Plugins

Keeps an unfinished task list visible across turns. DSH clears the todos projection at the start of every turn and expects the model to rewrite the whole list, so when a turn is interrupted the plan is durably in the session log but permanently gone from the UI — and todo_write is write-only, so neither the user nor the model can read it back. This plugin registers one ordinary session/event listener, remembers the last list per session, and when a turn/start arrives whose remembered list still has unfinished work, re-appends that list so the projection stops reporting an empty plan. No fork of tool-todo and no waterfall: a plugin cannot replace the projection (the registry throws on a stateVersion mismatch), so appending the event the fold reacts to is the reachable seam. Deliberately narrow — only unfinished lists, at most one event per turn start, fails open. Verified by a controlled A/B on real sessions: identical runs differing only in whether the plugin was installed, folded through the real projection definition. Zero dependencies.

Keeps one reasoning-only turn from making an entire session unusable. When a turn produces no visible text and no tool call, the assistant message is persisted empty, the gateway then rejects every later request in that session with 'content or tool_calls must be set', and the session is dead with no way back through chat. This plugin registers a single llm/stream waterfall listener and, only when the turn carried nothing visible, injects a small text block immediately before the terminal finish, so the persisted message is never empty. Ships the fixture DSH's own mock cannot express (a strictly reasoning-only SSE server, since llm-mock-server forbids an empty successText and always appends text after reasoning), plus an end-to-end reproduction that replays a real persisted session through DSH's own serializeMessages. Preventive, not a repair. Zero dependencies, one file, no process or filesystem access.

Modular DSH Web skin pack with official fallback, OpenBMC and UEFI appearances, and verified stable-Release updates

Automatic model-tier routing for DeepSeek Harness: one user instruction enters, one tier decision comes out — complex intent is planned on the strong tier and implemented on the cheap tier, simple intent stays cheap throughout, with a deterministic guard and a TTL escalation fallback.

DSH GitHub connector: review PRs, triage issues, debug CI, prepare changes (18 tools, gh-first + REST fallback, approval-gated writes).

DSH environment management: auto-installs a Rust/Tauri desktop host (downloaded per-OS from dsh-desktop-host releases, source build as fallback), one-click runtime updates, Windows platform compatibility, official DeepSeek billing, and a built-in code editor for the workspace.

Coding retrospect distilled memory for DSH: distills rejected solutions, pitfalls and constraints from each turn via a local LLM, tags them into MemOS (type:rejected_solution), and boosts weighted recall on coding tasks via agent/pre-step.

Browser notification plugin for DeepSeek Harness - system toasts when a task finishes, a run errors, or your input is needed (question / approval / plan review), click to open the session, per-type toggles, and no false 'done' while the main session waits on its subagents.

Background notification plugin: desktop notification, with a tab-title fallback, when an approval prompt or an ask-user-question card appears while the page is in a background tab. The host queues approval and question events behind a same-origin endpoint that the client polls every 3 seconds; clicking the notification focuses the page, and the tab title is restored once the page becomes visible.

Kimi Coding web search and web fetch providers for the DeepSeek Harness — one coding-plan key fills both ctx.web seams (POST /v1/search and /v1/fetch), a settings card in the GUI stores it via the credentials service, and the fallback chain reads the Kimi CLI config for zero-setup installs.

Saves a Windows clipboard image into a WSL file (clear ok:false + hint when clipboard has no image).

Parse Chinese e-invoice PDFs (XML attachment first, text-layer regex fallback) into structured JSON with amount cross-check, and generate expense reports (Markdown / xlsx).

Governs multi-model access, monthly credits, manual/quality/credit/auto routing, failure rerouting, and usage auditing for DSH.

Per-provider concurrency gate on the llm/stream waterfall: holds surplus model requests in a FIFO queue inside DSH so main agent, subagents and compaction take turns on a single-slot local server (llama.cpp --parallel 1) instead of timing out; maxConcurrent / maxQueued / queueTimeoutMs.

Routes DeepSeek Harness outbound traffic through a local Clash proxy, with automatic fallback to direct when the proxy is off.

Connects a Feishu self-built app over WebSocket, routes each chat to an independent DSH agent session, and streams replies through a Card JSON 2.0 card with plain-text fallback.

Text-to-image and image-to-image generation inside DSH through OpenAI-compatible endpoints, with an agent tool, multi-provider fallback, and a side-panel workspace with history.

LLM debug console for DeepSeek Harness: captures the full content of every model call at the llm/stream waterfall (options, stream chunks, usage, wire protocol) and shows it in a DevTools-style docked panel, with shared-prefix context diffing, wire-JSON reconstruction, JSON export and a runtime-tunable capture policy.

Reliable todo panel that survives restarts (official panel only re-renders on write — fixed via projection pre-warm) with three-state completion verification: evidence exists → verified, fake evidence → blocked, no evidence → unverified badge; settings toggle to fall back to official behavior.

Hybrid model router for DeepSeek Harness: declarative rules, user-configurable cost control (cost-first / quality-first / balanced), optional LLM task classifier, and failure fallback chains.

Add six languages (繁體中文, 日本語, 한국어, Français, Deutsch, Español) to the DeepSeek Harness Web UI language options, with hand-polished per-language translations and automatic fallbacks (Simplified-to-Traditional conversion for zh-TW, English for the others) covering upstream updates and third-party plugins.

Falsification-driven correction loop for agent sessions: risky work commits an evidence-checkable claim first, deterministic gates verify it, falsified claims force mutually exclusive replacement hypotheses with discriminating experiments, and every event lands in an append-only evidence ledger.

Per-file-type multimodal chains: preset processing models per wildcard convert image/video/audio files into prompt tokens before they reach the text-only session model, with per-preset fallback chains and a Multimodal settings page.

pi-web-access für DeepSeek Harness: Multi-Provider-Websuche mit automatischem Fallback (SearXNG → Tavily → Brave → schlüsselloses DuckDuckGo) plus Extraktion von URL-Inhalten, eingebunden in die Webschnittstelle, sodass das eingebaute web_search-Tool von piwa angetrieben wird. Search engin