Plugins

Prompt-polish template injector for DSH: click ✨ in the composer to inject a "polish my prompt" template (current draft auto-embedded), no send, no network, no model — the main agent rewrites with full context. Replaces narrative-prompt-polish (2026-09-14 migration).

Model-autonomous context compaction for DeepSeek Harness: a mind_map tool surveys the agent context surface (stable seq ids, roles, token weights, clear-boundary markers, the clear-mind playbook), a clear_mind tool replaces a chosen conversation-history span with a model-written checkpoint through the native compaction transaction, call and result pairs self-collapse into one-line tombstones, and a proactive reminder nudges the model when the context grows long or a turn runs too many steps.

Model-callable compact_agents tool that force-compacts context in every live session (main session, ordinary sub-agents and AgentTeams members alike), queues busy targets and compacts them when the turn ends, reports the shadowed node count and estimated tokens per target, and ships a browser form for the trigger ratio, retained ratio, controlled-phase output budget, auto-continue budget and an end-of-turn pre-compaction ratio that compacts near the trigger line once a turn ends, so a reply no longer opens with a wait for the summarizer.

Per-reply read-aloud with a multi-provider settings page — MiniMax or any OpenAI-compatible /audio/speech endpoint, with voice, emotion, speed and model selection plus an auto-read toggle.

Section icons for the settings navigation: replaces the uniform gear icons with per-section glyphs and fills the plugin-market card avatar slot with each plugin icon; a page reload after uninstall restores the official icons.

Evidence-governed experience memory for DeepSeek Harness

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.

DSH web plugin: a VS Code-style code editor view (file tree, Monaco, ripgrep search/replace, Markdown preview, git diff).

Model picker for the DSH web composer that replaces the model seat: a search-first popup whose left column indexes the suppliers and scrolls the grouped list, with a favorites group pinned at the top, image icons separating native from bridged vision input, and a reasoning-effort panel opened from the right zone of the trigger.

Optional Obscura headless-browser tools from WSL (obscura_status / obscura_fetch / obscura_mcp_hint). Not the same as dsh-wsl-browser win_open_url (Windows GUI browser), and not a replacement for dsh-wsl-fetch web_fetch.

Language-pack framework that rewrites model replies: A is an independent LLM rewrite, B injects FTS5 terminology retrieval into that rewrite, C applies local dictionary and regex rules, and composite runs a pipeline such as B then A then C. Ships a GitHub pack marketplace, a ZIP installer with path, size and file-type checks, and a mandatory risk disclaimer gate.

Plan review enhancement for DeepSeek Harness Web: reply to the plan-review card inline to have the model revise the plan, and browse, compare, edit and archive every plan version in a full-screen overlay.

Voice conversation mode for the DSH Web GUI: tap the mic to start a full-screen call, talk hands-free, and hear each reply read aloud as it streams. Bilingual (Chinese/English) UI, light and dark themes, adjustable speech rate, and switchable voices.

Adds an avatar to the left of each turn's final AI reply and to the right of each user message; upload your own image and set its size and circle or square frame.

Manageable one-tap text chips above the session composer; idle sessions queue the message and running sessions steer at the next safe boundary without rewriting the draft.

Browser patrol & website inspection plugin for DeepSeek Harness (DSH). Teach once, replay deterministic runbooks with managed Chromium, screenshots, credentials, checkpoints and selector recovery.

Builds a motor-repair replacement-parts BOM from the fault and motor rating, listing stock, lead time and domestic or imported alternatives.

Combines historical supplier quotes and service rates into onsite-repair, factory-overhaul, green-remanufacture and replacement options for a motor, then recommends one by total cost of ownership.

Adds an Execution graph conversation-view tab that renders each turn's prompts, replies, and tool calls/results as a foldable timeline with a detail panel and PNG export.

Image generation workbench and agent tool for DSH: one tool (ark9_generate_image) paints text-to-image and image-to-image via any OpenAI-compatible API. Channels are entirely user-configured - no bundled endpoint, you bring your own baseURL, key and model name (manual input supported). Agent-initiated generations are approval-gated by default: requests queue in the floating panel and nothing bills until you approve; approval is non-blocking (the agent replies "awaiting your approval" immediately, and results, denials or timeouts are injected back so the agent reports them in the next turn). Floating glass workbench with five tabs (Generate / Approvals / Prompts / History / About), aspect-ratio grid with quality-budget + 16px-alignment sizing, transparent background, batch up to 10 images as aggregated sub-tasks, prompt library with custom JSON sources fetched through a host proxy, persistent generation logs with one-click retry, multi-channel aggregation with per-channel model fetching, bilingual UI (zh/en), stroke-SVG icon set, config import/export. Registers as a Better Sidebar tab when dsh-better-sidebar is installed; stacks above the dsh-cua FAB when both are present.

Automatically sends a "continue" prompt when a reply is cut off at the output token limit, resuming long answers without manual input; aimed at self-hosted Ollama deployments with small output caps.

Injects a fixed system-prompt section so the agent always replies in Simplified Chinese — including the short narration between tool calls — unless you explicitly ask for another language. Applies to every session and preset without overriding the persona.

Three independent language channels for model output — tool-call descriptions (every tool-call card title, including run_code in PTC-style presets), thinking, and replies — each set to follow the web GUI language, force a specific language, or turn off (default: only descriptions, following the GUI). One global runtime-context prompt directive on the host; the client half reports the GUI language; a Settings plugin card switches the modes.