87 results
VCP 视觉通感协议插件 v2（官方 Slot API 版）：AI 正文直接输出裸 HTML，前端以 conversation.chat.node 的 assistant-step 替换渲染器接入 + 复用 v1 渲染引擎（assets/vendor/vcp-engine-v1.js，流式/字体锁权/作用域化/KaTeX·Mermaid 零重写）渲染 vcp-root 卡片，零 bundle 补丁，兼容 rc 与 alpha 全版本线；内置 IO 热窗（可视+预取渲染、远离降级占位）防历史卡累积卡顿；内置开源精选字体 + 设计规范注入
自进化插件：agent 在 session 内随对话按需给自己长出/剪掉能力 —— evolve_add 把 cordis 插件源码落盘并热挂载（下一 step 工具即可见），evolve_remove 可逆卸载，重启自动恢复
Think-step bounce pet for the DeepSeek Harness Web UI: a ball launches in from a random screen edge whenever a reasoning step starts, bounces around with energy loss, can be poked or flung with the mouse, and shrinks away when the think step ends
Embedded C/C++ firmware development toolbox — 4 agents, 8 skills covering FreeRTOS, ISR, NVM storage, Keil MDK, ARMCLANG, HardFault, state machines, architecture, LVGL patterns, and claim fact-checking. Ships a native DeepSeek Harness (dsh) bundle that injects the session-start gate into the first model step.
CAD visualization plugin for DeepSeek Harness: cad_view / cad_info tools with an embedded 3D (STL/OBJ/STEP/IGES) and 2D (DXF/SVG) viewer card in the Web UI
Self-contained Trellis workflow trigger for DeepSeek Harness: per-turn breadcrumb injection via agent/pre-step, a bundled skill provider, and DSH-native task/phase state resolution. No Trellis AGPL code is vendored; workflow semantics are rewritten and licensed MIT.
OpenTelemetry and Langfuse observability exporter for DeepSeek Harness: turn/step/tool/LLM spans, token and cost metrics, and sanitized LLM prompt/completion capture from the session/event stream, with async batching, bounded offline buffering, and retry with backoff.
One-click restart button for the DeepSeek Harness Web UI: a sidebar footer button that restarts the dsh web process (single click, no confirmation step). / DSH Web 界面一键重启按钮：侧边栏底部按钮，单击即重启 dsh web 进程（无需二次确认）。
LLM-as-a-Verifier for DeepSeek Harness: fine-grained reward, Probabilistic Pivot Tournament best-of-N selection, and per-step progress tracking as agent tools.
ADHD behavioral coaching skill for DeepSeek Harness. Guides readers through task execution - breaks tasks into micro-steps, manages overwhelm, provides launch rituals, calibrates time estimates, and recovers from self-blame.
Session execution analytics and audit reports for DeepSeek Harness. Understand how your agent actually worked: steps, tool calls, failures, repeated actions, token usage and verification signals.
DSH web bundle: pause the current turn at the next step boundary (graceful drain), persist everything, and resume later — including after a DSH restart.
Replay a DeepSeek Harness session at its original token cadence — an in-app playback theater with play/pause/step/speed/seek.
Multi-provider account balance display for the DeepSeek Harness Web GUI. Queries real billing APIs (DeepSeek, StepFun, Kimi Coding, OpenRouter, MiniMax, xAI) and shows balance/quota in front of the model selector. Providers without an API-level billing endpoint show a clear 'Balance unavailable' state.
An intent-calibrated DSH discussion mode that prevents complex conversations from drifting and turns them into evidence-based next steps.
DSH recap plugin: distills every model request (turn:step) of a session into one sentence, appended incrementally with a cache-friendly prompt (history sentences first, new delta last). Side-effect-free for the agent loop.
GPU-aware execution layer for DeepSeek Harness: gpu_status / gpu_exec / gpu_run_bg tools, per-step GPU context injection, automatic CUDA_VISIBLE_DEVICES card selection
过程精简：把 DSH Web GUI 里的 Think / 工具调用等技术过程行折叠成友好摘要，面向文字工作者的非技术呈现。Collapses technical process rows in the DeepSeek Harness Web GUI into a friendly summary pill.
DSH Web 插件：基于 DeepSeek 官方 BPE 分词表，在 composer 下方实时渲染 token 状态带，显示流式 TPS、输出 token 与首字延迟，并对比上次 step 的估算与实际偏差；纯插件自包含，不改 DSH 源码
Memory protocol enforcement plugin for DeepSeek Harness — forces memory_weave before tool calls, auto-ingests session turns, injects memory context into agent steps. Bridges the opencode-memory MCP server with a hard protocol gate.
DeepSeek Harness plugin: the agent publishes a project plan (steps) and a project framework (architecture graph); the browser renders the steps in a right-hand panel and the framework as a downloadable diagram in the center view.
DSH Web GUI 会话统计插件：在顶部「对话」「轨迹」之后新增「会话统计」tab，列出每个对话的轮数、步数、token 用量（输入/输出/缓存）、模型调用、工具调用与耗时等数据，点击会话可查看逐条调用明细。数据直接解析 ~/.dsh/sessions 下的会话日志（zstd），无需常驻监听，历史对话同样可查。
DeepSeek Harness 原生插件：让纯文本模型（如 DeepSeek）经 agent/pre-step 劫持 + resolveModelInfo 包装自动识别上传图片（qwen-vl），使纯文本对话也能“看见”图片。
Fold DSH Web execution steps into compact expandable process boxes while keeping dialogue and final answers outside.