Plugins

Multi-provider LLM API balances under the chat input, with presets for DeepSeek, OpenRouter and SiliconFlow.

Per-turn and session-wide LLM cost metering: a dollar cost line under each message, a session cumulative total with hover breakdown, and an LLM+web price auto-refresh tool.

Hand off focused context without waiting for or interrupting the active Agent: freeze a bounded Session snapshot, summarize it asynchronously with an isolated no-tool LLM, and deliver a provenance-bearing capsule to another Session or an authorized Room—quiet by default for Sessions, with explicit wakeup available.

Two-tier model router for DeepSeek Harness — LLM-Judge routing, multi-model fallback chains, exponential-backoff failover, and task-level orchestration (DSH adaptation of pi-shift-router)

Bridge the vercel-labs/skills ecosystem into dsh: LLM-driven skill search, install, and lifecycle for temp/project/global scopes

DeepSeek Harness plugin: query AI Monitor balance / usage from local SQLite — cost-aware agent, auto bookkeeping with the AI Monitor proxy.

Local Markdown wiki as long-term memory for DeepSeek Harness — ported from llmwiki

Tencent CodeBuddy plugin for DeepSeek Harness (dsh).

ProactiveAgent × DeepSeek Harness 插件组：主动记忆 + 主动建议接入 DSH（bundle 聚合包，dsh plugin add 一键安装）

Feishu (Lark) IM bridge for DeepSeek Harness via `dsh plugin add`; one DM user to one persistent dsh session, with full debugging docs.

Scans skills and MCP configs for prompt injection, homoglyphs, hidden Unicode, dangerous shell, and credential leaks.

Text hygiene as a dsh plugin: sanitize untrusted text, scan invisible characters, clean LLM formatting, and escape CSV formula injection.

Tencent CodeBuddy LLM adapter for DeepSeek Harness

Checks the agent’s own claims against the tool output saved in the same session: each sentence carrying a path, a number with a unit, inline code or a URL is graded supported, partial or no evidence, with the anchors it could not find listed. Coverage shows on the composer pill. Heuristic and local — no LLM call, no receipt log, nothing uploaded.

Connects one CLIProxyAPI gateway to DeepSeek Harness, speaks OpenAI Responses, Chat Completions, Anthropic Messages or Google Generative AI natively, and imports the gateway model catalog into a settings page where each model can be selected.

Local remaining-time, elapsed-time, Token-budget and reference-cost estimates for DeepSeek Harness, with local route calibration; for Windows DSH Desktop 0.2.2-rc.2.

Community fork of KyoMio/dsh-zen-remote (do not enable both simultaneously — shares the ~/.dsh/zen-remote state and /_dsh/zen-remote/* wire protocol, and auto-defers if dsh-zen-remote is active): keeps the pairing-code PWA with Web Push and multi-host session relay, and adds a desktop Local / Remote / All workspace switcher, remote directory browsing with folder creation and workspace rename/delete (/v1/workspace-manage), automatic custom LLM provider and API-key sync to paired hosts, and bidirectional cross-host @ session reference hydration (session-ref-bridge). The gateway binds 127.0.0.1 by default; the LAN/Tailscale mode binds 0.0.0.0 behind a private-IP wall, and a paired desktop client runs terminals and file previews with the host's OS-user permissions.

Lightweight cross-session reflection: at the end of each turn it writes 3-5 reusable facts (tool chain, file paths, config items, pitfalls) as a card in .dsh/reflections/{date}.md, and on a new session it injects the last 3 cards at the end of the system prompt. Extraction uses the real ctx.llm.stream; when no model route is available it falls back to a heuristic built only from the session log and labels the source honestly on the card. No timers, no quarantine, no approval flow; cards are plain Markdown you can edit or delete.

Grows a conversation into an editable card tree where each card is one self-contained question and answer, with child cards, same-level branches and in-card follow-ups, and an overview for dragging, re-hanging, cross-referencing or pruning the graph.

Google Cloud Vertex AI (Gemini) provider: ADC or service-account credentials, project and region settings, thought-signature replay for tool calls, rate limiting, and a settings page with a live connectivity test.

DeepSeek Harness sidebar for LiteLLM models, MCP servers, skills, and spend.

Replays the last request while a main session idles, so the provider prompt cache keeps its hit and the next real turn is billed at the cache-hit price instead of a cold re-prefill. Per-window toggle, per-task records, and an estimated net saving.

Shows the balance or Coding Plan quota of the active model provider in the dsh status line, model menu and Current Model label, with background-collected history and trend charts.

Prompt forge for DeepSeek Harness: one draft becomes three parallel rewrites, read for intent first and scored blind by the model, with the sentences each version added highlighted — then bind one back into the composer.