error-dump: A Tool for Dumping Model Request Errors to Disk
`dsh-request-error-dump` is a debugging plugin developed for DeepSeek Harness (DSH) that addresses the issue of hidden error messages when model requests fail. On request failure, it writes the original request body, upstream response body, network exceptions, and DSH context information to a local file, helping developers diagnose specific issues such as invalid API keys and rate limits. Its core features cover four sections—request information, response information, transport-layer exceptions, and DSH context—and support configuring capture scope, header redaction, and file management. By default, the plugin captures only model requests, has zero runtime dependencies, and is ready to use after installation and restart. Developers should observe privacy boundaries and avoid saving request bodies containing sensitive information in untrusted environments.
Read MoreDeepSeek Harness (DSH) Plugin: Native Windows Notifications (`dsh-windows-notify`)
`dsh-windows-notify` is a plugin that provides native Windows desktop notification support for DeepSeek Harness (DSH). Using the Toast mechanism, it推送 notifications when the Agent requires authorization, option selection, or task completion, and clicking one brings the browser window to the foreground. The plugin allows independent notification toggling in DSH settings and provides a persistent rule that forces item-by-item write-back of pending task statuses. It supports only Windows 10/11 environments and Web Profile, and can be installed via Bundle or manual mounting.
Read MoreDeepSeek Harness Plugin: Real-Time Dependency Graph (ewoowe/dsh-plugin-dependency-graph-plugin)
`ewoowe/dsh-plugin-dependency-graph-plugin` is a visualization plugin for DeepSeek Harness designed to intuitively display dependencies between plugins as a graph. The plugin fetches data in real time from the Cordis runtime and supports a dual-view mode, search by package name or service name, status filtering, and canvas interactions such as zooming, panning, and node selection. It uses a golden angle distribution layout to help developers troubleshoot service injection issues and understand the plugin architecture. The project requires Node.js 22.19.0 or later and supports isolated display for browser and Node.js runtimes.
Read Moredsh-glm-quota-dock: Real-Time Monitoring of GLM Coding Plan Quotas with DeepSeek Harness
dsh-glm-quota-dock is a plugin for DeepSeek Harness (DSH) designed to provide a persistent display of GLM Coding Plan quota usage. It displays, below the conversation input box, the usage percentages and reset countdowns for the 5-hour, weekly, and tool windows; clicking it opens a detail panel. It supports idle frequency reduction, version adaptation, and key security, and requires configuring the API endpoint and credential references, with changes taking effect after a restart.
Read Moredsh-shake: DeepSeek Harness Manual Context Reduction Plugin
dsh-shake is a third-party plugin for DeepSeek Harness (DSH) designed to reduce excessive token consumption caused by overly long conversation history. It frees up context space by replacing long tool outputs and large code/XML blocks with short placeholders, while enabling on-demand restoration of the original text through the `shake_read` tool. The plugin includes an intelligent protection mechanism that, by default, preserves the most recent 4,000 tokens of history and only cleans up content that meets the eligibility criteria. Users can manually trigger cleanup by sending the `/shake` command, with flexible configuration to effectively balance context management and content preservation.
Read MoreDeepSeek Harness Plugin: Sidebar Balance Monitoring
`dsh-balance` is a DeepSeek Harness (DSH) sidebar plugin maintained by etony668 for displaying the API balance in an intuitive way. The plugin shows the balance at the bottom of the sidebar, next to the settings button, and supports real-time queries, a 60-second cache, and balance alerts (turning red by default when the balance is below ¥2). It reads local credentials to ensure key security and adapts to light/dark themes. The recommended installation method is to run the `dsh plugin --profile web add @etony668/dsh-balance` command; after installation, configure the API key in settings to enable its use.
Read MoreDeepSeek Harness Local Privacy Gateway Plugin
The “Local Privacy Gateway” plugin for DeepSeek Harness operates under the MIT License and is designed to address compliance risks in data processing. It locally replaces sensitive information in conversations—such as names, phone numbers, and ID numbers—with placeholders, and scans CSV/XLSX files for anonymization, ensuring that data remains visible only locally and is not exposed to the model or logs. The plugin supports custom lexicons, stores data locally in `~/.dsh-privacy-gateway/`, and requires users to enable it manually and download desensitized copies.
Read MoreDeepSeek Harness Plugin dsh-nx: Secure Local Siemens NX Automation via MCP
The `dsh-nx` plugin for DeepSeek Harness (DSH) provides models with secure local Siemens NX automation capabilities via the MCP protocol. It supports part creation, sketching, feature generation, and STEP file export, and includes auxiliary tools such as health checks and state management. Installation is simple and prebuilt assets are supported. Note that this version is experimental: mock results are not written to actual files, and it must not be used in production environments. Developers can refer to the documentation for integration testing.
Read Moredsh-model-deploy
`dsh-model-deploy` is a plugin for DeepSeek Harness (DSH) designed to ensure that models meet performance targets on target hardware through benchmarking. It provides an end-to-end workflow from static model checks and environment probing to local and SSH remote benchmarking, and supports gate thresholds for result validation. Users need to install the required Python dependencies and ONNX Runtime (with GPU support) and configure SSH. The plugin offers model checking, environment probing, and performance testing via the CLI and DSH toolkit, supports dynamic input models, and detects execution provider fallbacks. The current version (V0.1) does not yet support backends such as TensorRT.
Read Moredsh-commands: DeepSeek Harness Filesystem Slash Command Discovery
The `dsh-commands` plugin automatically registers DeepSeek Harness commands by scanning Markdown files. After installation, it maps files in a specified directory to slash commands, supporting YAML frontmatter descriptions and hot reloading. When executed, it appends the body or parameters to the user message. With low priority, automatic skipping of invalid files, and configurable observer behavior, this plugin greatly simplifies command management.
Read Moredsh-lorebooklike-prompt-tool: A Declarative Prompt Injection Layer for DSH
`dsh-lorebooklike-prompt-tool` is a declarative prompt-injection plugin for DeepSeek Harness (DSH). It draws on SillyTavern's lorebook logic while removing role-playing–specific attributes and focusing on prompt engineering. Its core features include hierarchical anchors aligned with DSH's `SECTION_ORDERS`, a blue/green light mode, deterministic probability based on the FNV-1a algorithm, and three-surface regular-expression replacement. Compared with a standard SillyTavern lorebook, it adapts to DSH's dual injection surfaces (system/message) and enforces manually orchestrated rules to ensure audit traceability. The plugin has no runtime dependencies, is not intended for role-playing scenarios, and is designed to fill a gap in the DSH plugin ecosystem for system-prompt injection.
Read Moredsh-decisions-classifiers: OpenAI Decisions API Classifiers for DeepSeek Harness
The `dsh-decisions-classifiers` plugin for DeepSeek Harness (DSH) integrates the OpenAI Decisions API to provide agents with structured text classification and decision-making capabilities. It is integrated across four layers: Agent tools, Cordis services, callable gates, and suggestion cards, and includes a built-in classifier library covering harmful content detection, prompt injection defense, dangerous command identification, and intent analysis. Users can flexibly configure the API base URL, model, and thresholds, and enable a pre-execution callable gate mechanism to perform security interception before critical operations, thereby improving agent safety and controllability.
Read Moredsh-waibao: Delegating Codex Tasks to the DSH Desktop
`dsh-waibao` is a plugin for DeepSeek Harness (DSH) designed to address the difficulty of tracking command-line tasks. It visualizes the execution process in a desktop workspace, enabling features such as task delegation, context continuation, historical access, and completion callbacks. Developers run task files via the CLI, while the native DSH Agent handles execution and notifies Codex for acceptance. The plugin supports Windows, uses named pipe communication, and requires the Codex client, helping developers efficiently handle complex tasks.
Read More@erick0412-dev/dsh-restart: DeepSeek Harness Process Restart and Task Resumption Plugin
The DeepSeek Harness plugin `@erick0412-dev/dsh-restart` provides controlled process restart and state recovery capabilities, supporting restarts through both the user interface and AI tool calls. Compatible with DSH 0.1.7+, it adopts the new slot and Session v4 specifications and features cross-platform native monitoring. The installation command is `dsh plugin --profile web add @erick0412-dev/dsh-restart`, and after a restart, tasks can be automatically resumed to prevent context loss.
Read Moredsh-prompt-studio: DeepSeek Harness Prompt and Preset Evolution Workbench
`dsh-prompt-studio` is a prompt engineering and regression evaluation tool designed specifically for DeepSeek Harness, aiming to address the lack of objective grounds for adjusting Agent presets. It integrates real-time editing, a Git-like version timeline, and Golden Test regression testing, supporting draft saving, version rollback, and automated performance scoring. By enabling hot updates through dynamic Cordis injection and combining CLI installation with a visual workflow, it ensures the controllability and safety of Prompt iteration, making the optimization process for Agent presets more standardized and traceable.
Read Moredsh-codearts: Integrating Huawei Cloud CodeArts Models into DeepSeek Harness
`dsh-codearts` is a plugin maintained by elegant01, designed to address the isolation between DeepSeek Harness (DSH) and Huawei Cloud CodeArts models. By using OAuth2 login and automatic credential renewal, it seamlessly integrates CodeArts model capabilities into the DSH conversation flow. Core features include multi-account rotation to manage concurrency limits, automatic model discovery, usage statistics, and image input support. After installing the plugin, users can log in with one click in the DSH settings and call CodeArts models directly in conversations. The plugin also provides a command-line tool for managing status and running diagnostics. This project is for personal learning and research use only.
Read MoreAgentActa: Local AI Agent Request Log Panel
The DeepSeek Harness plugin `agent-acta` is a local resident service designed to aggregate fragmented logs from multiple AI agents. It supports 31 agent integrations and provides detailed analysis of per-turn duration, token usage, cache hit rates, and tool-call details. The plugin runs entirely locally, with no network reporting or configuration injection, and supports desktop floating cards, an MCP Server interface, full-text search, and single-turn log replay to help developers efficiently monitor agent runtime status and costs.
Read More