[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6PHFTKTYFD8N2MCB404FEXM":3,"skill-same:01M6PHFTKTYFD8N2MCB404FEXM":110},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":37,"linkOnly":38,"markdownContent":39,"markdownBody":40,"frontMatter":41,"refName":47,"commitSha":48,"skillDirectory":49,"forksCount":14,"repositoryHtmlUrl":50,"repositoryHomepageUrl":6,"ownerHtmlUrl":51,"ownerBlogUrl":52,"ownerDescription":6,"updatedAt":53,"install":54,"resources":67},"01M6PHFTKTYFD8N2MCB404FEXM","nv-reason-cxr",null,"Used for command-shape or live NV-Reason-CXR chest X-ray reasoning smoke tests. Not for diagnosis or clinical reporting.","headmktlucca-ui\u002Fneuroads-lp","headmktlucca-ui","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F247379838?v=4","01M6PDVAV6M1W96D1SR1PDF8CS","NeuroAds","01M6PDVAZ4P281GNDKD45CYTF5",0,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Fnv-reason-cxr\u002FSKILL.md",[],[19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36],"nv reason cxr","reason","shape","for","X-ray","diagnosis","clinical","tests","live","smoke tests","smoke","command","Bash","team","Ray","NVIDIA","reporting","reasoning","2026-11-10 10:12:30",false,"---\nname: nv-reason-cxr\ndescription: Used for command-shape or live NV-Reason-CXR chest X-ray reasoning smoke tests. Not for diagnosis or clinical reporting.\nlicense: Apache-2.0\nallowed-tools: Bash\nmetadata:\n  author: NVIDIA MedTech Team\n  tags:\n    - MedTech\n    - CXR\n    - reasoning\n---\n\n# NV-Reason-CXR\n\n## Purpose\n- Used for command-shape or live NV-Reason-CXR chest X-ray reasoning smoke tests. Not for diagnosis or clinical reporting.\n- Use the wrapper exactly as documented; do not replace the upstream entrypoint with a handwritten implementation.\n- Manifest I\u002FO: inputs are `chest_xray_image_or_fixture`; outputs are `result_json`.\n\n## Instructions\n- Read `skill_manifest.yaml` before changing arguments, side effects, or validation gates.\n- Run `scripts\u002Frun_nv_reason_cxr.py` through the documented command below; pass `--out-dir` only for generated fixtures or harness-managed artifact directories.\n- If a host agent exposes `run_script`, use `run_script(\"scripts\u002Frun_nv_reason_cxr.py\", args=[...])`; otherwise run the Bash\u002FPython command shown below.\n- Check the emitted JSON and paired verifier guidance before treating the run as evidence.\n- When reporting a completed run, return the full wrapper JSON or at minimum\n  the complete `output.response_text` exactly as emitted, including any\n  model-generated `\u003Cthink>...\u003C\u002Fthink>` and `\u003Canswer>...\u003C\u002Fanswer>` sections. Do\n  not collapse the result to labels unless the user explicitly asks for a\n  summary.\n\n## Available Scripts\n| Script | Purpose | Arguments |\n|---|---|---|\n| `scripts\u002Frun_nv_reason_cxr.py` | Primary entrypoint declared by skill_manifest.yaml. | `PATH_TO_CXR_OR_FIXTURE [--out-dir OUT_DIR] [--backend local\\|hf-space-api] [--mock] [--check-setup]` |\n\n## Prerequisites\n- Local backend requirements: GPU\u002FCUDA when declared by the manifest; Python packages listed in `runtime.side_effects.pip_packages`.\n- API backend requirements: public network access to the [Hugging Face Space](https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Fnvidia\u002Fnv-reason-cxr); no local PyTorch, Transformers, CUDA, model cache, or Hugging Face token.\n- Side effects: emits result JSON on stdout; may write generated fixture artifacts under the caller's `--out-dir`; may cache model assets under `~\u002F.cache\u002Fhuggingface\u002F` for local inference; and may contact `https:\u002F\u002Fhuggingface.co`, `https:\u002F\u002Fgithub.com`, or `https:\u002F\u002F*.hf.space` outside `--mock` mode.\n- Run commands from the repository root unless an existing section below says otherwise.\n\n## Limitations\n- This is a thin wrapper. Image preprocessing, model inference, and decoding are delegated to Hugging Face Transformers and the NV-Reason-CXR-3B model.\n- Output is not a diagnosis, clinical report, treatment recommendation, or triage decision. It is engineering evidence and must be reviewed by a qualified professional before any medical use.\n- The model may hallucinate findings, miss subtle abnormalities, misread support devices, or produce overconfident prose.\n- The committed fixture uses a generated synthetic PNG and deterministic mock response so CI can verify wrapper behavior without downloading model weights. Mock mode is not a substitute for model inference.\n- The `hf-space-api` backend depends on public Hugging Face Space availability and API compatibility.\n- Not for clinical deployment, clinical interpretation, autonomous diagnosis, treatment decisions.\n\n## Troubleshooting\n| Error | Cause | Fix |\n|---|---|---|\n| Missing dependency or import error | Runtime package drift from `skill_manifest.yaml`. | Install the packages declared in the manifest or use the documented setup command. |\n| CUDA unavailable from an agent but available in a user terminal | The agent sandbox, container, or job wrapper may not expose NVIDIA device nodes even when the same Python environment has CUDA-capable PyTorch installed. | Compare `python -c \"import torch; print(torch.cuda.is_available())\"` and `nvidia-smi` inside the agent context and in the user terminal. If only the agent context fails, rerun with GPU\u002Fdevice access, use the host terminal, or pass `--device cpu --allow-cpu` only for an explicit slow CPU test. |\n| API backend HTTP or schema error | The public Hugging Face Space may be unavailable, rate limited, or changed. | Re-run later or use `--backend local` when local dependencies and CUDA are available. |\n| Empty or schema-invalid output | Wrong input path, unsupported modality, or upstream failure. | Re-run with a known fixture and inspect the wrapper JSON plus stderr. |\n| Validation gate failure | Output violated a declared engineering invariant. | Keep the failed evidence pack and use the gate message to repair inputs or wrapper code. |\n\nRuns NVIDIA-Medtech [`NV-Reason-CXR-3B`](https:\u002F\u002Fgithub.com\u002FNVIDIA-Medtech\u002FNV-Reason-CXR)\nfor chest X-ray image interpretation through either the documented local\nHugging Face Transformers inference path or the public Hugging Face Space API.\nThe wrapper does not reimplement the model, image preprocessing, or decoding.\n\n\n## Exact Runnable Surface\n\nFor command-shape smoke tests and JSON fixtures, use this repo-root wrapper path exactly:\n\n```bash\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py PATH_TO_CXR_OR_FIXTURE --mock --out-dir OUT_DIR\n```\n\nFor local live image inference, omit `--mock` only when the user asks for live\nmodel inference. Local is the default backend:\n\n```bash\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py PATH_TO_CXR_OR_FIXTURE \\\n  --prompt \"Find abnormalities and support devices.\" \\\n  --backend local\n```\n\nFor public API inference without local model packages, use:\n\n```bash\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py PATH_TO_CXR_OR_FIXTURE \\\n  --prompt \"Find abnormalities and support devices.\" \\\n  --backend hf-space-api\n```\n\nDo not invent `Medical AI Skills run`, `eval_engine\u002Frun.py`, `infer.py`, or\n`python -m nv_reason_cxr` commands for ordinary user runs.\n\n## Preconditions\n\nFor `--backend local`, install the inference dependencies in the environment\nthat will run the skill:\n\n```bash\npip install torch==2.7.1 torchvision==0.22.1 transformers==4.56.1 Pillow\n```\n\nThe model weights are loaded from `nvidia\u002FNV-Reason-CXR-3B` through\nTransformers. They may download to the Hugging Face cache on first use.\nSet `TRANSFORMERS_OFFLINE=1` or pass `--local-files-only` only after the\nweights are already cached.\n\nCUDA is expected for practical inference. CPU execution may work for small\ntests but is slow and must be requested explicitly.\n\nFor `--backend hf-space-api`, no local PyTorch, Transformers, CUDA, model\ncache, or Hugging Face token is required. The backend sends the image and\nprompt to the public `nvidia\u002Fnv-reason-cxr` Hugging Face Space.\n\nCheck the local environment before downloading weights or running inference:\n\n```bash\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py --check-setup\n```\n\nThe setup report checks importable dependencies, CUDA visibility, Hugging Face\ncache state, and the recommended next step.\n\nOperational environment variables:\n\n| Variable | When to use |\n|---|---|\n| `MOCK_NV_REASON_CXR` | Set to `1` for deterministic command-shape smoke tests without model inference. |\n| `NV_REASON_CXR_MODEL` | Override the Hugging Face model id only for compatibility probes. |\n| `HF_HOME` | Point at a pre-populated Hugging Face cache. |\n| `HF_TOKEN` | Optional for local model downloads only when required by the local environment; not needed for the public API backend. |\n| `TRANSFORMERS_OFFLINE` | Set to `1` only after weights are already cached. |\n| `HF_HUB_OFFLINE` | Set to `1` only after Hugging Face assets are already cached. |\n\n## Prompt Routing\n\nChoose both the model prompt and the user-facing output mode before running\nthe wrapper. Routing order matters: exact model-prompt requests use\npass-through\u002Fraw-only mode first; otherwise report-generation requests take\nprecedence over general analysis and specific-question routing.\n\nUse pass-through\u002Fraw-only mode only when the user explicitly asks to send an\nexact prompt to the model, such as \"call the model with this prompt exactly:\n...\". Pass only that exact model prompt as `--prompt`.\n\nUse abnormality-analysis mode when the user asks to analyze, examine, or find\nabnormalities in a chest X-ray. Treat local image paths, uploaded filenames,\nbackend choices such as \"use API\" or \"use local\", output delivery instructions,\nand other agent orchestration text as wrapper instructions, not model prompt\ncontent. Do not include local filesystem paths, backend names, or \"use API\" in\n`--prompt` unless the user explicitly asks to send that exact text to the\nmodel. For ordinary abnormality-finding requests, use the documented prompt,\nusually `--prompt \"Find abnormalities and support devices.\"`, with the\nrequested backend.\n\nUse report-generation\u002Ftwo-call mode if the user asks to write, create, or\ngenerate a structured report, chest X-ray report, radiology report, or report.\nIf sufficient raw model context for the same image is already available,\nespecially output from `Find abnormalities and support devices.`, skip the context-gathering\ncall. Otherwise first run the wrapper with `--prompt \"Examine the chest\nX-ray.\"` to gather context, but do not show that first call. Then run the\nwrapper again with a multi-turn transcript prompt:\n\n```text\nUser: Find abnormalities and support devices.\n\nAssistant:\n\u003Craw model context>\n\nUser: Write a structured report.\n```\n\nTreat the second call as the completed run.\n\nUse default-prompt\u002Fcontext-answer mode when the user asks a specific question\nabout a finding, such as presence, count, location, or characterization, or\nmixes general analysis with specific questions. Run the wrapper with\n`--prompt \"Find abnormalities and support devices.\"` before answering the original question\nin plain text prefixed exactly with `Answer:`. Base the answer only on the raw\nmodel output context and the image.\n\n## Follow-up Handling\n\nFor follow-up questions about an image already analyzed in the conversation,\nreuse prior raw model context when it is sufficient. For report follow-ups, use\nreport-generation\u002Ftwo-call mode and skip directly to the second model call if\nthere is sufficient context. If prior context is insufficient and the same\nimage path or image bytes are available, call the wrapper again using the\nprompt routing rules above. If the image is no longer available, ask the user\nto reattach it.\n\nFor long multi-turn prompts that include prior raw model output, prefer a\nquoted Bash here-doc variable so XML-like tags, apostrophes, quotes, and\nnewlines are preserved:\n\n```bash\nIFS= read -r -d '' prompt \u003C\u003C'PROMPT'\nUser: Examine the chest X-ray.\n\nAssistant:\n\u003Craw model context>\n\nUser: Write a structured report.\nPROMPT\n\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py PATH_TO_CXR.png \\\n  --prompt \"$prompt\" \\\n  --backend hf-space-api\n```\n\nUse `IFS= read -r -d '' prompt \u003C\u003C'PROMPT'`, not command substitution, for long\npasted transcripts.\n\n## License\n\nThe upstream repository code is Apache-2.0. The model weights are released\nunder the NVIDIA OneWay Noncommercial License Agreement. Users are responsible\nfor complying with the model-weight terms before live inference.\n\n## Usage\n\nFrom Medical AI Skills repo root:\n\n```bash\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py PATH_TO_CXR.png \\\n  --prompt \"Find abnormalities and support devices.\" \\\n  --backend local\n```\n\nFor public API inference without installing model packages locally:\n\n```bash\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py PATH_TO_CXR.png \\\n  --prompt \"Find abnormalities and support devices.\" \\\n  --backend hf-space-api\n```\n\nFor user requests that include local path or backend instructions, keep those\ninstructions out of the model prompt:\n\n```text\nUser request: find abnormalities in ~\u002FDesktop\u002F363.jpg (use API)\n```\n\n```bash\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py ~\u002FDesktop\u002F363.jpg \\\n  --prompt \"Find abnormalities and support devices.\" \\\n  --backend hf-space-api\n```\n\nUse the wrapper script directly for agent-generated commands. Do not replace\nit with `eval_engine\u002Frun.py` unless the user explicitly asks to run the eval\nharness. Do not redirect stdout with `>` in generated commands: callers and\nthe eval harness read the wrapper's stdout JSON to verify the run. The direct\nrunnable surface is:\n\n```bash\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py PATH_TO_CXR_OR_FIXTURE \\\n  --mock \\\n  --out-dir runs\u002Fnv_reason_cxr_case\n```\n\n`PATH_TO_CXR_OR_FIXTURE` may be a PNG\u002FJPEG image or a JSON fixture. If the\nuser provides a JSON request such as\n`runs\u002F...\u002Fsynthetic_cxr_input.json`, pass that exact JSON path as the first\nargument. The script will load `generated:\u002F\u002Fsynthetic_chest_xray` fixtures,\ncreate the temporary PNG under the output directory, and emit JSON with the\nmodel response. Use `--mock` only for command-shape smoke tests or fixtures\nthat request mock mode; omit `--mock` for live model inference.\n\nFor JPEG input:\n\n```bash\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py PATH_TO_CXR.jpg \\\n  --prompt \"Describe the chest X-ray findings.\" \\\n  --backend local\n```\n\nFlags:\n\n- `--backend local|hf-space-api` — inference backend, default `local`.\n- `--model-id` — Hugging Face model id, default `nvidia\u002FNV-Reason-CXR-3B`.\n- `--device auto|cuda|cpu` — default `auto`, using CUDA when available.\n- `--allow-cpu` — required for live CPU inference; CPU runs can be very slow.\n- `--torch-dtype auto|float16|bfloat16|float32` — default `auto`, using\n  bfloat16 on CUDA and float32 on CPU, matching the published BF16 model.\n- `--max-new-tokens` — generation cap, default 2048.\n- `--local-files-only` — use only locally cached Hugging Face assets.\n- `--mock` — deterministic dry-run response for CI and wiring checks.\n- `--prompt-preset findings|comprehensive|educational|structured` — optional\n  known-good prompt presets from the model card\u002Fdemo behavior.\n- `--out-dir` — optional artifact directory. Required for generated JSON\n  fixtures; the eval harness passes it explicitly.\n\nThe tested local live path uses:\n\n- `AutoModelForImageTextToText.from_pretrained(..., dtype=torch.bfloat16).eval().to(\"cuda\")`\n- `AutoProcessor.from_pretrained(..., use_fast=True)`\n- PNG\u002FJPEG image input plus one text prompt\n- `max_new_tokens=2048` by default\n\nThe script emits JSON on stdout and writes no clinical report files. Direct\nPNG\u002FJPEG runs do not create a default output directory. Generated JSON fixtures\nrequire `--out-dir` for the temporary synthetic image. The result JSON records\ninput image metadata, prompt, model id, runtime mode, response text, and known\nlimitations. If `runtime.truncated_by_max_new_tokens` is `true`, rerun with a\nhigher `--max-new-tokens` value.\n\nReporting reminder: for both `local` and `hf-space-api` backends, follow the\ncompleted-run rule in Instructions.\n\nThe `hf-space-api` backend calls the fixed public Hugging Face Space at\n`https:\u002F\u002Fnvidia-nv-reason-cxr.hf.space` with a 300 second HTTP timeout.\n\n## Fixture Smoke Test\n\nThe committed fixture uses a generated synthetic PNG and mock mode so the\neval harness can verify the wrapper without downloading weights:\n\n```bash\npython eval_engine\u002Frun.py skills\u002Fnv-reason-cxr \\\n  --fixture skills\u002Fnv-reason-cxr\u002Ffixtures\u002Fsynthetic_cxr_input.json \\\n  --out runs\u002Fnv_reason_cxr_smoke\n```\n\n## Limits\n\nThis is research and engineering tooling only. It is not validated for\nclinical diagnosis, treatment decisions, triage, patient-facing reporting, or\nregulatory use. Model outputs can hallucinate, miss subtle findings, or\noverstate uncertainty. A qualified professional must review any use in a\nmedical workflow.\n","# NV-Reason-CXR\n\n## Purpose\n- Used for command-shape or live NV-Reason-CXR chest X-ray reasoning smoke tests. Not for diagnosis or clinical reporting.\n- Use the wrapper exactly as documented; do not replace the upstream entrypoint with a handwritten implementation.\n- Manifest I\u002FO: inputs are `chest_xray_image_or_fixture`; outputs are `result_json`.\n\n## Instructions\n- Read `skill_manifest.yaml` before changing arguments, side effects, or validation gates.\n- Run `scripts\u002Frun_nv_reason_cxr.py` through the documented command below; pass `--out-dir` only for generated fixtures or harness-managed artifact directories.\n- If a host agent exposes `run_script`, use `run_script(\"scripts\u002Frun_nv_reason_cxr.py\", args=[...])`; otherwise run the Bash\u002FPython command shown below.\n- Check the emitted JSON and paired verifier guidance before treating the run as evidence.\n- When reporting a completed run, return the full wrapper JSON or at minimum\n  the complete `output.response_text` exactly as emitted, including any\n  model-generated `\u003Cthink>...\u003C\u002Fthink>` and `\u003Canswer>...\u003C\u002Fanswer>` sections. Do\n  not collapse the result to labels unless the user explicitly asks for a\n  summary.\n\n## Available Scripts\n| Script | Purpose | Arguments |\n|---|---|---|\n| `scripts\u002Frun_nv_reason_cxr.py` | Primary entrypoint declared by skill_manifest.yaml. | `PATH_TO_CXR_OR_FIXTURE [--out-dir OUT_DIR] [--backend local\\|hf-space-api] [--mock] [--check-setup]` |\n\n## Prerequisites\n- Local backend requirements: GPU\u002FCUDA when declared by the manifest; Python packages listed in `runtime.side_effects.pip_packages`.\n- API backend requirements: public network access to the [Hugging Face Space](https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Fnvidia\u002Fnv-reason-cxr); no local PyTorch, Transformers, CUDA, model cache, or Hugging Face token.\n- Side effects: emits result JSON on stdout; may write generated fixture artifacts under the caller's `--out-dir`; may cache model assets under `~\u002F.cache\u002Fhuggingface\u002F` for local inference; and may contact `https:\u002F\u002Fhuggingface.co`, `https:\u002F\u002Fgithub.com`, or `https:\u002F\u002F*.hf.space` outside `--mock` mode.\n- Run commands from the repository root unless an existing section below says otherwise.\n\n## Limitations\n- This is a thin wrapper. Image preprocessing, model inference, and decoding are delegated to Hugging Face Transformers and the NV-Reason-CXR-3B model.\n- Output is not a diagnosis, clinical report, treatment recommendation, or triage decision. It is engineering evidence and must be reviewed by a qualified professional before any medical use.\n- The model may hallucinate findings, miss subtle abnormalities, misread support devices, or produce overconfident prose.\n- The committed fixture uses a generated synthetic PNG and deterministic mock response so CI can verify wrapper behavior without downloading model weights. Mock mode is not a substitute for model inference.\n- The `hf-space-api` backend depends on public Hugging Face Space availability and API compatibility.\n- Not for clinical deployment, clinical interpretation, autonomous diagnosis, treatment decisions.\n\n## Troubleshooting\n| Error | Cause | Fix |\n|---|---|---|\n| Missing dependency or import error | Runtime package drift from `skill_manifest.yaml`. | Install the packages declared in the manifest or use the documented setup command. |\n| CUDA unavailable from an agent but available in a user terminal | The agent sandbox, container, or job wrapper may not expose NVIDIA device nodes even when the same Python environment has CUDA-capable PyTorch installed. | Compare `python -c \"import torch; print(torch.cuda.is_available())\"` and `nvidia-smi` inside the agent context and in the user terminal. If only the agent context fails, rerun with GPU\u002Fdevice access, use the host terminal, or pass `--device cpu --allow-cpu` only for an explicit slow CPU test. |\n| API backend HTTP or schema error | The public Hugging Face Space may be unavailable, rate limited, or changed. | Re-run later or use `--backend local` when local dependencies and CUDA are available. |\n| Empty or schema-invalid output | Wrong input path, unsupported modality, or upstream failure. | Re-run with a known fixture and inspect the wrapper JSON plus stderr. |\n| Validation gate failure | Output violated a declared engineering invariant. | Keep the failed evidence pack and use the gate message to repair inputs or wrapper code. |\n\nRuns NVIDIA-Medtech [`NV-Reason-CXR-3B`](https:\u002F\u002Fgithub.com\u002FNVIDIA-Medtech\u002FNV-Reason-CXR)\nfor chest X-ray image interpretation through either the documented local\nHugging Face Transformers inference path or the public Hugging Face Space API.\nThe wrapper does not reimplement the model, image preprocessing, or decoding.\n\n\n## Exact Runnable Surface\n\nFor command-shape smoke tests and JSON fixtures, use this repo-root wrapper path exactly:\n\n```bash\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py PATH_TO_CXR_OR_FIXTURE --mock --out-dir OUT_DIR\n```\n\nFor local live image inference, omit `--mock` only when the user asks for live\nmodel inference. Local is the default backend:\n\n```bash\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py PATH_TO_CXR_OR_FIXTURE \\\n  --prompt \"Find abnormalities and support devices.\" \\\n  --backend local\n```\n\nFor public API inference without local model packages, use:\n\n```bash\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py PATH_TO_CXR_OR_FIXTURE \\\n  --prompt \"Find abnormalities and support devices.\" \\\n  --backend hf-space-api\n```\n\nDo not invent `Medical AI Skills run`, `eval_engine\u002Frun.py`, `infer.py`, or\n`python -m nv_reason_cxr` commands for ordinary user runs.\n\n## Preconditions\n\nFor `--backend local`, install the inference dependencies in the environment\nthat will run the skill:\n\n```bash\npip install torch==2.7.1 torchvision==0.22.1 transformers==4.56.1 Pillow\n```\n\nThe model weights are loaded from `nvidia\u002FNV-Reason-CXR-3B` through\nTransformers. They may download to the Hugging Face cache on first use.\nSet `TRANSFORMERS_OFFLINE=1` or pass `--local-files-only` only after the\nweights are already cached.\n\nCUDA is expected for practical inference. CPU execution may work for small\ntests but is slow and must be requested explicitly.\n\nFor `--backend hf-space-api`, no local PyTorch, Transformers, CUDA, model\ncache, or Hugging Face token is required. The backend sends the image and\nprompt to the public `nvidia\u002Fnv-reason-cxr` Hugging Face Space.\n\nCheck the local environment before downloading weights or running inference:\n\n```bash\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py --check-setup\n```\n\nThe setup report checks importable dependencies, CUDA visibility, Hugging Face\ncache state, and the recommended next step.\n\nOperational environment variables:\n\n| Variable | When to use |\n|---|---|\n| `MOCK_NV_REASON_CXR` | Set to `1` for deterministic command-shape smoke tests without model inference. |\n| `NV_REASON_CXR_MODEL` | Override the Hugging Face model id only for compatibility probes. |\n| `HF_HOME` | Point at a pre-populated Hugging Face cache. |\n| `HF_TOKEN` | Optional for local model downloads only when required by the local environment; not needed for the public API backend. |\n| `TRANSFORMERS_OFFLINE` | Set to `1` only after weights are already cached. |\n| `HF_HUB_OFFLINE` | Set to `1` only after Hugging Face assets are already cached. |\n\n## Prompt Routing\n\nChoose both the model prompt and the user-facing output mode before running\nthe wrapper. Routing order matters: exact model-prompt requests use\npass-through\u002Fraw-only mode first; otherwise report-generation requests take\nprecedence over general analysis and specific-question routing.\n\nUse pass-through\u002Fraw-only mode only when the user explicitly asks to send an\nexact prompt to the model, such as \"call the model with this prompt exactly:\n...\". Pass only that exact model prompt as `--prompt`.\n\nUse abnormality-analysis mode when the user asks to analyze, examine, or find\nabnormalities in a chest X-ray. Treat local image paths, uploaded filenames,\nbackend choices such as \"use API\" or \"use local\", output delivery instructions,\nand other agent orchestration text as wrapper instructions, not model prompt\ncontent. Do not include local filesystem paths, backend names, or \"use API\" in\n`--prompt` unless the user explicitly asks to send that exact text to the\nmodel. For ordinary abnormality-finding requests, use the documented prompt,\nusually `--prompt \"Find abnormalities and support devices.\"`, with the\nrequested backend.\n\nUse report-generation\u002Ftwo-call mode if the user asks to write, create, or\ngenerate a structured report, chest X-ray report, radiology report, or report.\nIf sufficient raw model context for the same image is already available,\nespecially output from `Find abnormalities and support devices.`, skip the context-gathering\ncall. Otherwise first run the wrapper with `--prompt \"Examine the chest\nX-ray.\"` to gather context, but do not show that first call. Then run the\nwrapper again with a multi-turn transcript prompt:\n\n```text\nUser: Find abnormalities and support devices.\n\nAssistant:\n\u003Craw model context>\n\nUser: Write a structured report.\n```\n\nTreat the second call as the completed run.\n\nUse default-prompt\u002Fcontext-answer mode when the user asks a specific question\nabout a finding, such as presence, count, location, or characterization, or\nmixes general analysis with specific questions. Run the wrapper with\n`--prompt \"Find abnormalities and support devices.\"` before answering the original question\nin plain text prefixed exactly with `Answer:`. Base the answer only on the raw\nmodel output context and the image.\n\n## Follow-up Handling\n\nFor follow-up questions about an image already analyzed in the conversation,\nreuse prior raw model context when it is sufficient. For report follow-ups, use\nreport-generation\u002Ftwo-call mode and skip directly to the second model call if\nthere is sufficient context. If prior context is insufficient and the same\nimage path or image bytes are available, call the wrapper again using the\nprompt routing rules above. If the image is no longer available, ask the user\nto reattach it.\n\nFor long multi-turn prompts that include prior raw model output, prefer a\nquoted Bash here-doc variable so XML-like tags, apostrophes, quotes, and\nnewlines are preserved:\n\n```bash\nIFS= read -r -d '' prompt \u003C\u003C'PROMPT'\nUser: Examine the chest X-ray.\n\nAssistant:\n\u003Craw model context>\n\nUser: Write a structured report.\nPROMPT\n\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py PATH_TO_CXR.png \\\n  --prompt \"$prompt\" \\\n  --backend hf-space-api\n```\n\nUse `IFS= read -r -d '' prompt \u003C\u003C'PROMPT'`, not command substitution, for long\npasted transcripts.\n\n## License\n\nThe upstream repository code is Apache-2.0. The model weights are released\nunder the NVIDIA OneWay Noncommercial License Agreement. Users are responsible\nfor complying with the model-weight terms before live inference.\n\n## Usage\n\nFrom Medical AI Skills repo root:\n\n```bash\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py PATH_TO_CXR.png \\\n  --prompt \"Find abnormalities and support devices.\" \\\n  --backend local\n```\n\nFor public API inference without installing model packages locally:\n\n```bash\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py PATH_TO_CXR.png \\\n  --prompt \"Find abnormalities and support devices.\" \\\n  --backend hf-space-api\n```\n\nFor user requests that include local path or backend instructions, keep those\ninstructions out of the model prompt:\n\n```text\nUser request: find abnormalities in ~\u002FDesktop\u002F363.jpg (use API)\n```\n\n```bash\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py ~\u002FDesktop\u002F363.jpg \\\n  --prompt \"Find abnormalities and support devices.\" \\\n  --backend hf-space-api\n```\n\nUse the wrapper script directly for agent-generated commands. Do not replace\nit with `eval_engine\u002Frun.py` unless the user explicitly asks to run the eval\nharness. Do not redirect stdout with `>` in generated commands: callers and\nthe eval harness read the wrapper's stdout JSON to verify the run. The direct\nrunnable surface is:\n\n```bash\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py PATH_TO_CXR_OR_FIXTURE \\\n  --mock \\\n  --out-dir runs\u002Fnv_reason_cxr_case\n```\n\n`PATH_TO_CXR_OR_FIXTURE` may be a PNG\u002FJPEG image or a JSON fixture. If the\nuser provides a JSON request such as\n`runs\u002F...\u002Fsynthetic_cxr_input.json`, pass that exact JSON path as the first\nargument. The script will load `generated:\u002F\u002Fsynthetic_chest_xray` fixtures,\ncreate the temporary PNG under the output directory, and emit JSON with the\nmodel response. Use `--mock` only for command-shape smoke tests or fixtures\nthat request mock mode; omit `--mock` for live model inference.\n\nFor JPEG input:\n\n```bash\npython skills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py PATH_TO_CXR.jpg \\\n  --prompt \"Describe the chest X-ray findings.\" \\\n  --backend local\n```\n\nFlags:\n\n- `--backend local|hf-space-api` — inference backend, default `local`.\n- `--model-id` — Hugging Face model id, default `nvidia\u002FNV-Reason-CXR-3B`.\n- `--device auto|cuda|cpu` — default `auto`, using CUDA when available.\n- `--allow-cpu` — required for live CPU inference; CPU runs can be very slow.\n- `--torch-dtype auto|float16|bfloat16|float32` — default `auto`, using\n  bfloat16 on CUDA and float32 on CPU, matching the published BF16 model.\n- `--max-new-tokens` — generation cap, default 2048.\n- `--local-files-only` — use only locally cached Hugging Face assets.\n- `--mock` — deterministic dry-run response for CI and wiring checks.\n- `--prompt-preset findings|comprehensive|educational|structured` — optional\n  known-good prompt presets from the model card\u002Fdemo behavior.\n- `--out-dir` — optional artifact directory. Required for generated JSON\n  fixtures; the eval harness passes it explicitly.\n\nThe tested local live path uses:\n\n- `AutoModelForImageTextToText.from_pretrained(..., dtype=torch.bfloat16).eval().to(\"cuda\")`\n- `AutoProcessor.from_pretrained(..., use_fast=True)`\n- PNG\u002FJPEG image input plus one text prompt\n- `max_new_tokens=2048` by default\n\nThe script emits JSON on stdout and writes no clinical report files. Direct\nPNG\u002FJPEG runs do not create a default output directory. Generated JSON fixtures\nrequire `--out-dir` for the temporary synthetic image. The result JSON records\ninput image metadata, prompt, model id, runtime mode, response text, and known\nlimitations. If `runtime.truncated_by_max_new_tokens` is `true`, rerun with a\nhigher `--max-new-tokens` value.\n\nReporting reminder: for both `local` and `hf-space-api` backends, follow the\ncompleted-run rule in Instructions.\n\nThe `hf-space-api` backend calls the fixed public Hugging Face Space at\n`https:\u002F\u002Fnvidia-nv-reason-cxr.hf.space` with a 300 second HTTP timeout.\n\n## Fixture Smoke Test\n\nThe committed fixture uses a generated synthetic PNG and mock mode so the\neval harness can verify the wrapper without downloading weights:\n\n```bash\npython eval_engine\u002Frun.py skills\u002Fnv-reason-cxr \\\n  --fixture skills\u002Fnv-reason-cxr\u002Ffixtures\u002Fsynthetic_cxr_input.json \\\n  --out runs\u002Fnv_reason_cxr_smoke\n```\n\n## Limits\n\nThis is research and engineering tooling only. It is not validated for\nclinical diagnosis, treatment decisions, triage, patient-facing reporting, or\nregulatory use. Model outputs can hallucinate, miss subtle findings, or\noverstate uncertainty. A qualified professional must review any use in a\nmedical workflow.\n",{"name":5,"license":15,"metadata":42,"description":7,"allowed-tools":31},{"tags":43,"author":46},[44,45,36],"MedTech","CXR","NVIDIA MedTech Team","main","7a940b8e5af16c95e5c190a524421d32fc7445a9",".agent\u002Fskills\u002Fnv-reason-cxr","https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp","https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui","https:\u002F\u002Fneuroads.com.br","2026-08-12 00:42:44",{"source":8,"skillName":5,"commands":55,"prompt":65,"zipAvailable":66},[56,59,62],{"tool":57,"command":58},"npx","npx skills add headmktlucca-ui\u002Fneuroads-lp --skill nv-reason-cxr",{"tool":60,"command":61},"bunx","bunx skills add headmktlucca-ui\u002Fneuroads-lp --skill nv-reason-cxr",{"tool":63,"command":64},"pnpm","pnpm dlx skills add headmktlucca-ui\u002Fneuroads-lp --skill nv-reason-cxr","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Fnv-reason-cxr\u002FSKILL.md",true,[68,74,78,82,88,92,96,102,106],{"path":69,"resourceType":70,"resourceTypeDesc":71,"sizeBytes":72,"sourceHtmlUrl":73},"BENCHMARK.md","REFERENCE","参考文档",3980,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Fnv-reason-cxr\u002FBENCHMARK.md",{"path":75,"resourceType":70,"resourceTypeDesc":71,"sizeBytes":76,"sourceHtmlUrl":77},"evals\u002Fevals.json",1269,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Fnv-reason-cxr\u002Fevals\u002Fevals.json",{"path":79,"resourceType":70,"resourceTypeDesc":71,"sizeBytes":80,"sourceHtmlUrl":81},"fixtures\u002Fsynthetic_cxr_input.json",387,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Fnv-reason-cxr\u002Ffixtures\u002Fsynthetic_cxr_input.json",{"path":83,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":86,"sourceHtmlUrl":87},"scripts\u002Frun_nv_reason_cxr.py","SCRIPT","脚本",33314,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Fnv-reason-cxr\u002Fscripts\u002Frun_nv_reason_cxr.py",{"path":89,"resourceType":70,"resourceTypeDesc":71,"sizeBytes":90,"sourceHtmlUrl":91},"skill_manifest.yaml",7379,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Fnv-reason-cxr\u002Fskill_manifest.yaml",{"path":93,"resourceType":70,"resourceTypeDesc":71,"sizeBytes":94,"sourceHtmlUrl":95},"skill-card.md",3813,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Fnv-reason-cxr\u002Fskill-card.md",{"path":97,"resourceType":98,"resourceTypeDesc":99,"sizeBytes":100,"sourceHtmlUrl":101},"skill.oms.sig","OTHER","其他",5785,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Fnv-reason-cxr\u002Fskill.oms.sig",{"path":103,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":104,"sourceHtmlUrl":105},"tests\u002Ftest_nv_reason_cxr.py",7721,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Fnv-reason-cxr\u002Ftests\u002Ftest_nv_reason_cxr.py",{"path":107,"resourceType":70,"resourceTypeDesc":71,"sizeBytes":108,"sourceHtmlUrl":109},"validators\u002Foutput_schema.json",2894,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Fnv-reason-cxr\u002Fvalidators\u002Foutput_schema.json",[111,137,161,187,205,240],{"skillNo":112,"name":113,"nameZh":6,"description":114,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":115,"licenseFileUrl":6,"sourceHtmlUrl":116,"categories":117,"keywords":126,"publishedAt":37},"01M6PTDCZTXTGSTX4RFK901YSZ","xunit-skill","Generates xUnit.net tests in C#. Covers Fact\u002FTheory, constructor injection, IClassFixture, and FluentAssertions. Use when user mentions \"xUnit\", \"[Fact]\", \"[Theory]\", \"Assert.Equal\", \"C# xUnit\". Triggers on: \"xUnit\", \"[Fact]\", \"[Theory]\", \"Assert.Equal C#\", \"xUnit.net\".","MIT","https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fxunit-skill\u002FSKILL.md",[118,123],{"categoryNo":119,"categoryName":120,"taxonomyType":121,"taxonomyTypeDesc":122},"01JBSEEDCAT000000000000008","测试与质量","SCENARIO","使用场景",{"categoryNo":124,"categoryName":125,"taxonomyType":121,"taxonomyTypeDesc":122},"01JBSEEDCAT000000000000023","Skill与智能体开发",[127,128,129,130,131,132,26,133,134,135,136],"xunit","xunit skill","testing","injection","user","use","triggers","unit","unit-testing","C#",{"skillNo":138,"name":139,"nameZh":6,"description":140,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":115,"licenseFileUrl":6,"sourceHtmlUrl":141,"categories":142,"keywords":145,"publishedAt":37},"01M6PTCB59ZTG5FTMWGPYEMMBP","xcuitest-skill","Generates XCUITest UI tests for iOS\u002FiPadOS apps in Swift. Apple's native testing framework for reliable, fast UI automation. Supports local simulators and TestMu AI cloud real devices. Use when user mentions \"XCUITest\", \"XCTest\", \"iOS UI test\", \"Swift test\", \"XCUIApplication\". Triggers on: \"XCUITest\", \"XCTest UI\", \"iOS UI test\", \"Swift UI test\", \"XCUIApplication\", \"TestMu\".","https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fxcuitest-skill\u002FSKILL.md",[143,144],{"categoryNo":119,"categoryName":120,"taxonomyType":121,"taxonomyTypeDesc":122},{"categoryNo":124,"categoryName":125,"taxonomyType":121,"taxonomyTypeDesc":122},[146,147,129,148,149,150,151,152,132,153,154,155,156,157,26,133,158,159,22,160],"xcuitest","xcuitest skill","automation","test","testing framework","ios","devices","mobile","mobile-testing","ui test","local","apple","apps","native","objective",{"skillNo":162,"name":163,"nameZh":6,"description":164,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":115,"licenseFileUrl":6,"sourceHtmlUrl":165,"categories":166,"keywords":171,"publishedAt":37},"01M6PTATMNTY9NF4J7KSR76YK6","world-builder","Create and manage generative game zones for Dorothy's Pokemon-style world. Use this skill when creating, updating, or designing game worlds via MCP tools.","https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fworld-builder\u002FSKILL.md",[167,170],{"categoryNo":168,"categoryName":169,"taxonomyType":121,"taxonomyTypeDesc":122},"01JBSEEDCAT000000000000015","视觉设计",{"categoryNo":124,"categoryName":125,"taxonomyType":121,"taxonomyTypeDesc":122},[172,173,174,175,176,177,178,179,132,180,181,182,183,184,185,22,186],"builder","world","world builder","MCP","create","mcp tools","Tools","game","pokemon","manage","style","via","theme","zone","generative",{"skillNo":188,"name":189,"nameZh":6,"description":190,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":115,"licenseFileUrl":6,"sourceHtmlUrl":191,"categories":192,"keywords":195,"publishedAt":37},"01M6PT9Z92JTSQEHE1QR0NNW5F","webdriverio-skill","Generates WebdriverIO (WDIO) automation tests in JavaScript or TypeScript. Supports local and TestMu AI cloud. Use when user mentions \"WebdriverIO\", \"WDIO\", \"wdio.conf\", \"browser.url\", \"$\", \"$$\". Triggers on: \"WebdriverIO\", \"WDIO\", \"wdio\", \"browser.$\".","https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fwebdriverio-skill\u002FSKILL.md",[193,194],{"categoryNo":119,"categoryName":120,"taxonomyType":121,"taxonomyTypeDesc":122},{"categoryNo":124,"categoryName":125,"taxonomyType":121,"taxonomyTypeDesc":122},[196,129,197,148,198,199,200,201,131,132,202,156,26,133,203,204],"webdriverio skill","typescript","javascript","browser","cloud","e2e","e2e-testing","URL","conf",{"skillNo":206,"name":207,"nameZh":6,"description":208,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":115,"licenseFileUrl":6,"sourceHtmlUrl":209,"categories":210,"keywords":220,"publishedAt":37},"01M6PT9NVK36ZWCQ69YXK3FR4K","web-design-guidelines","Web platform design and accessibility guidelines. Use when building web interfaces, auditing accessibility, implementing responsive layouts, or reviewing web UI code. Triggers on tasks involving HTML, CSS, web components, WCAG compliance, responsive design, or web performance.","https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fweb\u002FSKILL.md",[211,214,215],{"categoryNo":212,"categoryName":213,"taxonomyType":121,"taxonomyTypeDesc":122},"01JBSEEDCAT000000000000009","前端开发",{"categoryNo":168,"categoryName":169,"taxonomyType":121,"taxonomyTypeDesc":122},{"categoryNo":216,"categoryName":217,"taxonomyType":218,"taxonomyTypeDesc":219},"01JBSEEDCAT000000000000032","设计师","PROFESSION","适用职业",[221,222,223,224,225,226,227,228,229,230,231,232,233,234,235,236,237,238,133,239],"design","web","guidelines","web design guidelines","web design","css","performance","components","web components","tasks","code","web performance","ui","platform design","html","responsive design","responsive","platform","implementing",{"skillNo":241,"name":242,"nameZh":6,"description":243,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":115,"licenseFileUrl":6,"sourceHtmlUrl":244,"categories":245,"keywords":248,"publishedAt":37},"01M6PT9A3JKRE3X34276F2FW5R","watchos-design-guidelines","Apple Human Interface Guidelines for Apple Watch. Use when building watchOS apps, complications, or workout features. Triggers on tasks involving Watch UI, Digital Crown, glanceable interfaces, or wrist-based interactions.","https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fwatchos\u002FSKILL.md",[246,247],{"categoryNo":168,"categoryName":169,"taxonomyType":121,"taxonomyTypeDesc":122},{"categoryNo":216,"categoryName":217,"taxonomyType":218,"taxonomyTypeDesc":219},[221,223,249,230,233,132,250,251,157,238,252,133,158,253,22,254,255],"watchos design guidelines","workout","human","watch","interface","platform-design","features"]