Plugins

Native Rapid-MLX provider for DeepSeek Harness (dsh) — dsh reads model facts from the server instead of your settings.yaml.

Automatic context compression for DeepSeek Harness: compresses tool outputs before they reach the model and keeps every lossy compression reversible via CCR retrieval tools.

Model selector for the DSH composer that keeps the native button and menu and takes over only when the small dropdown would open: search across providers, probe any model with a real streaming request to measure TTFT and TPS, and browse a leaderboard of up to 100 rows built from OpenRouter's catalogue, where clicking a row searches local channels of that name.

Knowledge base as a canvas in the right sidebar: notes are nodes, and nine tools let the model read, write, link and remove them. A node can carry pictures and documents, and reading one returns their text — Office files extracted in place — while pictures can be handed over as images.

DSH input assistant for next-message prediction, draft autocomplete (text completion), and prompt polishing/rewrite. Session-local suggestions; Tab accepts into the draft without sending. Requires a configured model; calls may incur charges. Source-available; modifications require prior written permission.

Cross-session persona and diary for DeepSeek Harness: a local gate decides whether a turn is worth remembering, then one extra model call turns it into a first-person diary entry plus an impression of the user, where every impression carries the specific moment that produced it. Later conversations inject about 2000 tokens of that as a self-memory block in three registers: a persona core, a section for judgements the model wrote itself, and the diary. There is a tool for the model to record a judgement in its own words, and a plain markdown hand-note file the human can write into which the plugin only ever reads. A sidebar panel lets you read, forget or export the archive, and forgetting demotes an entry rather than deleting it, so an old one still surfaces on a strong topical match. Local-only, secrets are stripped before anything is sent, and the archive is a plain folder you can read, edit or carry to another account. Zero runtime dependencies, no build step.

Google Antigravity (agy) OAuth auth + model access plugin for DeepSeek Harness: multi-account pool, 429 rotation, device fingerprinting, CLI and web login.

Continual self-refinement loop: persistent memory, periodic review-and-refine, cross-session shared knowledge, and automatic rollback through a model-callable harness_refine tool.

Reads ComfyUI PNG metadata in a model-facing comfy_png_inspect tool: point it at one PNG or a whole output folder and it reports, per image, the LoRA execution chain with every strength value, sampler settings, base model, CLIP, VAE, latent size, both prompts and bypassed nodes.

Adds OpenCode Zen free models to DeepSeek Harness as one provider, backed by a managed, isolated OpenCode runtime, with a Settings page that lists each model and tests it.

Brings WorkBuddy desktop-app experts into DSH: a settings-page market scans the local WorkBuddy expert directory (experts land there after you summon them in WorkBuddy) read-only for browsing, search and category filtering, with one-click install (exported to ~/.dsh/experts/ as role prompt + skills + avatar), update and uninstall; a picker capsule next to the input box switches the current session's expert — swapping role prompt, skills and an optional tool whitelist wholesale while keeping full session history, applied at the next model-request boundary.

Floating macOS-glass DeepSeek usage panel for the DSH web UI: live official balance + token usage (today, session, 24 h smooth trend, per model) without opening platform.deepseek.com/usage. · DSH 毛玻璃用量面板：官方余额 + Token 用量

Agent governance kingdom for DeepSeek Harness: territories, role bindings (bind/unbind/session with model/agent identity), and a plan→assign→execute→review task loop where Worker claims never auto-complete (Claim ≠ Fact), with a local GUI operations console.

Telegram mobile remote that attaches the phone to a live DeepSeek Harness Web session: workspace-then-session picker (/sessions), /last for the previous Q/A, /model switch, and native Rich Message replies.

Liang Saint Slider as the DeepSeek Harness model + thinking-effort selector.

Detailed model usage and cost analysis, with per-session usage stats and usage queries for GLM, MiniMax, OpenCode Go, DeepSeek and more.

Aggregated multi-upstream web search for DeepSeek Harness: replaces native web_search with model-native grounding and agent search APIs, source attribution, and rich citation cards.

Thinking-loop guard for dsh: observes the llm/stream waterfall and breaks a model that degrades into a loop. Detects two per-call shapes (reasoning-only calls, restated-material calls) and re-fires instead of latching; plus two mid-stream breakers that en

Custom modes / assistant manager: give each mode its own system prompt in the settings page — a custom prompt that is a plain file the agent loop re-reads before every model step, so an edit applies to the running session on its next step — and start it from the official standard / PTC / minimal / Cordis composition, toggling the plugin rows it mounts one by one (a custom plugin composition per mode), renaming it as you like. Several assistants coexist with their own prompts, so one machine can keep a coding assistant, a chat companion and a roleplay (RP) persona side by side — a multi-agent setup you pick from per new session. Rows left untouched stay byte-identical to the shipped ones, platform conditions included.

Kiro model provider with Builder ID, Identity Center, Google, GitHub and credential-import sign-in, automatic token and profile refresh, account usage, live model discovery, selectable reasoning efforts, streaming tool calls, provider stop-reason and context-overflow handling, and token accounting that uses Kiro's exact counters when it sends them and otherwise scales the context-usage percentage it does send by the model's advertised window.

Project file explorer in the details column (multi-select, drag-move, upload, download, rename, trash, system-open) with an in-session read-only preview tab for Markdown/HTML/JSON/CSV/notebook/PDF + code highlighting, a live model throughput badge, and a multiline GoalBar.

AI-for-Science research workflow skill for DeepSeek Harness, aimed at researchers and grad students using agentic AI: multi-angle literature review with per-angle files, experiment design with a caveat list, figures and paper reading; each job returns files plus a cross-verification brief for a second model. Zero deps, MIT.

Analyze usage across models, providers, workspaces, and time ranges — including custom start/end dates over the full retained history — with live cache-hit visibility and separate model and provider breakdowns. Includes hourly single-day trends, daily cross-day trends, a 53-week GitHub-style heatmap, usage streaks, model and workspace Token-share donut charts with Token, cost, and percentage details, independent combinable filters, cache-efficiency trends, DeepSeek account balance, workspace aliases, CSV export, and a durable ledger that preserves successfully flushed usage after session deletion — and after a workspace is deleted, where its recorded usage is summed into one Deleted row — and rebuilds incrementally — on restart, unchanged sessions are applied from the ledger without re-reading their logs (a per-session log revision is the change signal), and only changed/new events are folded. The workspace registry is kept fresh automatically through the DSH domain/changed probe: only added/removed workspaces are reprocessed incrementally, unchanged workspaces reuse their computed aggregates and ledger with zero rescan, and usage is strictly limited to DSH-registered workspaces (unregistered cwds are ignored). After a scan, lightweight revision status fetches full history only when the Host/data state changes and exposes non-sensitive sync health (revision reuse, rereads, ledger recovery, and failures). Token accounting is explicit: input is fresh (cache read/write and reasoning are separate buckets) and zero-usage replays never overwrite recorded usage. Chinese uses local time, English uses UTC; API access is loopback-only.

Not another memory plugin: ODR is an evolving-worlds self-improvement loop (Dream-RSI) — attempts are recorded into a Discovery Tree, and between runs thousands of strategy variants are replayed against that recorded world; a strategy, recipe or lesson is promoted only after surviving counterfactual replay, never on the model's word alone, and Sentinel gates tool calls before the budget burns; the LLM stays frozen, installable from npm as dsh-open-dream-rsi.