Plugin

Plugin DeepSeek Harness: traduce in cinese i nomi dei livelli di «推理等级» (reasoning effort) e inietta la visualizzazione del livello di ragionamento per i modelli di terze parti personalizzati collegati (il DSH ufficiale non lo inietta per impostazione predefinita); il livello scelto viene scritto nel corpo della richiesta in uscita.

Plugin di ancoraggio RL del primo turno di DeepSeek V4 per DeepSeek Harness Standard: dopo la promotion ripristina tutte le capacità e le Skill.

Controlli componibili di lavoro, ragionamento, qualità e auto-evoluzione revisionata da umani per DeepSeek Harness Web.

Controlli componibili di lavoro, ragionamento e qualità per DeepSeek Harness Web.

Configurazione delle capacità dei modelli nella GUI web di DSH: nell'editor del catalogo dei modelli di «Impostazioni → Modelli», imposta direttamente per ogni modello configurato se supporta le immagini e il livello di pensiero (reasoning efforts), scrivendo nel namespace delle impostazioni llm-pi-ai.

Scegli l'impegno di ragionamento (reasoning effort) per modelli aggiuntivi in DeepSeek Harness, con Off e cinque livelli standard.

Plugin DSH: memoria Honcho v3 automatica a doppio peer, richiamo contestuale, ricerca, ragionamento, conclusioni e diagnostica dello stato

Compila automaticamente livelli di sforzo di ragionamento, finestre di contesto, limiti di output e modalità immagine per i modelli di provider non ufficiali (personalizzati), con dati da models.dev.

Trigger del modello nel composer in stile ufficiale con popup flottante del cursore dello sforzo di ragionamento.

Uses a configurable system-prompt section to request a chosen reasoning language (Simplified Chinese by default), and expands thinking rows by default while preserving rows the user manually collapses.

Remote DSH access that needs no hosting of your own — the Feishu DM chat is the interface over the bot's outbound long connection, with cards for workspace, session (switch and rename), model, reasoning effort, permission preset and balance, and tool approvals and agent questions answered in that same chat.

OpenRouter provider that reads the live /models list, so new models appear without a harness release: models sorted by name with input/output prices, per-model reasoning levels, image input, reasoning_details replay, Anthropic prompt caching, and actual spend per session including subagents from OpenRouter usage.cost, shown under the Web UI composer and by /cost.

Tiered compaction hub: replaces the single 80% compaction cliff with a configurable ladder (default 30/50/70/90) whose tiers escalate from gentle to maximum consolidation laws, folds a deterministic keyword index into every checkpoint so facts accumulate across cycles, and summarizes through a model chain — primary plus up to 3 fallbacks, each with its own reasoning effort. Adds a Compaction module sub-tab and a Settings card; toggling off always restores stock compaction.

OpenCode Go provider that verifies a model against the gateway rather than treating a recommendation as a measurement: the models.dev protocol suggestion is probed before it is used, reasoning levels follow a measured contract, and requests run through an ordered protocol fallback chain. Context window, max output and image-input numbers are declared by a runtime-refreshed models.dev layer rather than probed, with operator overrides above both. Several OpenCode Go credentials can be stored and switched from the settings page.

Wire-level LLM inspector: captures the raw HTTP request and response bodies of model provider calls and attributes each capture to the session that caused it through the llm/stream waterfall. The request pane renders the body as a collapsible tree with per-node sizes — one measured request was 98.6% a single 973-entry array — and the response pane reassembles the OpenAI-style SSE chunk stream into reasoning, content, tool calls and usage. Shown in a Conversation View tab and a loopback viewer page.

Telegram remote control for DeepSeek Harness: list conversations with titles and workspaces, resume paused persisted sessions on demand (preset remount), turn-tracked reply relay, first-line reasoning summaries, inline-button approve/reject with outcome edits, and a bot command menu.

A CSS skin that restyles the reasoning-tier control drawn by hytime/dsh-thinking-effort into a blue-to-violet pill with a starfield and a sheen sweep, with matching light and dark palettes. Requires that plugin to be installed; it adds no UI of its own.

Replaces the chat input model picker with a provider-grouped, collapsible list: groups models by provider, folds vision-bridge mirror providers back into their upstream, and carries the reasoning-effort options each model declares in settings. Tested on DSH Desktop only; the DSH web build is untested.

Zen Mode view tab and foreground time tracker for DeepSeek Harness: while the agent works, a calm conversation view shows running/done status, your latest user message, live activity hints and the final AI reply in markdown with one-line turn stats, plus screen-watching (focus) time tracking with daily/weekly Zen scores, a chat-hide toggle that hides tool calls and reasoning rows, auto enter/exit Zen on task start/complete, an Alt+Z shortcut and a Zen settings panel — pure client-side, stored in localStorage, no network.

Declare any number of LLM providers, each with its own endpoint, credential and per-model parameters (reasoning-effort vocabulary, image input, retry policy), across the OpenAI Chat, OpenAI Responses, Anthropic Messages and Google Gemini wire formats, configured from a settings page.

Codex-style DeepSeek Harness model and reasoning selector with model-advertised effort levels, DSH-native themes, left-clipped radiation effects, and copy-ready reasoning-effort declaration guidance for custom-provider models.

Sets the DeepSeek flash model's reasoning effort per step — thinking off for short cheap prompts, low for a plain request, high for engineering work — and raises it only on repeated tool failures. The model itself never changes; effort max is opt-in.

Smooth think-row experience for the DSH Web UI: think rows expand as a capped 24-line preview that glides to the bottom while reasoning streams, collapse with a short height animation on settle, and the main conversation view follows with a reader-safe smooth chase (no snaps, no yanks).

DSH plugin: resumes a turn that ended successfully with reasoning only, no text and no tool call, by sending one bounded follow-up