Flux et automatisation

DSH 插件：把网页大模型接入 Agent 工作流——真实浏览器驱动，支持代码审查、翻译、红队等多种角色。

Surveys local code and GitHub/npm for reusable implementations before development, evaluates reuse versus rewrite, and checks architecture coupling.

Evidence-first self-evolution controller: bounded Cordis candidate generation, one-shot real-Loader admission, Harbor evaluation, and a crash-resumable journaled lineage.

Per-workspace project kanban board: nine kanban_* model tools for agent-driven planning plus a browser board UI, with disk persistence, undo, templates, and archiving.

Runs scheduled coding tasks in isolated DeepSeek Harness sessions, managed from Settings or the agent.

GPT plans dependency DAGs, DeepSeek workers execute tasks with adaptive parallelism, and an optional GPT reviewer audits incremental Git diffs and drives repair rounds.

Cron scheduler with a sidebar UI: run shell, Python or Node scripts, skills, or agent tasks on a schedule, with push notifications and session archiving.

Adds persistent execution contracts, recursive work graphs, critical clarification, and evidence gates for long or underspecified Harness tasks.

Un plugin de méta-agent natif au modèle pour DeepSeek Harness : le modèle synthétise à l'exécution des workflows spécifiques à la tâche à partir de cinq patrons de workflow agentique (chaînage de prompts, workers parallèles, routeur, superviseur, boucle d'évaluation) et coordonne les outils et les sous-agents.

Ajoute des artefacts de spécification, des skills, des commandes, une implémentation adossée à des objectifs, et un contexte de progression des tâches.

Tâches planifiées basées sur les réglages, exécutées à heure fixe ou pendant les heures creuses de DeepSeek, avec des planifications ponctuelles et quotidiennes adossées à un état de tâche durable.

Redéfinitions par appel du modèle, du fournisseur, de la persona et du toolFilter pour la délégation aux sous-agents, avec des références @preset: et des identifiants composites fournisseur/modèle.

Project stack detection and capability recommendations for DeepSeek Harness.

A local-first, human-gated authorial film workflow for DeepSeek Harness

Google Antigravity CLI (agy) as a DSH product subagent: delegate tasks to a Gemini coding agent with your Antigravity login (no API key), with conversation follow-ups, background streaming progress, image analysis by path, and a pop-up login window.

Cron/one-shot scheduled tasks: run shell commands or fire webhooks on a schedule, with optional ServerChan/DingTalk/Feishu/webhook delivery of results.

Provider-neutral CI/CD capability for DeepSeek Harness with GitHub Actions support, bounded failure evidence, source-aware diagnostics, and approval-gated rerun/cancel controls.

Shut down Windows automatically after DeepSeek Harness tasks complete — a floating arm button plus a cancellable countdown banner (shutdown /a), zero runtime dependencies.

Minimal prompt x full PTC tools: one-line persona, Code Mode SDK, subagents, workflows, web search, and Windows Git Bash.

Long-task orchestration loop: Codex plans/scores/reviews, DeepSeek executes, Kimi supplements (vision acceptance, summarization, cross-check).

Dispatch self-contained tasks from DSH to 11 external agent CLIs (Codex, Claude Code, TraeCode, OpenCode, Gemini, Cursor, Kimi, Qwen, Copilot, WorkBuddy, Grok): a host-only bundle registering the conductor_dispatch tool plus an auto-triggered conductor skill.

Proactive scheduling: an adaptive heartbeat plus a value network (urgency, relevance, interruption cost) decides when the agent should speak unprompted.

A structured multi-panelist review workflow for the DeepSeek Harness: N panelists score in isolated subagents, blind to each other; a chair reconciles disagreement by anchor match then Δ-level adjudication; an independent critic audits the process. Works for design, code, paper or project review.

A project-agnostic multi-agent collaboration protocol for the DeepSeek Harness: loop guards that stop runaway agent loops, a six-field handoff format, risk-tiered review routing, and a fixed verify→commit→push→review→close sequence. One skill instead of the same rules copy-pasted into every agent's instructions.