プラグイン
DeepSeek-Harness プラグインを閲覧・絞り込み・インストールできます。
690 件のプラグインが見つかりました
dsh-profile-lab
young-tim/dsh-profile-lab
ガイド付きケース作成、アサーション、Web比較レポート、復旧、ポリシーゲートにより、再現可能なDSHプロファイルおよびパッチ実験を実行。
dsh-hover-approve
wydddddcool/dsh-hover-approve
DSH Web のサイドバーの各セッション行に承認/回答バブルを固定表示し、会話を開かずに認可、質問、プランレビュー、ブロックされた目標をワンクリックで処理できます。
dsh-xiaohongshu-viral-note
xuboboo/dsh-xiaohongshu-viral-note
Xiaohongshu/REDのバズノート用エージェントスキルをバンドル:ホットノート調査、ノートの生成/書き換え、検証、認証済みアカウント分析、QRログイン、制御された公開。
gh-tea-npm-dsh
tommyfang2077/gh-tea-npm-dsh
GitHub (gh)、Gitea (tea)、npm CLI 向け DeepSeek Harness (dsh) プラグイン:検出 / インストール、ガイド付き認証 / 設定(npm トークン ローテーションを含む)、イシューおよびパッケージ ツール。
dsh-plugin-qr-connect
mervyn-teo/dsh-plugin-qr-connect
Settingsの上部にQR-codeボタンを配置。認証付きリバースプロキシを通じてスマートフォンがWeb UIに接続できるようにします。
dsh-web-auth
summersec/dsh-web-auth
DeepSeek Harness Web GUI 用のトランスポート層認証ゲート:サーバーサイドセッション、HttpOnly Cookie、IP ベースのログインスロットリング、scrypt パスワード CLI。
dsh-messaging
534119219/dsh-messaging
DeepSeek Harness の統合メッセージングゲートウェイ:27 の IM プラットフォーム(Telegram、QQ、WeChat、Discord、WhatsApp、Feishu など)を 1 つのプラグインに、QR スキャン認証、プラットフォームごとのワークスペース、スラッシュコマンド、Web セットアップダイアログを備える。
dsh-http-tools
lussey820/dsh-http-tools
HTTP/API デバッグツールセット:完全制御の HTTP リクエスト(method/headers/body/auth)、curl コマンドの解析とワンステップ実行、レスポンスを並べて比較できるセッション内リクエスト履歴、ページネーション対応の Link ヘッダー、DeepSeek ブランドの UI カード(オプションの dsh-http-tools-ui)。
DSH-CODEX-SUBSCRIPTION-POOL
eve-146t/dsh-codex-subscription-pool
DSHでChatGPT/Codexサブスクリプションを利用。画像生成とWeb検索をサポート。
dsh-subscription-auth
khellendros97/dsh-subscription-auth
サブスクリプション OAuth ログイン。ChatGPT/Claude/Grok/Kimi のサブスクリプションアカウント(API キーではなく)を使用でき、モデルを自動検出します。
dsh-codex-provider-plugin
damonbao/dsh-codex-provider-plugin
DeepSeek Harness 向けの OpenAI Codex OAuth プロバイダーと、ネイティブな設定 UI。
dsh-codex-agent-bridge
je00/dsh-codex-agent-bridge
DeepSeek Harness のエージェントを、認証済みの Codex App Server セッションに接続します。
dsh-memory-gate
git121995/dsh-memory-gate
CBDC 権限ゲーティング付きの、範囲を限定したローカルメモリ。SQLite + FTS5 によるクレーム管理、説明可能な use/verify/ignore 判定によるスコープ付き想起と完全な監査ログ、/memory コマンド、1回あたり最大3クレーム/1200文字の注入に対応し、追加のモデル呼び出しは不要です。
dsh-doc-defrag
chrischjh/dsh-doc-defrag
Turns a fragmented Markdown docs folder into a few authoritative documents and archives the rest; never deletes files and rolls back byte-for-byte via a per-run journal, with or without git.
weread-dsh
wangjianghu/weread-dsh
WeRead notes, deeply wired into DSH and AI workflows: incremental Markdown sync and export of highlights, ideas and reviews, full-text search with chapter provenance, reading insights (streaks, deep-read ratio, category radar, monthly report), SM-2 note roaming, recent-reading context injected into every model turn, an in-session weread agent tool (sync/search/insights/roaming/stats/store/recommend), an @微信读书 book picker in the composer, clickable tool-result cards, and a cross-host MCP server (9 tools). Three login channels (QR / cookie / official API key), multi-account with instant switching, credentials stay on-device; the loopback API is host-caller authenticated and fail-closed, and the single-reader window ships selected text straight into a DSH session via ask-DSH. Requires DSH >=0.2.0.
dsh-codex
goodjin/dsh-codex
Adds Codex (ChatGPT subscription) models to DeepSeek Harness as a model row, calling the Codex /responses endpoint directly with the local Codex CLI OAuth credential, and shows the official usage windows as a quota pill under the composer. Its HTTP routes have no DSH access check, so a web page open in your browser could call them.
dsh-task-signal
tangyixiao/dsh-task-signal
Task-completion sounds with Codex/ZCode styles, six synthesized tones and custom WAV; per-event controls, host-persisted settings, Windows native background playback, browser playback, quiet hours and subagent filtering. Version 0.1.2 requires synchronous Host request-policy approval and authenticated owner admission before HTTP settings or audio actions; missing or invalid admission services fail closed. Original application recordings are imported locally and are not redistributed. Version 0.1.3 has been verified with DSH 0.2.2-rc.5.
dsh-session-quickbar
879745uhjun/dsh-session-quickbar
Takes over the quick-action strip on the right of every sidebar session row: choose per session which buttons show and in what order, reuse built-in star, todo, pin, archive and copy-id actions, and keep a two-way channel with the sessions that author plugins. Its HTTP routes have no DSH access check, so a web page open in your browser could call them.
dsh-plugins (dsh-remote-relay)
skillpie/dsh-plugins
Reach any DeepSeek Harness host by its 8-digit code through a relay origin: a browser-surface remote control with relay-prefixed GUI passthrough, per-host WebSocket muxing, a visitor fetch/WebSocket shim that rewrites same-origin calls, owner-only HTTP APIs, and optional concealment of the code in the URL via an origin-level Service Worker. Off by default; it must be enabled from the owner panel in the sidebar. Once enabled, it connects out to the author's relay dsh.skillpie.cn, and whoever holds the code gets full control of the GUI on this machine.
dsh-admin-gate
fish133/dsh-admin-gate
Password gate in front of the DeepSeek Harness web UI: admins sign in with a password stored only as a scrypt salted hash, and guests enter without one but are held to an allowlist of /api endpoints and agent tools; several admin accounts may exist, each with its own password; three modes (admin-only, guest, open), and closing guest entry drops every guest already inside; conversation ownership is recorded per sessionId so each account sees only its own conversations, with list and search filtering rather than refusing so "exists but you cannot see it" never leaks. The gate is enforced on the host, not in the page: connection.admit, admitConnection and authorizeIndex decide who a visitor is and issue a Peer carrying that identity, connection/request answers 401 or 403, WebSocket streams are checked when they open, tools.guard() keeps guest-triggered agent turns inside the tool allowlist, and the gate page at /admin-gate is mounted outside /api so it does not need the harness cookie. Five failed logins within ten minutes lock the door for five minutes, and state is written to a temporary file and renamed so an interrupted write cannot leave a half-written JSON. The default allowlists deliberately exclude session/prompt and the read, bash and write tools, since those run as the owner. Its HTTP routes have no DSH access check, so a web page open in your browser could call them.
dsh-vizro
lion231226/dsh-vizro
Vizro dashboards inside DeepSeek Harness: seven native tools for the model contract, production-path validation, data profiling, scaffolding, preview servers and health checks, six authoring skills, and a workbench panel. No MCP server.
dsh-memory (kb-panel)
shehuilaoliumang/dsh-memory
Governed long-term memory for dsh. Each turn is appended to an append-only raw layer; a semantic judge distils it into one Markdown note per memory inside a plain vault, and only notes you confirm are ever injected, within a five-slot budget shared by global and project stores, with explicit supersede to retire old conclusions. Ships a memory page that explains why each note was or was not injected, local BM25 search with pinyin fallback, project stores that stay read-only until you authorize writes, and two-step deletion (a terminal-issued ticket plus a one-time key). The plugin ships inside this repository next to its host-independent rule engine, so it installs from a clone and is pointed at your vault. Its HTTP routes pass DSH Connection admission for Host and Origin before its own loopback and token checks.
dsh-conversation-lock
xiaohuangbo/dsh-conversation-lock
Session-level restriction plugin for DeepSeek Harness: the model can close a conversation behind a yellow notice with countdown, disable the Stop button and both response ratings, exclude the session from official Session Log uploads while turning Session OTel off, and block copying of the transcript — each gated by a user-authorized settings switch.
dsh-novel-studio
linque0/dsh-novel-studio
Novel Studio for DeepSeek Harness: a book library, an outline, character and lore cards and a manuscript editor in one workspace, plus studio_* tools that let the agent read a work, file review findings and save candidate text only after the author confirms. Experimental.