Plugins

Mini-jeu d'attente grille de Schulte dynamique : cliquez 1..N dans l'ordre sur une grille statique ou une roulette tournante pendant que le modèle réfléchit.

Frontend Pi TUI (framework de terminal à rendu différentiel) : markdown en streaming, repli de la réflexion, cartes d'outils, commandes slash, superpositions d'approbation/question, magasin de session dsh partagé.

Plugin DeepSeek Harness : bascule du mode réflexion par requête (enable_thinking) pour les modèles de la famille Qwen3 sur des endpoints compatibles OpenAI.

Moves reasoning that a provider leaks into the visible text channel back into the reasoning channel, and emits a text block when a whole turn of answer text lands only in the reasoning channel. Catches three shapes: `<thinking>`-tagged text, text that continues a reasoning tail which ended mid-token, and an answer-only reasoning turn that stops. Healthy streams pass through unchanged, block order is preserved, and each of the four rules can be switched off in one JSON config.

Qwen mode for DSH: manages local GenieX and Alibaba Cloud QwenCloud providers, normalizes thinking and sampling, detects model capabilities, and shows per-turn latency and decode speed. Its bundle patch replaces the permission plugin's preset table with reply-only, read-only, workspace-write and danger-full-access (the last two as in stock), so any custom presets are overridden.

Connect a locally installed agy CLI to DeepSeek Harness 0.2.2-rc.1, with separate thinking levels, quota display, and Harness-owned tools. Community fork of dannndouglas/dsh-antigravity-acp, beta, verified on Windows x64 only. Needs the agy CLI signed in with your own Google account; installing makes new chats default to Agy models (messages go through agy to Google), switches web search to the agy provider and replaces the built-in compaction. Do not enable it alongside other forks that register the agy-local-acp provider.

The row that is running expands itself and folds back when it settles. A running turn keeps its whole procedure in one height-capped window that scrolls itself; a finished turn merges into the built-in finished header, leaving only the final answer. It works by driving the official disclosure rows rather than drawing its own panel, leaves rows you opened by hand alone, and restores the official settings on uninstall.

Delivers one required user-role runtime context that makes the agent reason in Simplified Chinese and keep a whale-loli in-thinking persona (calls itself 本鱼, lazy-arrogant tone) in both reasoning and the final reply, while proper nouns, identifiers, paths, commands, URLs and error text stay verbatim; persona, extra proper nouns and context order are configurable. Enabled by default for every agent in the profile, subagents included.

Imports chats captured from chat.deepseek.com by the repo's companion Edge extension (loaded unpacked in Edge) into DeepSeek Harness as new sessions; thinking is saved as a sidecar thinking.md, and trimmed, branched or very long chats also get the full transcript.md.

Per-turn reasoning-effort routing that keeps the current provider/model: judges each user message (retry or correction, explicit intent, length, code blocks, errors, task scope) and sets reasoningEffort to a level the model actually declares, so Magpie and other OpenAI-compatible models work too. Ships a composer chip with an auto/manual switch and per-session decision history.

Replaces the composer model picker with a model + thinking-effort picker and an Auto level; Auto scores each turn by how far the conversation has drifted from the main objective and by repeated failures (one extra small judge call per turn), picks the effort from that score, and labels every finished turn with the effort it sent.

Turns scattered thinking notes into a growing constellation of themes, claims, relations and open questions, plus a written synthesis of the argument.

Injects a Chinese-thinking reminder around the system prompt and the per-step runtime snapshot, so the agent reasons in Simplified Chinese instead of drifting back to English as the transcript grows.

Replaces the composer Effort switch with a Claude Desktop-style five-level slider (Low/Medium/High/Max/Ultracode): a pill opens a popover above the composer, dragging snaps with a spring and writes once, and Ultracode maps to the model's maximum reasoning effort. Ships the Ultracode WebGL2 energy grid, a per-session Ultracode mode, and passes modifier-click through to the shipped level list.

Fournisseur SAP AI Core avec OAuth partagé, cinq choix de modèles Qwen, budgets de raisonnement, streaming et appels d’outils ; validé sur cn40.

[LLM] Permet à DSH de synchroniser les modalités d’entrée et le niveau de raisonnement depuis l’amont magpie.

Plugin DeepSeek Harness : demande à chaque agent de rédiger son raisonnement privé en chinois simplifié.

Curseur composer inspiré de Claude Desktop pour régler l’effort de raisonnement du modèle par clic, glissement ou clavier.

Bundle Windows x64 incluant le runtime ACP officiel Antigravity. Délègue des tâches texte ponctuelles avec choix de l’authentification, du modèle et du niveau de réflexion. Utilise host peers pour Harness >=0.2.1-rc.3 <0.2.2-0, y compris les RC 0.2.1 ultérieures et la version stable 0.2.1.

L’historique du composer restaure tout l’état de saisie : contenu, pièces jointes image/fichier, références @ et commandes slash, ainsi que modèle, effort de raisonnement, autorisation et preset d’agent utilisés.

Menu à trois points dans l’en-tête de session : télécharger le journal, passer chaque transcription entre Normal et Verbose et définir ce que Verbose développe (blocs de réflexion, appels d’outils, sortie du terminal).

Fournisseur de modèles Cohere pour DSH : protocole compatible OpenAI par défaut, voie native v2 Cohere pour citations/RAG/budget de réflexion. Adaptateur direct, sans proxy.

Interface DSH monochrome et calme aux accents lavande, animations de démarrage et de changement, espace de réflexion, skills à portée définie et connexions locales facultatives à la voix et à l’animal.