Plugins

Inline Command Code plan usage badge in the DSH composer toolbar beside the model selector: 5h and weekly window usage plus remaining credits, clicking the badge opens an upward detail popover with reset countdowns; reuses the commandcode/report remote from @mars-sea/dsh-commandcode-provider.

Deterministic loop guard: denies a tool call repeated with identical arguments and caps per-agent calls, in code outside the model; incident log lets a parent interrupt a stuck subagent.

Installs a skill-router skill that semantically searches a local skill corpus on demand, keeping the corpus out of the per-turn model catalog.

GLM (Zhipu bigmodel) web search provider - routes ctx.web search through Zhipu's Anthropic-compatible endpoint with the native web_search server tool, billing your existing GLM quota or Coding Plan instead of the DeepSeek platform. Includes a translation layer for Zhipu's non-standard tool_result response layout.

Probes Windows-host Ollama, LM Studio, vLLM, and llama-server from WSL, compares ctx, and reports /v1/models apiReady versus TCP.

Reports Docker CLI, context, and daemon reachability in WSL, plus vLLM :8000 /v1/models health and GPU runtime hints.

Text-to-image and image-to-image generation via OpenRouter's unified Image API: configurable model aliases with parameters gated against the live model capability listing, reference-image inputs, a settings card, and inline chat display of results.

Install and call external Agent CLIs (Codex, Claude Code) from DSH without touching the system install. Each CLI gets its own Provider, model, reasoning effort and permission preset, callable directly or as a native subagent.

Codex-shaped coding tools and agent presets for GPT models on DeepSeek Harness: exec_command, write_stdin, apply_patch and view_image, plus OpenAI Chat Completions and Responses API routes, bridged to DSH-native plan, ask, approval and compaction. Long-running commands become DSH background jobs with bounded head/tail output readable through job_output, and each subagent type can be pinned to its own preset, provider, model and reasoning effort.

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.

Neutralize tool calls with invalid JSON arguments on the wire so one malformed model generation cannot brick a session against strict OpenAI-compatible servers.

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

Command Code LLM provider for DSH Desktop: registers the commandcode route with a live model catalog, configurable request and stream-idle timeouts, multi-account rotation on 429/401, plan-aware model filtering, an OAuth login flow and a bilingual usage and billing panel.

Add missing reasoning-effort declarations to custom pi-ai models through DSH settings, enabling the native effort picker and handling models added later.

Registers Kimi, Grok, Codex, and MiniMax as DeepSeek Harness model routes, vendor subagent recipes, and kernel_status / kernel_run / kernel_switch tools.

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.

Rule-based multi-provider model routing for DeepSeek Harness with pre-first-token failover, cooldown circuit-breaking, usage accounting, and a status API.

A lightweight plugin that adds model thinking token statistics to the bottom Dock and the end of each conversation.

Browse and govern the skill library from a floating-ball panel with three tabs — skills, call ledger and quick start. You choose the skill root yourself through a native directory chooser (defaults to ~/.dsh/skills, no path is shipped with the plugin); skills are grouped into two levels of categories, with manual, keyword-based and AI grouping available and a tree or flat view toggle, and grouping is purely visual so no skill file is ever moved. Descriptions are read from each SKILL.md frontmatter, search filters by skill name, description or original text, and a card opens the full SKILL.md. Skills can be switched on and off per card, per category or for the whole library — the toggle rewrites disable-model-invocation in that SKILL.md and writes a SKILL.md.bak-toggle-<timestamp> backup before saving, rolling back if the write fails; disabled cards grey out until restored. The call ledger separates automatic from manual entries, records skill successes and failures by listening to tools/result events, and reports call counts, success rate, trend and a ranking, reusing an existing ledger file found in the library. Hot reload picks up skill edits without a restart. The DSH settings page gains a floating-ball section that shows or hides the three balls, offers one-click restore of a ball dragged off-screen, and picks the shared 130-skin catalogue for all three balls.

Workspace-first enhanced sidebar for dsh Web that shadows the built-in workspace browser at low priority (one-click toggle back to the official one) and adds soft workspace archive/restore, per-workspace SVG icons and accent colors, drag reorder and title/path/session-content search, plus a ⌘K Spotlight palette fuzzy-searching workspaces and sessions; the host half ships read-only workspace_find/workspace_list model tools and /workspace-find /workspace-list slash commands. Bilingual zh/en.

One-click search-engine switcher beside the DeepSeek Harness model picker; changes the dsh-free-search provider, handy on overseas university networks that block the DeepSeek Official search endpoint.

Composer code-quote collapse for the DSH Web UI: pasting a `path:line` header followed by code folds the input into one compact inline quote chip, and at send time the `agent/pre-step` hook injects the full pasted snapshot to the model as a separate context message. 0.5.x also folds sidebar file-selections that insert only a `path:lines` line (large selections), fetching the code from disk by session cwd, and preempts Ctrl+V paste interception.

Raises Node's HTTP headers/body timeouts for the whole DSH process (default 30 min, proxy-aware) so Ollama and LM Studio models that stay silent while thinking or generating a large tool call are not cut off at 5 minutes; replaces the node_modules undici patch.

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.