horizonrobotics
- 71 skills
- 0 followers
- 6 hours ago last updated
- ▌ Workflow · horizonrobotics bundleHigh-level workflow skill for long-horizon task decomposition, dependency management, monitoring, and replanning in OpenClaw / HoloAgent robot systems. **Use this skill when:** - The user gives a multi-step task - The task spans navigation, manipulation, monitoring, or recovery - The task involves one or more robots - The task requires dependency ordering or replanning - The user asks for long-horizon execution logic or orchestration
- ▌ Arm Skill · horizonrobotics bundleSkill for triggering predefined robot arm actions through the official OpenClaw control interfaces. **Use this skill when:** - The user explicitly requests an arm action - The task requires a predefined arm skill such as wave, grasp, or reset - The user asks how to trigger arm actions through HTTP or ROS topics - The task involves manipulation but only through documented arm skill interfaces
- ▌ Robot Service · horizonrobotics bundleCanonical skill for OpenClaw robot service interfaces, covering robot_bridge on each robot and multi_robot_ctl on the control center. **Use this skill when:** - The user wants to trigger robot actions through HTTP services - The user asks about robot_bridge or multi_robot_ctl - The task involves remote navigation, arm control, or semantic navigation - The task requires multi-robot fan-out control - The user needs the official external control surface instead of internal FIFO details
- ▌ Sem Nav Skill · horizonrobotics bundleSkill for semantic navigation requests using floor, room, and object level targets. **Use this skill when:** - The user asks the robot to go to a room, landmark, or object - The task refers to semantic targets instead of waypoint IDs - The user asks about `/api/semantic_nav` or `/chat_loc_pub`
- ▌ Rel Move Skill · horizonrobotics bundleSkill for relative robot movement commands through the documented OpenClaw interfaces. **Use this skill when:** - The user asks the robot to move forward, backward, left, or right relatively - The task requires a short relative adjustment instead of named-point navigation - The user asks about `/api/relative_nav` or `/relative_nav`
- ▌ Codereview Family · horizonrobotics bundleStructured code review family for explicit heavy review requests and architecture-sensitive review/evaluation requests. Route to PR/MR review, local changeset review, or architecture review based on the review target and intent. Do not use for routine self-checks or casual bug scans.
- ▌ Feature Dev · horizonrobotics bundleGuided feature development with codebase understanding and architecture focus. Use when implementing a non-trivial feature that benefits from structured discovery, exploration, architecture comparison, implementation, and review.
- ▌ Code Explorer · horizonroboticsDeeply analyzes existing codebase features by tracing execution paths, mapping architecture layers, understanding patterns and abstractions, and documenting dependencies to inform new development.
- ▌ Code Reviewer · horizonroboticsReviews code for bugs, logic errors, security vulnerabilities, code quality issues, and adherence to project conventions, using confidence-based filtering to report only high-priority issues that truly matter.
- ▌ Prmr Codereview · horizonrobotics bundleReview a GitHub PR or GitLab MR when the task is a PR/MR review or comment flow. Use this skill for PR/MR-specific transport, metadata, and comment workflows.
- ▌ Code Architect · horizonroboticsDesigns feature architectures by analyzing existing codebase patterns and conventions, then providing comprehensive implementation blueprints with specific files to create or modify, component designs, data flows, and build sequences.
- ▌ Architecture Review · horizonrobotics bundleReview local code, diffs, modules, directories, or design changes for consequential architecture issues when the user explicitly asks for architecture review, when an upstream review routes architecture-sensitive scope here, or when the current review/evaluation request is primarily architectural.
- ▌ Changeset Codereview · horizonrobotics bundleReview a commit, branch diff, staged diff, working tree diff, patch, or file set for validated high-signal bugs and scoped guidance violations in a local changeset review.
- ▌ Ruff Format · horizonroboticsFormat HoloMotion Python files with the Ruff binary selected by train.env. Use when the user asks to format Python code, run Ruff formatting, check formatting, format changed files, or explicitly format the whole project.
- ▌ Diagnose Holomotion · horizonroboticsDiagnose HoloMotion errors across environment setup, motion conversion, retargeting, training, evaluation, export, simulation, offline deployment, and live teleoperation. Use when a command fails, hangs, produces invalid motion or policy output, differs between stages, or has an unknown HoloMotion root cause.
- ▌ Prepare Motion Data · horizonroboticsConvert a supported or custom human-motion source into HoloSMPL and HoloRetarget data for HoloMotion training. Use when users bring their own BVH, SMPL/SMPL-X, device export, video-derived motion, or another motion format and need conversion, validation, retargeting, or a new source adapter.
- ▌ Train Motion Policy · horizonroboticsTrain, fine-tune, resume, evaluate, and export a HoloMotion motion policy. Use for dataset selection, configuration preflight, checkpoint lineage, distributed training, evaluation coverage, export parity, and preparation for simulation or deployment.
- ▌ Deploy Offline Motion · horizonroboticsDeploy and troubleshoot a HoloMotion policy for offline real-robot motion tracking. Use for deployment preflight, no-action validation, model or motion substitution, controller startup, and offline robot-runtime diagnosis.
- ▌ Isaaclab Source Lookup · horizonroboticsLook up Isaac Lab APIs and implementations in the source tree associated with the project's active training environment. Use for source-level questions about Isaac Lab packages without importing simulator modules.
- ▌ Run Live Teleoperation · horizonroboticsSet up, run, validate, or troubleshoot a supported HoloMotion live-teleoperation pipeline. Use for source readiness, reference processing, streaming, timing, visualization, policy integration, controller transitions, and real-robot teleoperation problems.
- ▌ Benchmark Holomotion Latency · horizonrobotics bundleBenchmark HoloMotion inference with deployment-equivalent artifacts, runtime state, preprocessing, I/O, and hardware. Use for latency, throughput, cache, refinement, export, quantization, target-runtime, or framework-versus-deployment comparisons.
- ▌ Holomotion Train Interpreter · horizonroboticsRun HoloMotion Python scripts or inspect packages with the project training environment. Use for training, evaluation, motion conversion, project Python commands, dependency checks, or any task that must use the environment selected by train.env.
- ▌ Sim Runner · horizonroboticsRun SAPIEN-based simulation rendering from EmbodiedGen layout outputs using sim-cli. Use this skill whenever users ask to load a generated layout.json into simulation, render interactive scene videos, control camera/render settings, or enable robot grasp trajectory rendering.
- ▌ Room Creator · horizonroboticsGenerate indoor rooms (single room or house) and export URDF/USD by wrapping embodied_gen/scripts/room_gen/gen_room.py. Use when users ask to create rooms with seed control, choose room type and complexity, run generation/export stages, or run reproducible room generation jobs (batch runs can be done by wrapping this command in an outer loop/script).
- ▌ Asset Creator · horizonroboticsCreate 3D assets with EmbodiedGen using img3d-cli, text3d-cli, and texture-cli. Use this skill whenever users ask to generate assets from images/text, texture existing meshes, run retry/seed controlled generation, or choose the proper asset-generation CLI from mixed requirements.
- ▌ Asset Process · horizonrobotics bundleScale and rotate complete URDF-based 3D assets, including OBJ, GLB, collision meshes, Gaussian splats, and height metadata. Use this skill whenever a user asks to resize, rotate, reorient, or otherwise transform an EmbodiedGen asset while keeping its related files synchronized.
- ▌ Layout Creator · horizonroboticsGenerate interactive 3D layouts from task descriptions with EmbodiedGen using layout-cli. Use this skill whenever users ask to build task-driven 3D scenes, batch-generate layouts from task files, tune layout generation retries/seeds, or produce simulator-ready layout outputs from background scene lists.
- ▌ Asset Converter · horizonroboticsConvert EmbodiedGen URDF assets to simulator-specific formats (USD/MJCF/URDF) using embodied_gen.data.asset_converter APIs. Use this skill whenever users ask to export assets for IsaacSim, MuJoCo, Genesis, IsaacGym, PyBullet, or SAPIEN, batch-convert URDF assets, or choose the correct converter/source_type per simulator.
- ▌ Asset Retrieval · horizonrobotics bundleRetrieve existing EmbodiedGen assets from a configurable dataset index CSV by natural-language descriptions and return matching `.urdf` paths. Use when users describe an asset they want, ask to find one or several existing URDF assets, or need a fast lookup over the local asset index instead of scanning folders manually.
- ▌ Spatial Computing · horizonrobotics bundleVisualizes floorplans from URDF scene files and inserts/removes 3D assets with collision-aware placement on surfaces. Supports semantic instance matching via LLM (e.g., "put lamp on bookshelf", "delete sofa in living room"). Use when working with URDF/USD indoor scenes, floorplan visualization, object placement/deletion, or room-level scene editing.
- ▌ Scene3d Deployment · horizonroboticsDeploy and troubleshoot the EmbodiedGen scene3d pipeline. Use when bash install.sh scene3d, tiny-cuda-nn, fused-ssim, txt2panoimg, or scene3d-cli fails on RTX 4090/5090.
- ▌ Background Creator · horizonroboticsGenerate background 3D scenes with EmbodiedGen using scene3d-cli. Use this skill whenever users ask to create room/indoor background scenes from text prompts, pre-generate backgrounds for layout-cli, or control scene3d generation quality/runtime with retry, seed, and gs3d settings.
- ▌ Affordance Deployment · horizonroboticsDeploy and troubleshoot the EmbodiedGen affordance pipeline. Use when bash install.sh affordance, pointnet2_ops, torch-scatter, or flash-attn fails on RTX 4090/5090, or when affordance-cli cannot import.
- ▌ Embodiedgen Deployment · horizonrobotics bundleDiagnose EmbodiedGen installation failures on shared machines and RTX 4090/5090 systems. Use for broken conda environments, user-site package leakage, missing build tools, CUDA/PyTorch architecture mismatches, or xformers failures.
- ▌ Ucp · horizonrobotics bundle地平线 J6 UCP 技能入口，通过路由机制将用户意图分发至对应子 skill，涵盖模型性能评测、HBM 推理部署、推理代码生成、Perfetto Trace 采集与分析等能力。
- ▌ Hmct Workflow · horizonrobotics bundleHMCT 模型转换与精度调优总入口。根据用户意图自动路由： (1) 提供了校准数据 → 调用 reference/run_build.py build 执行完整量化构建； (2) 未提供校准数据 → 调用 reference/run_build.py check 使用随机数据快速验证转换流程； (3) 用户希望进行精度调优 → 转交 j6-hmct-cosine-similarity-tuning SKILL 执行多阶段调优； (4) 用户希望进行单项精度 debug 分析（节点灵敏度、数据分布、累积误差等）→ 调用 hmct-debugger CLI 执行对应分析工具。 当用户提示词中出现 HMCT、模型转换、模型量化、PTQ、精度调优、cosine similarity、节点灵敏度、数据分布、累积误差、debug 等关键词时应触发此 Skill。
- ▌ Horizon Router · horizonrobotics bundleOpenExplorer 工具链入口 Skill，处理 PTQ/QAT 量化编译、板端部署、性能精度评估等请求，并将任务路由到对应的 Horizon 子 Skill。
- ▌
- ▌ J6 Hbdk Compile · horizonrobotics bundle通用模型编译 skill。基于 YAML 配置文件驱动，支持 ONNX 和 BC 模型格式。支持从 PTQ config 自动提取输入预处理参数。工作流：生成配置 -> 默认必须经用户确认 -> 再执行编译与报告。Use when the user mentions 模型编译、compile、导出 hbm、生成上板产物、pyramid 输入、resizer 输入。
- ▌ J6 Board Monitor · horizonrobotics bundleJ6 开发板资源监控与推理期间资源采集。当用户需要监控 BPU 占用率、DDR 带宽、内存使用时触发。支持三种场景：(1) Scenario A：CV 模型在指定帧率（如 10Hz）推理期间同步采集 BPU/DDR/内存数据；(2) Scenario B：独立监控板端硬件资源（无推理负载）；(3) Scenario C：LLM 模型循环推理期间同步采集资源数据，使用 simple_demo_request 保持模型持续运行。关键词：BPU 监控、DDR 带宽、内存使用、资源监控、设定帧率推理、LLM 推理监控、hrt_ucp_monitor、hrut_ddr、simple_demo_request、板端资源评估。注意：不要用 hbm_infer/gRPC 做高频推理监控（通信开销太大）。
- ▌ J6 Ucp Hbm Infer · horizonrobotics bundleGenerate X86-side Python client code for hbm_infer, the Python SDK that connects to BPU boards via gRPC to deploy and run HBM model inference. Use this skill when the user wants to write Python code using hbm_infer client APIs — including HbmRpcSession (standard/flexible mode), HTensor for transmission optimization, output_config for output filtering, and related utilities like get_input_info, get_output_info, get_profile, compress_option, core_id. Triggers on writing or debugging Python inference scripts, optimizing data transfer between X86 and board, or any question about the hbm_infer Python client API.
- ▌ J6 Plugin Export · horizonrobotics bundle使用 horizon_plugin_pytorch 的 hbdk4.export 将 QAT 模型导出为 HBIR IR 模块。以独立导出脚本的形式执行，不在训练或评测脚本中添加导出逻辑。
- ▌ J6 Ucp Model Perf Eval · horizonrobotics bundle自动化 hrt_model_exec perf 板端性能评测。触发条件：模型性能测试、benchmark、性能评估、板端测试、thread_num/core_id 参数扫描、吞吐量/延迟对比、远程部署 hrt_model_exec 运行 perf。
- ▌ Horizon Tc UI · horizonrobotics bundleOpenExplorer 工具链 CLI 包的集成 Skill。适用于处理 hb_compile、hb_model_info、 hb_verifier、hb_analyzer、hb_config_generator、hb_eval_preprocess 等工具相关问题， 包括 YAML 配置编写、ONNX/Caffe 模型编译、PTQ 量化、HBIR/HBM 产物生成、 模型信息查看、精度验证、性能分析、评估数据预处理和板端部署验证。 当用户提到 OpenExplorer、horizon_tc_ui、工具链、编译模型、生成 YAML、量化、 calibration、HBIR、HBM、march、fast-perf、check 模式、板端推理、 定点/浮点 BC 模型判断等场景时，应优先使用本 Skill。
- ▌ J6 Plugin Adaptation · horizonrobotics bundle为用户的浮点 PyTorch 模型适配地平线机器人公司的 QAT 工具 `horizon_plugin_pytorch`。这是一个编排型 skill，必须按固定顺序依次调用：`j6-plugin-set-march` → `j6-plugin-insert-quant-dequant` → `j6-plugin-dynamic-block` → `j6-plugin-prepare` → `j6-plugin-set-fake-quantize`。
- ▌ J6 Plugin Graph Diff · horizonrobotics bundle对比两份 FX Graph 计算图文件，精确定位计算图结构和算子参数差异，并根据 diff 报告在原始模型代码中找到对应的差异位置。
- ▌ J6 Ucp Infer Generating · horizonrobotics bundle提供基于 UCP 的模型推理 C++ 代码生成能力。当用户询问 UCP/DNN 推理接口怎么用、某个 API 的参数含义、如何加载模型、如何分配 tensor 内存、如何提交推理任务、Cache 同步怎么做，或要求只生成推理流程中某个模块的代码时使用。
- ▌ Board Detection · horizonrobotics板卡硬件平台检测 Skill。当任务涉及板端运行、板端推理、远端 HBM、性能压测或 BPU 实测，且 .horizon/.env.board 不存在或不完整时触发。自动完成板卡 IP 获取、SSH 架构探测、平台信息写入。
- ▌ J6 Plugin Hbdk Generating · horizonrobotics bundle为基础网络结构生成从量化到编译的完整流程代码（set_march → Quant/DeQuant → 量化配置 → prepare → 校准 → QAT → export → convert → remove_io_op → statistics → compile HBM）。当用户需要同时覆盖量化和编译多个步骤时触发，如"帮我写量化编译全流程代码"、"Conv+BN+ReLU 量化部署"、"地平线量化编译"、"基础结构量化到 HBM"。即使用户没有明确说"全流程"，只要涉及从量化到编译的多个步骤都应触发。如果用户只需要量化或只需要编译，应路由到对应子 skill。关键词："量化编译"、"量化部署"、"全流程"、"set_march 到 HBM"、"地平线量化"、"Horizon 量化编译"、"基础结构量化"、"QAT 量化编译"。
- ▌ Oe Package Install · horizonrobotics bundleOE 包本地安装 Skill。在 oe-package-detection 完成后触发，询问用户是否本地安装。若用户同意，则检测本地 Python / CUDA / PyTorch 版本，匹配 OE 包内 whl，创建 venv 全量安装，更新 .env 为 EXECUTION_MODE=local。
- ▌ J6 Plugin Precision Tuning · horizonrobotics bundle当用户遇到 Horizon Plugin PyTorch 精度调优问题时使用。本 skill 聚焦 PyTorch 侧精度调优，不处理 export、convert、compile 或板端一致性问题。
- ▌ J6 Ucp Perfetto Trace Catcher · horizonrobotics bundleCapture UCP (Horizon Robotics inference SDK) Perfetto traces from a J6 development board. Use this skill whenever the user wants to grab UCP traces, capture Perfetto traces on a Horizon dev board, debug UCP scheduling issues with trace data, or pull .pftrace files from a remote board. Also trigger when the user mentions UCP performance analysis, UCP trace, hrt_model_exec tracing, or BPU trace capture on J6/S1000 platforms. Do NOT use for DSP trace or Chrome trace.
- ▌ J6 Plugin Consistency Debug · horizonrobotics bundle当用户遇到 Horizon Plugin PyTorch 训练部署一致性问题（QAT 模型正常但 BC/HBM 掉点、export/convert/compile 阶段精度偏差）时使用。本 skill 引导用户按 qat.pt -> qat.export.pt -> qat.bc -> quantized.bc -> hbm 分段定位问题，并在每个阶段帮助适配工具、分析工具产出物。
- ▌ J6 Ucp Perfetto Trace Analysis · horizonrobotics bundleUse this skill whenever the user wants to quickly locate, diagnose, or analyze likely performance bottlenecks in a UCP model inference Perfetto trace. Trigger on requests about UCP inference traces, `.pftrace` files, inference latency, slow inference, pipeline stalls, CPU/BPU gaps, OpInfer delays, dispatch or response delay, low effective occupancy, critical-path investigation, or trace-based performance triage in the UCP inference path. This skill is specialized for UCP / UCP-like inference-chain performance analysis and fast bottleneck localization, not generic Perfetto profiling unrelated to UCP inference.
- ▌ Oe Package Detection · horizonroboticsOE 包环境检测 Skill。当任务涉及量化、编译、部署等工具链操作，且 .horizon/.env.oe-package 不存在时触发。自动完成 OE 包路径定位、版本采集、本地环境匹配检查、GPU/CPU Docker 判定，并将结果写入 .env 文件。
- ▌ J6 Plugin Model Check Result · horizonrobotics bundle在 horizon_plugin_pytorch 量化配置检查中，读取并分析 prepare 产出的 model_check_result.txt，定位模型结构与 qconfig 问题（算子融合、共享模块、qconfig 异常、averaging_constant 非默认值、算子输入浮点检查）。
- ▌ J6 Hmct Cosine Similarity Tuning · horizonrobotics bundleHMCT PTQ 精度调优工作流。自动按 INT8→INT16→dual-int16→FP16 逐级验证上限， 结合节点敏感度分析渐进回退，找到满足 Cosine Similarity 阈值（默认 >=0.99）的 最小混精度 quant_config，产出达标配置与调优报告。 触发关键词：HMCT 精度调优、PTQ 调优、量化精度、cosine similarity 不达标、混合精度。
- ▌ Hb Analyzer Performance · horizonrobotics bundle使用 hb_analyzer 分析 J5/J6 模型性能。当用户提到模型文件（.onnx/.bc/.hbm）并想了解性能、延时、带宽、BPU利用率、瓶颈、实测、对比时触发。常见场景：评测模型、检查性能、找瓶颈、板端测试、分析 perf JSON。关键词："性能"、"延时"、"带宽"、"BPU"、"瓶颈"、"太慢"、"实测"、"对比"、"评测"、"hb_analyzer"。即使用户没明确说"分析"，只要提到模型+性能相关词就应该触发。
- ▌ Oe LLM Package Install · horizonrobotics bundleOE-LLM 包本地安装 Skill。在 oe-llm-package-detection 完成后触发，询问用户是否本地安装。若用户同意，则检测本地 Python / CUDA / PyTorch 版本，匹配 OE-LLM 包内 whl，创建 venv 全量安装，更新 .env 为 EXECUTION_MODE=local。
- ▌ Oe LLM Package Detection · horizonroboticsOE-LLM 包环境检测 Skill。当任务涉及 LLM 量化、LLM 压缩、LLM 编译、板端 LLM 推理等 LLM 工具链操作，且 .horizon/.env.oe-llm-package 不存在时触发。自动完成 OE-LLM 包路径定位、版本采集、本地环境匹配检查、GPU Docker 判定，并将结果写入 .env 文件。
- ▌ Llmcompression Add Model · horizonrobotics bundle为 llm_compression 框架新增 LLM/VLM 模型支持。当用户需要在 llm_compression/models/ 中接入新模型时触发。
- ▌ Lightcompress Quant Explore · horizonrobotics bundle执行 LightCompress 量化实验并生成精度报告。当用户要求运行量化实验、量化精度测试或 PPL 评估时触发。
- ▌ Llmcompression Operations · horizonroboticsllm_compression 日常操作 Skill。覆盖校准(calib.sh)、GPU精度评测(torch_eval.sh)、HBM编译(compile.sh)、板端精度评测(hbm_rpc_eval.sh)、量化分析(quant_analysis.sh)的标准用法。触发关键词：校准、calib、torch_eval、compile、hbm_rpc_eval、quant_analysis、LLM编译、LLM量化评测、板端推理评测。注意：llm_compression 是独立于 OE 的工具包，不走 OE/OE-LLM 包检查流程。新增模型支持请路由到 llmcompression-add-model。
- ▌ J6 Plugin Prepare · horizonrobotics bundle在适配 horizon_plugin_pytorch 的过程中对浮点模型执行 prepare（仅添加 prepare 调用；qconfig_setter 固定为全部双 int8 模板；不包含 dynamic_block 相关修改）。
- ▌
- ▌
- ▌ J6 Plugin Dynamic Block · horizonrobotics bundle在适配 horizon_plugin_pytorch（prepare/JIT_STRIP）时，自动为动态控制流/动态循环中涉及算子替换或融合的逻辑添加 dynamic_block 标注，避免 scope/替换错乱与 forward 报错。
- ▌ J6 Plugin Set Fake Quantize · horizonrobotics bundle在适配 horizon_plugin_pytorch 的量化流程中，为模型设置 fake quantize 状态（QAT/CALIBRATION/VALIDATION）。只添加/调用 set_fake_quantize，不做其他修改。
- ▌ J6 Hbdk Export Compile · horizonrobotics为量化后的模型生成导出编译流程代码（export QAT BC → convert → remove_io_op → statistics 检查 CPU 算子 → compile HBM）。务必在用户提到导出编译、HBM 编译、模型部署编译、QAT BC 导出、remove_io_op、CPU 算子检查、模型编译部署、HBM 生成时触发此 skill，即使用户只提到其中一个步骤，只要涉及从 QAT BC 到 HBM 的任何环节都应触发。
- ▌ J6 Plugin Quantization · horizonrobotics bundle为基础网络结构生成量化流程代码（set_march → 插入 Quant/DeQuant → 配置量化参数 → prepare → 校准 → QAT 训练）。务必在用户提到模型量化、量化流程、QAT 校准、Horizon 量化适配、HistogramObserver/MinMaxObserver 配置、量化参数配置、校准训练、QuantStub 插入时触发此 skill，即使用户只问其中一个步骤，只要涉及 Horizon 量化流程的任何环节都应触发。
- ▌ J6 Plugin Insert Quant Dequant · horizonrobotics bundle为浮点 PyTorch 模型插入部署边界 QuantStub/DeQuantStub（horizon_plugin_pytorch）。满足：quant/dequant 是部署边界；每个输入/输出各自独立 stub；QuantStub 初始化不设置 scale。