Updated 3mo ago
by HorizonRobotics
OE-Skills
Structured skill packs for AI agents (Claude Code, Codex, Cursor) to operate the Horizon AI toolchain — HBDK compilation, HMCT/Plugin quantization, UCP on-board inference, and LLM compression.
24 GitHub stars (24)
Steady at 24 stars — no new stars in the 16 days since Sep 30.
- 30 skills (30 skills found in its SKILL.md files)
- 4 forks (4 forks)
- 3mo ago last update (Last updated 3mo ago)
Install
Skills CLI
npx skills add HorizonRobotics/OE-Skills
Asks which skills and agents to install. Just one: npx skills add HorizonRobotics/OE-Skills --skill hmct-workflow — add -g for every project.
Claude Code
git clone https://github.com/HorizonRobotics/OE-Skillscp -r "OE-Skills/horizon/skills/hmct" ~/.claude/skills/
Copies the hmct-workflow skill into ~/.claude/skills/ for every project (use .claude/skills/ for one project). Repeat for the other skills listed below.
Plugin
/plugin marketplace add HorizonRobotics/OE-Skills
Run inside Claude Code if the repository ships a plugin marketplace, then pick plugins from /plugin.
Skills can run code on your machine — read the SKILL.md first.
What it does
In the author's own words, from each skill's SKILL.md — showing the first 30 of 37 files.
-
hmct-workflow
HMCT 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/skills/hmct/SKILL.md on GitHub (opens in a new tab) -
horizon-router
OpenExplorer 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。
horizon/skills/horizon-router/SKILL.md on GitHub (opens in a new tab) -
ucp
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。
horizon/skills/ucp/SKILL.md on GitHub (opens in a new tab) -
skills
HBDK4模型编译工具的使用指南索引,按任务场景组织
horizon/skills/hbdk/hbdk-manual/SKILL.md on GitHub (opens in a new tab) -
j6-hbdk-compile
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。
horizon/skills/hbdk/j6-hbdk-compile/SKILL.md on GitHub (opens in a new tab) -
j6-hmct-cosine-similarity-tuning
HMCT 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。
horizon/skills/hmct/j6-hmct-cosine-similarity-tuning/SKILL.md on GitHub (opens in a new tab) -
board-detection
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。
horizon/skills/horizon-router/board-detection/SKILL.md on GitHub (opens in a new tab) -
oe-llm-package-detection
OE-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。
horizon/skills/horizon-router/oe-llm-package-detection/SKILL.md on GitHub (opens in a new tab) -
oe-llm-package-install
OE-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。
horizon/skills/horizon-router/oe-llm-package-install/SKILL.md on GitHub (opens in a new tab) -
oe-package-detection
OE 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。
horizon/skills/horizon-router/oe-package-detection/SKILL.md on GitHub (opens in a new tab) -
oe-package-install
OE 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。
horizon/skills/horizon-router/oe-package-install/SKILL.md on GitHub (opens in a new tab) -
hb-analyzer-performance
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。
horizon/skills/horizon_tc_ui/hb-analyzer-performance/SKILL.md on GitHub (opens in a new tab) -
horizon-tc-ui
OpenExplorer 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。
horizon/skills/horizon_tc_ui/horizon-tc-ui/SKILL.md on GitHub (opens in a new tab) -
j6-plugin-adaptation
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`。
horizon/skills/plugin/j6-plugin-adaptation/SKILL.md on GitHub (opens in a new tab) -
j6-plugin-consistency-debug
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。
horizon/skills/plugin/j6-plugin-consistency-debug/SKILL.md on GitHub (opens in a new tab) -
j6-plugin-export
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。
horizon/skills/plugin/j6-plugin-export/SKILL.md on GitHub (opens in a new tab) -
j6-plugin-graph-diff
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。
horizon/skills/plugin/j6-plugin-graph-diff/SKILL.md on GitHub (opens in a new tab) -
j6-plugin-hbdk-generating
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"。
horizon/skills/plugin/j6-plugin-hbdk-generating/SKILL.md on GitHub (opens in a new tab) -
j6-plugin-model-check-result
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)。
horizon/skills/plugin/j6-plugin-model-check-result/SKILL.md on GitHub (opens in a new tab) -
j6-plugin-precision-tuning
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。
horizon/skills/plugin/j6-plugin-precision-tuning/SKILL.md on GitHub (opens in a new tab) -
j6-board-monitor
J6 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)。
horizon/skills/ucp/j6-board-monitor/SKILL.md on GitHub (opens in a new tab) -
j6-ucp-hbm-infer
Generate 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, ge…
horizon/skills/ucp/j6-ucp-hbm-infer/SKILL.md on GitHub (opens in a new tab) -
j6-ucp-infer-generating
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。
horizon/skills/ucp/j6-ucp-infer-generating/SKILL.md on GitHub (opens in a new tab) -
j6-ucp-model-perf-eval
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/skills/ucp/j6-ucp-model-perf-eval/SKILL.md on GitHub (opens in a new tab) -
j6-ucp-perfetto-trace-analysis
Use 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 per…
horizon/skills/ucp/j6-ucp-perfetto-trace-analysis/SKILL.md on GitHub (opens in a new tab) -
j6-ucp-perfetto-trace-catcher
Capture 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 cap…
horizon/skills/ucp/j6-ucp-perfetto-trace-catcher/SKILL.md on GitHub (opens in a new tab) -
lightcompress-batch-quantize
批量执行 LightCompress 量化实验并生成对比表。
horizon/skills/llm/lightcompress/lightcompress-batch-quantize/SKILL.md on GitHub (opens in a new tab) -
lightcompress-quant-explore
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。
horizon/skills/llm/lightcompress/lightcompress-quant-explore/SKILL.md on GitHub (opens in a new tab) -
llmcompression-add-model
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。
horizon/skills/llm/llm_compression/llmcompression-add-model/SKILL.md on GitHub (opens in a new tab) -
llmcompression-operations
llm_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。
horizon/skills/llm/llm_compression/llmcompression-operations/SKILL.md on GitHub (opens in a new tab)
No skill matches that filter.
Review before you install
A skill is a set of instructions and scripts your agent will follow, and it can run code on your machine. Read the SKILL.md and any scripts first — this listing is not a security review.