Skill Claude CodeCodex
App Builder — generate complete, runnable fullstack WebUI applications (FastAPI backend + pure HTML/CSS/JS frontend) around on-device AI Model Packs (OCR, TTS, ASR, Super-Resolution, etc.). Use this skill when the user wants to build, generate, modify, or debug a standalone WebUI app around any installed Model Pack.
Skill Claude CodeCodex
One sentence: what the model does + what it takes as input + what it produces as output.
Skill Claude CodeCodex
MeloTTS-zh: Chinese TTS from myshell-ai, fully on Snapdragon X Elite NPU (QNN HTP FP16). Four QNN context binaries, all NPU — no CPU vocoder.
Skill Claude CodeCodex
Real-ESRGAN x4plus: image/video super-resolution from xinntao/Real-ESRGAN, running on Snapdragon X Elite NPU (QNN HTP FP16). Single QNN context binary.
Skill Claude CodeCodex
Whisper-base: 74M-param encoder/decoder ASR (OpenAI), on Snapdragon QNN HTP, INT8. Accepts one audio clip (WAV/MP3/FLAC/WEBM, mono 16 kHz, ≤120 s) → structured JSON with detected language, task, full text, and time-aligned segments.
Skill Claude CodeCodex
Streaming RNN-T (k2-fsa / sherpa-onnx) for Mandarin Chinese on Snapdragon QNN HTP with INT8 quantization. Three QNN binaries: encoder.bin (streaming Zipformer encoder), decoder.bin (prediction network), joiner.bin (combines encoder embedding × decoder state → token distribution). Accepts one audio clip (WAV/WEBM, mono…
Skill Claude CodeCodex
A professional programming assistant supporting code writing, debugging, review, refactoring, and performance optimization, as well as open-source repository analysis.
Skill Claude CodeCodex
QAI ModelBuilder. Tools and workflows for model conversion, inspection, operator patching, quantization, and inference validation of self-converted models on Qualcomm platform. Use this skill when working with custom ONNX/PyTorch models — export to ONNX, convert to QNN/SNPE DLC, FP16/FP32/INT8 quantization, operator…
Skill Claude CodeCodex
Diagnose and fix failures in the ONNX → C++/bin → DLL → context-binary (.bin) chain on WoS ARM64: graph-name mismatches, missing VS ARM64 environment, missing HTP runtime files, architecture mismatches, and 0-byte/bins/ output traps. Root causes are usually environment/config — check those first before touching the…
Skill Claude CodeCodex
Fix WoS ARM64 environment: VS ARM64/MSBuild failures, two-Python split (x64 3.10 conversion vs ARM64 3.13 inference), missing deps, protobuf conflicts, PyTorch --index-url rules. All paths from qairtenv.json (generated by Setup.bat) — never hardcode.
Skill Claude CodeCodex
Fix failures that occur while exporting a PyTorch model to ONNX (before QNN conversion): the basicsr / functionaltensor import error, and conversion failures caused by training-only branches (auxiliary classifiers) that emit dynamic-shape operators QAIRT 2.45 cannot convert (ReshapeOp::calculateShape). Fixes are…
Skill Claude CodeCodex
Diagnose runtime failures and wrong results when running .bin/.dlc/.dll through qaiappbuilder / QNNContext: silent load crashes, stale-artifact low cosine, NCHW/NHWC input mismatch, multi-model buffer collisions, and Linux HTP transport/version mismatch. Inference on NPU MUST go through qaiappbuilder/QNNContext …
Skill Claude CodeCodex
Replace unsupported operators (Einsum, GridSample, ScatterND, Mod, Floor, Ceil, Round) with QNN-compatible equivalents from supported base ops (MatMul, Reshape, Transpose, Concat, Div, Mul, Sub, Where, Add), in-memory only. Patch at whichever stage fails.
Skill Claude CodeCodex
Enforce "Do Not Modify QAIRT SDK Files" (B9), diagnose a damaged qnn-context-binary-generator.exe (0-byte / WinError 193) read-only, and recover the single damaged file from the kept SDK zip / launcher-script backup without a 2 GB reinstall.
Skill Claude CodeCodex
Model Hub — download pre-exported models from Qualcomm AI Hub, run inference on-device, and export them to App Builder as ready-to-import Packs. Use this skill when the user wants to use an AI Hub prebuilt package (.bin / .dlc / ONNX) without doing custom conversion. NOT for self-converted models — use model-builder…
Skill Claude CodeCodex
Generate fully editable, high-quality PPTX with python-pptx based on the user's topic, materials, or business goals; defaults to a general-purpose "dark gilded elegant" visual system, with adaptive layout capabilities for cover, table of contents, section/transition, viewpoint statement, image-text mixed layout…
Skill Claude CodeCodex
Use when working with the browser tool — navigating or inspecting a page, clicking through a flow, filling and submitting a form, acting inside a logged-in session, or screenshotting a rendered page — and especially when a page will not cooperate — a selector matches nothing, content is missing because it renders with…
Skill Claude CodeCodex
Use when working with the computer tool — driving the Windows desktop, clicking or dragging anything on screen, opening or focusing an app, or locating a UI element — and especially when a computer action did not do what was expected — a click or drag missed a target that looked right, a drag turned into a selection…
Skill Claude CodeCodex
163 email tool for sending, reading, searching, and replying to emails via 163.com (NetEase Mail).
Skill Claude CodeCodex
Fetch and summarize academic papers from arxiv.org. Use for arxiv.org paper URLs (e.g. arxiv.org/abs/...). Triggers include 'read/summarize this arxiv paper', 'read paper', 'summarize paper'. Note — arxiv URLs must use this skill, NOT summarizenews.
Skill Claude CodeCodex
Query weather forecasts for any city. Uses wttr.in service, no API key required.
Skill Claude CodeCodex
GenieAPIService technical documentation retrieval. Find guides on platform deployment, model configuration, and API usage. GenieAPIService is an OpenAI-compatible API service that enables running large language models(include LLM & VLM model) locally on Qualcomm(高通) Windows on Snapdragon, Android, and Linux platforms.…
Skill Claude CodeCodex
QAI AppBuilder technical documentation retrieval. Find guides on installation, Python/C++ APIs, and model deployment examples. QAI AppBuilder is a rapid AI application development framework designed to simplify the deployment of QNN models on NPU (HTP) across Qualcomm(高通) Windows on Snapdragon, Android, and Linux…
Skill Claude CodeCodex
AIPC, AI Porting Conversion. Tools and workflows for QAIRT/AIPC project setup, model conversion, inspection, operator patching, quantization, context-binary generation, and inference on Qualcomm platforms. Use this skill when creating or initializing an AIPC project, exporting AI models to ONNX, converting ONNX models…