[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6PK9M4774QZ2DCQF20N1Z2R":3,"skill-same:01M6PK9M4774QZ2DCQF20N1Z2R":141},{"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":23,"publishedAt":44,"linkOnly":45,"markdownContent":46,"markdownBody":47,"frontMatter":48,"refName":58,"commitSha":59,"skillDirectory":60,"forksCount":14,"repositoryHtmlUrl":61,"repositoryHomepageUrl":6,"ownerHtmlUrl":62,"ownerBlogUrl":63,"ownerDescription":6,"updatedAt":64,"install":65,"resources":78},"01M6PK9M4774QZ2DCQF20N1Z2R","tao-analyze-gaps-visual-changenet",null,"Performs gap analysis on NVIDIA TAO Visual ChangeNet (VCN) Classify experiments by invoking the data-services container (`tao_toolkit.data_services` from `versions.yaml`) directly via `docker run … gap_analysis vcn_aoi …` — picks the optimal decision threshold, ranks per-sample weakness, and emits a top-K weakest parquet expanded per-lighting for downstream augmentation. Use when analyzing VCN classification failures, picking SDA augmentation targets, auditing PASS\u002FNO_PASS boundary cases, or running DEFT gap analysis on an AOI ChangeNet model.","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\u002Ftao-analyze-gaps-visual-changenet\u002FSKILL.md",[18],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000011","运维与部署","SCENARIO","使用场景",[24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43],"visual","tao analyze gaps visual changenet","changenet","gaps","tao","analyze","Bash","model","decision","read","image","sample","classify","top","at","bank","analysis","data","analyzing","run","2026-11-10 10:12:30",false,"---\nname: tao-analyze-gaps-visual-changenet\ndescription: Performs gap analysis on NVIDIA TAO Visual ChangeNet (VCN) Classify experiments by invoking the data-services\n  container (`tao_toolkit.data_services` from `versions.yaml`) directly via `docker run … gap_analysis vcn_aoi …` — picks\n  the optimal decision threshold, ranks per-sample weakness, and emits a top-K weakest parquet expanded per-lighting for\n  downstream augmentation. Use when analyzing VCN classification failures, picking SDA augmentation targets, auditing\n  PASS\u002FNO_PASS boundary cases, or running DEFT gap analysis on an AOI ChangeNet model.\nlicense: Apache-2.0\ncompatibility: Requires docker + nvidia-container-toolkit and a CUDA GPU. Pulls the `tao_toolkit.data_services` image declared in `versions.yaml` at the skill bank root.\nmetadata:\n  author: NVIDIA Corporation\n  version: \"0.3.0\"\nallowed-tools: Read Bash\ntags:\n- data\n- rca\n- vcn\n- aoi\n---\n\n# TAO VCN Classify Gap Analysis Skill\n\nYou are an analyst for NVIDIA TAO VCN Classify (Visual Component Net) inference results. Your job is to identify the **weakest samples per ground-truth label** by measuring signed distance from the decision threshold *in the wrong direction*, then surface them for downstream augmentation or relabeling.\n\nThis skill is intentionally lightweight. VCN's classify head is a single-score binary boundary (PASS vs NO_PASS by `siamese_score`), so the analysis is computational, not investigative. The whole computation lives behind one direct `docker run` invocation against the `tao_toolkit.data_services` image declared in `versions.yaml` (resolved at runtime — see Setup). The container's entrypoint takes `\u003Ccategory> \u003Caction> [hydra overrides...]`; we pass `gap_analysis vcn_aoi key=value …`. You do **not** need subagents, multi-phase image audits, or component-type clustering — VCN does not expose those dimensions. View only a small set of representative weak samples to qualify the gaps after the container returns.\n\nThe CLI surface can shift between data-services container builds. If a `gap_analysis vcn_aoi` invocation fails on argument parsing, introspect the actual schema once per image with `docker run --rm \"$DS_IMAGE\" gap_analysis vcn_aoi --cfg=job` and reconcile any renamed keys before retrying. See `references\u002Ftroubleshooting.md` for the key-rename reconciliation and the full pitfalls list. The output parquet name is `kpi_gaps.parquet`.\n\n---\n\n## Inputs\n\n1. **Experiment result directory** — contains `inference\u002Finference.csv` (required columns `input_path`, `object_name`, `label`, `siamese_score`). Pass the **directory** (e.g. `inference\u002Flatest\u002F`), not the CSV file.\n2. **Training code\u002Fconfig directory** — contains the VCN train YAML; the container reads `dataset.classify.input_map` and `dataset.classify.image_ext` for per-lighting expansion.\n3. **Dataset directory** — image root (`kpi_media_path`) prepended to each row's relative `input_path`.\n4. **Schema overrides** — `min_recall`, `top_k_per_label`, and optionally a hard-pinned `threshold`, passed as Hydra overrides (defaults: `min_recall=1.0`, `top_k_per_label=50`, `threshold=-1.0` meaning sweep). **`top_k_per_label` must be a positive integer** — omitting it flips the container into \"below-threshold filter\" mode, which at `min_recall=1.0` returns only PASS misclassifications and zero NO_PASS rows.\n\nSee `references\u002Fparameters-and-artifacts.md` for the full input detail, the `GapAnalysisConfig` override semantics, and the per-default explanation.\n\n---\n\n## Setup\n\nThe threshold sweep, weakness ranking, and per-lighting expansion all run inside the `tao_toolkit.data_services` image declared in `versions.yaml`. Resolve the concrete URI once at the top of the run, then confirm Docker, the NVIDIA container toolkit, and a GPU are present and ensure the image is cached:\n\n```bash\n# Resolve tao_toolkit.data_services → concrete nvcr.io\u002F... URI from versions.yaml\nDS_IMAGE=$(python3 -c \"import yaml,os; print(yaml.safe_load(open(os.environ['TAO_SKILL_BANK_PATH']+'\u002Fversions.yaml'))['images']['tao_toolkit']['data_services'])\")\necho \"DS_IMAGE=$DS_IMAGE\"\n\ndocker info > \u002Fdev\u002Fnull && echo \"OK: docker\"\nnvidia-smi > \u002Fdev\u002Fnull && echo \"OK: GPU\"\ndocker image inspect \"$DS_IMAGE\" > \u002Fdev\u002Fnull \\\n  || docker pull \"$DS_IMAGE\"\n```\n\n`TAO_SKILL_BANK_PATH` is exported by the plugin's `session_start` hook. If it is unset (e.g. running outside the Claude Code plugin), point it at the skill-bank repo root before resolving.\n\nA GPU is required (the same image is used across the AOI loop and other actions assume CUDA is present). Aborting early on a GPU-less host saves a confusing late error.\n\n**Path mounting.** Every host path the container reads or writes — `inference.csv`, the train YAML, the dataset image root, and the output dir — must be bind-mounted. The simplest pattern is to mount the workspace root with **identical paths** inside and outside the container so absolute paths in args resolve the same on both sides:\n\n```bash\nWORKSPACE=\u003Cabsolute path that contains inference.csv, train YAML, dataset images, and the output dir>\nDOCKER=\"docker run --gpus all --rm --ipc=host --user $(id -u):$(id -g) -v $WORKSPACE:$WORKSPACE -w $WORKSPACE $DS_IMAGE\"\n```\n\nIf `inference.csv`, the train YAML, and the dataset images live in different roots, pass multiple `-v` flags — but every absolute path you pass in args must resolve inside the container.\n\n**CLI overrides cover the common case.** `min_recall`, `top_k_per_label`, and optionally `threshold` are passed as Hydra overrides on the command line; defaults baked into the container (`min_recall=1.0`, `top_k_per_label=50`, `threshold=-1.0` to sweep) handle most runs. If the container also accepts a spec file via `-e \u003Cspec>` (verify with `--cfg=job`), passing one is a convenience, not a requirement — override only what you need.\n\n---\n\n## Method\n\nThe whole skill is a single `docker run` invocation followed by a small visual spot-check. The container does Steps 1–4 internally (threshold sweep, weakness scoring, top-K selection, per-lighting expansion). You handle Step 5 (visual spot-check) directly with the Read tool.\n\n### Step 1–4 — Run the container\n\n```bash\n$DOCKER gap_analysis vcn_aoi \\\n    inference_results_dir=\u003Cexp_dir>\u002Finference\u002F\u003Clabel>\u002F \\\n    train_config=\u003Cexp_dir>\u002Ftrain.yaml \\\n    kpi_media_path=\u003Cdataset_root> \\\n    results_dir=\u003Crca_results_dir> \\\n    top_k_per_label=50\n```\n\n> **Always pass `top_k_per_label`.** This is the argument that switches the container\n> from the default \"samples below threshold\" filter into proper top-K-per-label\n> ranking. At `min_recall=1.0` the threshold is by construction at-or-below every\n> NO_PASS score, so the below-threshold filter returns ONLY misclassified PASS rows\n> and zero NO_PASS rows — useless as an augmentation queue. With `top_k_per_label`\n> set to a positive integer (either in the spec or as a Hydra override), the\n> container computes signed weakness against the threshold for every row and\n> surfaces the K weakest **per ground-truth label**, which is the per-label ranked\n> output downstream steps consume.\n\nThe container sweeps every unique `siamese_score` (plus one value just below the minimum), keeps candidates with NO_PASS recall ≥ `min_recall` (tolerance `1e-12`), picks the best-F1 threshold (tie-break: precision, then threshold value), scores signed weakness per row, takes the top `top_k_per_label` per ground-truth label, and expands each into one row per lighting. See `references\u002Fparameters-and-artifacts.md` for the exact computation, the override defaults, and the artifact table.\n\nIf **no** candidate threshold meets the recall target, the container exits non-zero and writes `unreachable_kpi.txt` into `results_dir` explaining which recall the model can actually achieve. In that case, stop the analysis after the docker call, write a one-section report explaining the model fundamentally cannot reach the KPI at any operating point, and recommend retraining or relabeling — skip the visual spot-check.\n\n**Container writes into `results_dir`:** `kpi_gaps.parquet` (top-K weakest per label, expanded per lighting; columns `filepath`, `label`, `siamese_score`, `weakness`), `threshold.txt`, `metrics.json`, `weak_samples_breakdown.txt`, and `unreachable_kpi.txt` (only when the recall target is unreachable). See `references\u002Fparameters-and-artifacts.md` for the per-artifact contents. Print the container's stdout summary (chosen threshold, kept-row counts, per-label breakdown) to your own stdout so the script-check hook can verify the run produced output.\n\n### Step 5 — Visual spot check (small, fixed)\n\nSkip this step if `unreachable_kpi.txt` exists in `results_dir` — there is nothing meaningful to spot-check when the model can't reach the KPI at any threshold.\n\nOtherwise, use the Read tool to **view** the test images for:\n\n- The 5 weakest PASS samples (the top of the \"PASS misclassified as NO_PASS\" pile) — pick by sorting `kpi_gaps.parquet` rows where `label == 'PASS'` by `weakness` descending.\n- The 5 weakest NO_PASS samples (the top of the \"NO_PASS misclassified as PASS\" pile) — same, with `label != 'PASS'`.\n\n`kpi_gaps.parquet` is already expanded per-lighting (multiple rows per sample). For the spot check, deduplicate to one row per (input_path, object_name) — pick the row whose `filepath` uses the FIRST lighting from the train YAML (one image per sample is enough — VCN's classify head sees all lightings stacked, but for human spot-check one is representative).\n\nClassify each viewed sample as exactly one of:\n- **mislabeled** — visual content disagrees with the CSV label\n- **edge case** — genuinely ambiguous boundary case\n- **data quality** — corrupted, dark, wrong crop, bad framing\n- **systematic** — model has learned the wrong feature (the image looks \"obviously PASS\u002FNO_PASS\" but the model disagrees)\n\nCopy each viewed image (resized to 128×128 if PIL is available, otherwise just copy) into `\u003Cresults_dir>\u002Frca_images\u002F` so it can be embedded inline in the report.\n\nThis is the **only** image inspection required. Do not view dozens of images, do not run failure mode clustering, do not audit goldens — VCN does not have golden images.\n\n---\n\n## Reference invocation\n\nThe paste-and-edit end-to-end recipe (workspace, four paths, two numeric knobs, spec-file write, docker run, and the stdout sanity print that surfaces row counts for the script-check hook) lives in `references\u002Frecipe.md`. Use it verbatim, editing only the workspace, paths, and knobs.\n\n---\n\n## Outputs\n\nWrite everything into a timestamped folder under the experiment result directory: `\u003Cexperiment_result_dir>\u002Frca_results\u002FYYYY-MM-DD_HHMMSS\u002F`. The container's outputs (`kpi_gaps.parquet`, `threshold.txt`, `metrics.json`, `weak_samples_breakdown.txt`, and `unreachable_kpi.txt` when applicable) go straight there; the visual spot-check writes `rca_images\u002F`; the packaging hook adds `rca_config\u002F` and `claude_session.jsonl` automatically when `RCA_Report.md` is written. See `references\u002Fparameters-and-artifacts.md` for the full folder tree.\n\nAt the start of the run, get the real timestamp by running `date +%Y-%m-%d_%H%M%S` in Bash. Do NOT hardcode or guess. If the user specifies a custom output path, use that instead but maintain the same internal structure.\n\n---\n\n## Common pitfalls\n\nThe most consequential failure is **forgetting `top_k_per_label` when `min_recall=1.0`** — at that recall the chosen threshold sits at or below every NO_PASS score, so the fallback below-threshold filter matches ONLY misclassified PASS rows and `kpi_gaps.parquet` ends up with zero NO_PASS rows. Always include an explicit positive `top_k_per_label`. The full pitfalls list (spec file outside `$WORKSPACE`, unresolved `???` sentinels, wrong\u002Funpulled image tag, path-mount mismatch, `unreachable_kpi.txt` handling, missing `inference.csv` columns, missing train-YAML keys, `kpi_media_path` prefix mismatch, no GPU inside the container) and the CLI-drift reconciliation are in `references\u002Ftroubleshooting.md`.\n\n---\n\n## Report Structure\n\nWrite the RCA report into the timestamped output folder. It is a 7-section computational gap analysis (Verdict, Threshold Selection, Weakness Distribution, Top-K Weakest Samples, Visual Spot Check, Per-Label Breakdown, Recommended Actions), 1000–1800 words, with the confusion-matrix and per-label tables filled from `metrics.json` and the spot-check rows from `kpi_gaps.parquet`. When `unreachable_kpi.txt` exists, replace sections 3–6 with one short section quoting that file and collapse section 7 to a single retrain-or-relabel recommendation. See `references\u002Frca-report-structure.md` for the complete skeleton with every section heading, table layout, and the unreachable-KPI variant.\n\n---\n\n## Execution Order\n\n1. Resolve `DS_IMAGE` from `versions.yaml` (`images.tao_toolkit.data_services`), then run `docker info`, `nvidia-smi`, and `docker image inspect \"$DS_IMAGE\"` (pulling if missing) once to confirm the environment. Abort with a clear message if any fail.\n2. Run `date +%Y-%m-%d_%H%M%S` to get the timestamp; create `\u003Cexperiment_result_dir>\u002Frca_results\u002F\u003Ctimestamp>\u002F`.\n3. Write `vcn_aoi_spec.yaml` into the timestamped dir with `min_recall` and `top_k_per_label` filled in. Keep it under `$WORKSPACE` so the `-e` path resolves inside the container.\n4. Run `docker run … \"$DS_IMAGE\" gap_analysis vcn_aoi -e vcn_aoi_spec.yaml inference_results_dir=… train_config=… kpi_media_path=… output_dir=…`. The container writes `kpi_gaps.parquet`, `threshold.txt`, `metrics.json`, `weak_samples_breakdown.txt` into `results_dir`. Print the chosen threshold and kept-row counts to stdout so the script-check hook can verify the run produced output.\n5. If `unreachable_kpi.txt` exists, skip Step 6 and write the abridged report. Otherwise continue.\n6. Pick 10 weak samples (5 weakest PASS + 5 weakest NO_PASS) from `kpi_gaps.parquet`, view each test image with Read, classify, and copy each into `rca_images\u002F`.\n7. Write `RCA_Report.md` last — writing it triggers the packaging hook, which copies session logs and skill config alongside.\n","# TAO VCN Classify Gap Analysis Skill\n\nYou are an analyst for NVIDIA TAO VCN Classify (Visual Component Net) inference results. Your job is to identify the **weakest samples per ground-truth label** by measuring signed distance from the decision threshold *in the wrong direction*, then surface them for downstream augmentation or relabeling.\n\nThis skill is intentionally lightweight. VCN's classify head is a single-score binary boundary (PASS vs NO_PASS by `siamese_score`), so the analysis is computational, not investigative. The whole computation lives behind one direct `docker run` invocation against the `tao_toolkit.data_services` image declared in `versions.yaml` (resolved at runtime — see Setup). The container's entrypoint takes `\u003Ccategory> \u003Caction> [hydra overrides...]`; we pass `gap_analysis vcn_aoi key=value …`. You do **not** need subagents, multi-phase image audits, or component-type clustering — VCN does not expose those dimensions. View only a small set of representative weak samples to qualify the gaps after the container returns.\n\nThe CLI surface can shift between data-services container builds. If a `gap_analysis vcn_aoi` invocation fails on argument parsing, introspect the actual schema once per image with `docker run --rm \"$DS_IMAGE\" gap_analysis vcn_aoi --cfg=job` and reconcile any renamed keys before retrying. See `references\u002Ftroubleshooting.md` for the key-rename reconciliation and the full pitfalls list. The output parquet name is `kpi_gaps.parquet`.\n\n---\n\n## Inputs\n\n1. **Experiment result directory** — contains `inference\u002Finference.csv` (required columns `input_path`, `object_name`, `label`, `siamese_score`). Pass the **directory** (e.g. `inference\u002Flatest\u002F`), not the CSV file.\n2. **Training code\u002Fconfig directory** — contains the VCN train YAML; the container reads `dataset.classify.input_map` and `dataset.classify.image_ext` for per-lighting expansion.\n3. **Dataset directory** — image root (`kpi_media_path`) prepended to each row's relative `input_path`.\n4. **Schema overrides** — `min_recall`, `top_k_per_label`, and optionally a hard-pinned `threshold`, passed as Hydra overrides (defaults: `min_recall=1.0`, `top_k_per_label=50`, `threshold=-1.0` meaning sweep). **`top_k_per_label` must be a positive integer** — omitting it flips the container into \"below-threshold filter\" mode, which at `min_recall=1.0` returns only PASS misclassifications and zero NO_PASS rows.\n\nSee `references\u002Fparameters-and-artifacts.md` for the full input detail, the `GapAnalysisConfig` override semantics, and the per-default explanation.\n\n---\n\n## Setup\n\nThe threshold sweep, weakness ranking, and per-lighting expansion all run inside the `tao_toolkit.data_services` image declared in `versions.yaml`. Resolve the concrete URI once at the top of the run, then confirm Docker, the NVIDIA container toolkit, and a GPU are present and ensure the image is cached:\n\n```bash\n# Resolve tao_toolkit.data_services → concrete nvcr.io\u002F... URI from versions.yaml\nDS_IMAGE=$(python3 -c \"import yaml,os; print(yaml.safe_load(open(os.environ['TAO_SKILL_BANK_PATH']+'\u002Fversions.yaml'))['images']['tao_toolkit']['data_services'])\")\necho \"DS_IMAGE=$DS_IMAGE\"\n\ndocker info > \u002Fdev\u002Fnull && echo \"OK: docker\"\nnvidia-smi > \u002Fdev\u002Fnull && echo \"OK: GPU\"\ndocker image inspect \"$DS_IMAGE\" > \u002Fdev\u002Fnull \\\n  || docker pull \"$DS_IMAGE\"\n```\n\n`TAO_SKILL_BANK_PATH` is exported by the plugin's `session_start` hook. If it is unset (e.g. running outside the Claude Code plugin), point it at the skill-bank repo root before resolving.\n\nA GPU is required (the same image is used across the AOI loop and other actions assume CUDA is present). Aborting early on a GPU-less host saves a confusing late error.\n\n**Path mounting.** Every host path the container reads or writes — `inference.csv`, the train YAML, the dataset image root, and the output dir — must be bind-mounted. The simplest pattern is to mount the workspace root with **identical paths** inside and outside the container so absolute paths in args resolve the same on both sides:\n\n```bash\nWORKSPACE=\u003Cabsolute path that contains inference.csv, train YAML, dataset images, and the output dir>\nDOCKER=\"docker run --gpus all --rm --ipc=host --user $(id -u):$(id -g) -v $WORKSPACE:$WORKSPACE -w $WORKSPACE $DS_IMAGE\"\n```\n\nIf `inference.csv`, the train YAML, and the dataset images live in different roots, pass multiple `-v` flags — but every absolute path you pass in args must resolve inside the container.\n\n**CLI overrides cover the common case.** `min_recall`, `top_k_per_label`, and optionally `threshold` are passed as Hydra overrides on the command line; defaults baked into the container (`min_recall=1.0`, `top_k_per_label=50`, `threshold=-1.0` to sweep) handle most runs. If the container also accepts a spec file via `-e \u003Cspec>` (verify with `--cfg=job`), passing one is a convenience, not a requirement — override only what you need.\n\n---\n\n## Method\n\nThe whole skill is a single `docker run` invocation followed by a small visual spot-check. The container does Steps 1–4 internally (threshold sweep, weakness scoring, top-K selection, per-lighting expansion). You handle Step 5 (visual spot-check) directly with the Read tool.\n\n### Step 1–4 — Run the container\n\n```bash\n$DOCKER gap_analysis vcn_aoi \\\n    inference_results_dir=\u003Cexp_dir>\u002Finference\u002F\u003Clabel>\u002F \\\n    train_config=\u003Cexp_dir>\u002Ftrain.yaml \\\n    kpi_media_path=\u003Cdataset_root> \\\n    results_dir=\u003Crca_results_dir> \\\n    top_k_per_label=50\n```\n\n> **Always pass `top_k_per_label`.** This is the argument that switches the container\n> from the default \"samples below threshold\" filter into proper top-K-per-label\n> ranking. At `min_recall=1.0` the threshold is by construction at-or-below every\n> NO_PASS score, so the below-threshold filter returns ONLY misclassified PASS rows\n> and zero NO_PASS rows — useless as an augmentation queue. With `top_k_per_label`\n> set to a positive integer (either in the spec or as a Hydra override), the\n> container computes signed weakness against the threshold for every row and\n> surfaces the K weakest **per ground-truth label**, which is the per-label ranked\n> output downstream steps consume.\n\nThe container sweeps every unique `siamese_score` (plus one value just below the minimum), keeps candidates with NO_PASS recall ≥ `min_recall` (tolerance `1e-12`), picks the best-F1 threshold (tie-break: precision, then threshold value), scores signed weakness per row, takes the top `top_k_per_label` per ground-truth label, and expands each into one row per lighting. See `references\u002Fparameters-and-artifacts.md` for the exact computation, the override defaults, and the artifact table.\n\nIf **no** candidate threshold meets the recall target, the container exits non-zero and writes `unreachable_kpi.txt` into `results_dir` explaining which recall the model can actually achieve. In that case, stop the analysis after the docker call, write a one-section report explaining the model fundamentally cannot reach the KPI at any operating point, and recommend retraining or relabeling — skip the visual spot-check.\n\n**Container writes into `results_dir`:** `kpi_gaps.parquet` (top-K weakest per label, expanded per lighting; columns `filepath`, `label`, `siamese_score`, `weakness`), `threshold.txt`, `metrics.json`, `weak_samples_breakdown.txt`, and `unreachable_kpi.txt` (only when the recall target is unreachable). See `references\u002Fparameters-and-artifacts.md` for the per-artifact contents. Print the container's stdout summary (chosen threshold, kept-row counts, per-label breakdown) to your own stdout so the script-check hook can verify the run produced output.\n\n### Step 5 — Visual spot check (small, fixed)\n\nSkip this step if `unreachable_kpi.txt` exists in `results_dir` — there is nothing meaningful to spot-check when the model can't reach the KPI at any threshold.\n\nOtherwise, use the Read tool to **view** the test images for:\n\n- The 5 weakest PASS samples (the top of the \"PASS misclassified as NO_PASS\" pile) — pick by sorting `kpi_gaps.parquet` rows where `label == 'PASS'` by `weakness` descending.\n- The 5 weakest NO_PASS samples (the top of the \"NO_PASS misclassified as PASS\" pile) — same, with `label != 'PASS'`.\n\n`kpi_gaps.parquet` is already expanded per-lighting (multiple rows per sample). For the spot check, deduplicate to one row per (input_path, object_name) — pick the row whose `filepath` uses the FIRST lighting from the train YAML (one image per sample is enough — VCN's classify head sees all lightings stacked, but for human spot-check one is representative).\n\nClassify each viewed sample as exactly one of:\n- **mislabeled** — visual content disagrees with the CSV label\n- **edge case** — genuinely ambiguous boundary case\n- **data quality** — corrupted, dark, wrong crop, bad framing\n- **systematic** — model has learned the wrong feature (the image looks \"obviously PASS\u002FNO_PASS\" but the model disagrees)\n\nCopy each viewed image (resized to 128×128 if PIL is available, otherwise just copy) into `\u003Cresults_dir>\u002Frca_images\u002F` so it can be embedded inline in the report.\n\nThis is the **only** image inspection required. Do not view dozens of images, do not run failure mode clustering, do not audit goldens — VCN does not have golden images.\n\n---\n\n## Reference invocation\n\nThe paste-and-edit end-to-end recipe (workspace, four paths, two numeric knobs, spec-file write, docker run, and the stdout sanity print that surfaces row counts for the script-check hook) lives in `references\u002Frecipe.md`. Use it verbatim, editing only the workspace, paths, and knobs.\n\n---\n\n## Outputs\n\nWrite everything into a timestamped folder under the experiment result directory: `\u003Cexperiment_result_dir>\u002Frca_results\u002FYYYY-MM-DD_HHMMSS\u002F`. The container's outputs (`kpi_gaps.parquet`, `threshold.txt`, `metrics.json`, `weak_samples_breakdown.txt`, and `unreachable_kpi.txt` when applicable) go straight there; the visual spot-check writes `rca_images\u002F`; the packaging hook adds `rca_config\u002F` and `claude_session.jsonl` automatically when `RCA_Report.md` is written. See `references\u002Fparameters-and-artifacts.md` for the full folder tree.\n\nAt the start of the run, get the real timestamp by running `date +%Y-%m-%d_%H%M%S` in Bash. Do NOT hardcode or guess. If the user specifies a custom output path, use that instead but maintain the same internal structure.\n\n---\n\n## Common pitfalls\n\nThe most consequential failure is **forgetting `top_k_per_label` when `min_recall=1.0`** — at that recall the chosen threshold sits at or below every NO_PASS score, so the fallback below-threshold filter matches ONLY misclassified PASS rows and `kpi_gaps.parquet` ends up with zero NO_PASS rows. Always include an explicit positive `top_k_per_label`. The full pitfalls list (spec file outside `$WORKSPACE`, unresolved `???` sentinels, wrong\u002Funpulled image tag, path-mount mismatch, `unreachable_kpi.txt` handling, missing `inference.csv` columns, missing train-YAML keys, `kpi_media_path` prefix mismatch, no GPU inside the container) and the CLI-drift reconciliation are in `references\u002Ftroubleshooting.md`.\n\n---\n\n## Report Structure\n\nWrite the RCA report into the timestamped output folder. It is a 7-section computational gap analysis (Verdict, Threshold Selection, Weakness Distribution, Top-K Weakest Samples, Visual Spot Check, Per-Label Breakdown, Recommended Actions), 1000–1800 words, with the confusion-matrix and per-label tables filled from `metrics.json` and the spot-check rows from `kpi_gaps.parquet`. When `unreachable_kpi.txt` exists, replace sections 3–6 with one short section quoting that file and collapse section 7 to a single retrain-or-relabel recommendation. See `references\u002Frca-report-structure.md` for the complete skeleton with every section heading, table layout, and the unreachable-KPI variant.\n\n---\n\n## Execution Order\n\n1. Resolve `DS_IMAGE` from `versions.yaml` (`images.tao_toolkit.data_services`), then run `docker info`, `nvidia-smi`, and `docker image inspect \"$DS_IMAGE\"` (pulling if missing) once to confirm the environment. Abort with a clear message if any fail.\n2. Run `date +%Y-%m-%d_%H%M%S` to get the timestamp; create `\u003Cexperiment_result_dir>\u002Frca_results\u002F\u003Ctimestamp>\u002F`.\n3. Write `vcn_aoi_spec.yaml` into the timestamped dir with `min_recall` and `top_k_per_label` filled in. Keep it under `$WORKSPACE` so the `-e` path resolves inside the container.\n4. Run `docker run … \"$DS_IMAGE\" gap_analysis vcn_aoi -e vcn_aoi_spec.yaml inference_results_dir=… train_config=… kpi_media_path=… output_dir=…`. The container writes `kpi_gaps.parquet`, `threshold.txt`, `metrics.json`, `weak_samples_breakdown.txt` into `results_dir`. Print the chosen threshold and kept-row counts to stdout so the script-check hook can verify the run produced output.\n5. If `unreachable_kpi.txt` exists, skip Step 6 and write the abridged report. Otherwise continue.\n6. Pick 10 weak samples (5 weakest PASS + 5 weakest NO_PASS) from `kpi_gaps.parquet`, view each test image with Read, classify, and copy each into `rca_images\u002F`.\n7. Write `RCA_Report.md` last — writing it triggers the packaging hook, which copies session logs and skill config alongside.\n",{"name":5,"tags":49,"license":15,"metadata":53,"description":7,"allowed-tools":56,"compatibility":57},[41,50,51,52],"rca","vcn","aoi",{"author":54,"version":55},"NVIDIA Corporation","0.3.0","Read Bash","Requires docker + nvidia-container-toolkit and a CUDA GPU. Pulls the `tao_toolkit.data_services` image declared in `versions.yaml` at the skill bank root.","main","7a940b8e5af16c95e5c190a524421d32fc7445a9",".agent\u002Fskills\u002Ftao-analyze-gaps-visual-changenet","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":66,"prompt":76,"zipAvailable":77},[67,70,73],{"tool":68,"command":69},"npx","npx skills add headmktlucca-ui\u002Fneuroads-lp --skill tao-analyze-gaps-visual-changenet",{"tool":71,"command":72},"bunx","bunx skills add headmktlucca-ui\u002Fneuroads-lp --skill tao-analyze-gaps-visual-changenet",{"tool":74,"command":75},"pnpm","pnpm dlx skills add headmktlucca-ui\u002Fneuroads-lp --skill tao-analyze-gaps-visual-changenet","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Ftao-analyze-gaps-visual-changenet\u002FSKILL.md",true,[79,85,89,95,99,103,107,111,115,119,123,127,131,135],{"path":80,"resourceType":81,"resourceTypeDesc":82,"sizeBytes":83,"sourceHtmlUrl":84},"BENCHMARK.md","REFERENCE","参考文档",4401,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Ftao-analyze-gaps-visual-changenet\u002FBENCHMARK.md",{"path":86,"resourceType":81,"resourceTypeDesc":82,"sizeBytes":87,"sourceHtmlUrl":88},"evals\u002Fevals.json",868,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Ftao-analyze-gaps-visual-changenet\u002Fevals\u002Fevals.json",{"path":90,"resourceType":91,"resourceTypeDesc":92,"sizeBytes":93,"sourceHtmlUrl":94},"hooks\u002F_parse-stdin.sh","SCRIPT","脚本",1471,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Ftao-analyze-gaps-visual-changenet\u002Fhooks\u002F_parse-stdin.sh",{"path":96,"resourceType":91,"resourceTypeDesc":92,"sizeBytes":97,"sourceHtmlUrl":98},"hooks\u002Frca-artifacts-check.sh",5018,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Ftao-analyze-gaps-visual-changenet\u002Fhooks\u002Frca-artifacts-check.sh",{"path":100,"resourceType":91,"resourceTypeDesc":92,"sizeBytes":101,"sourceHtmlUrl":102},"hooks\u002Frca-label-coverage.sh",3490,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Ftao-analyze-gaps-visual-changenet\u002Fhooks\u002Frca-label-coverage.sh",{"path":104,"resourceType":91,"resourceTypeDesc":92,"sizeBytes":105,"sourceHtmlUrl":106},"hooks\u002Frca-package.sh",2964,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Ftao-analyze-gaps-visual-changenet\u002Fhooks\u002Frca-package.sh",{"path":108,"resourceType":91,"resourceTypeDesc":92,"sizeBytes":109,"sourceHtmlUrl":110},"hooks\u002Frca-script-check.sh",5787,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Ftao-analyze-gaps-visual-changenet\u002Fhooks\u002Frca-script-check.sh",{"path":112,"resourceType":91,"resourceTypeDesc":92,"sizeBytes":113,"sourceHtmlUrl":114},"hooks\u002Frca-section-check.sh",3334,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Ftao-analyze-gaps-visual-changenet\u002Fhooks\u002Frca-section-check.sh",{"path":116,"resourceType":81,"resourceTypeDesc":82,"sizeBytes":117,"sourceHtmlUrl":118},"references\u002Fparameters-and-artifacts.md",6363,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Ftao-analyze-gaps-visual-changenet\u002Freferences\u002Fparameters-and-artifacts.md",{"path":120,"resourceType":81,"resourceTypeDesc":82,"sizeBytes":121,"sourceHtmlUrl":122},"references\u002Frca-report-structure.md",3322,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Ftao-analyze-gaps-visual-changenet\u002Freferences\u002Frca-report-structure.md",{"path":124,"resourceType":81,"resourceTypeDesc":82,"sizeBytes":125,"sourceHtmlUrl":126},"references\u002Frecipe.md",2083,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Ftao-analyze-gaps-visual-changenet\u002Freferences\u002Frecipe.md",{"path":128,"resourceType":81,"resourceTypeDesc":82,"sizeBytes":129,"sourceHtmlUrl":130},"references\u002Ftroubleshooting.md",3549,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Ftao-analyze-gaps-visual-changenet\u002Freferences\u002Ftroubleshooting.md",{"path":132,"resourceType":81,"resourceTypeDesc":82,"sizeBytes":133,"sourceHtmlUrl":134},"skill-card.md",3987,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Ftao-analyze-gaps-visual-changenet\u002Fskill-card.md",{"path":136,"resourceType":137,"resourceTypeDesc":138,"sizeBytes":139,"sourceHtmlUrl":140},"skill.oms.sig","OTHER","其他",7009,"https:\u002F\u002Fgithub.com\u002Fheadmktlucca-ui\u002Fneuroads-lp\u002Fblob\u002Fmain\u002F.agent\u002Fskills\u002Ftao-analyze-gaps-visual-changenet\u002Fskill.oms.sig",[142,167,192,218,236,271],{"skillNo":143,"name":144,"nameZh":6,"description":145,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":146,"licenseFileUrl":6,"sourceHtmlUrl":147,"categories":148,"keywords":155,"publishedAt":44},"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",[149,152],{"categoryNo":150,"categoryName":151,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000008","测试与质量",{"categoryNo":153,"categoryName":154,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000023","Skill与智能体开发",[156,157,158,159,160,161,162,163,164,165,166],"xunit","xunit skill","testing","injection","user","use","tests","triggers","unit","unit-testing","C#",{"skillNo":168,"name":169,"nameZh":6,"description":170,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":146,"licenseFileUrl":6,"sourceHtmlUrl":171,"categories":172,"keywords":175,"publishedAt":44},"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",[173,174],{"categoryNo":150,"categoryName":151,"taxonomyType":21,"taxonomyTypeDesc":22},{"categoryNo":153,"categoryName":154,"taxonomyType":21,"taxonomyTypeDesc":22},[176,177,158,178,179,180,181,182,161,183,184,185,186,187,162,163,188,189,190,191],"xcuitest","xcuitest skill","automation","test","testing framework","ios","devices","mobile","mobile-testing","ui test","local","apple","apps","native","for","objective",{"skillNo":193,"name":194,"nameZh":6,"description":195,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":146,"licenseFileUrl":6,"sourceHtmlUrl":196,"categories":197,"keywords":202,"publishedAt":44},"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",[198,201],{"categoryNo":199,"categoryName":200,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000015","视觉设计",{"categoryNo":153,"categoryName":154,"taxonomyType":21,"taxonomyTypeDesc":22},[203,204,205,206,207,208,209,210,161,211,212,213,214,215,216,190,217],"builder","world","world builder","MCP","create","mcp tools","Tools","game","pokemon","manage","style","via","theme","zone","generative",{"skillNo":219,"name":220,"nameZh":6,"description":221,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":146,"licenseFileUrl":6,"sourceHtmlUrl":222,"categories":223,"keywords":226,"publishedAt":44},"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",[224,225],{"categoryNo":150,"categoryName":151,"taxonomyType":21,"taxonomyTypeDesc":22},{"categoryNo":153,"categoryName":154,"taxonomyType":21,"taxonomyTypeDesc":22},[227,158,228,178,229,230,231,232,160,161,233,186,162,163,234,235],"webdriverio skill","typescript","javascript","browser","cloud","e2e","e2e-testing","URL","conf",{"skillNo":237,"name":238,"nameZh":6,"description":239,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":146,"licenseFileUrl":6,"sourceHtmlUrl":240,"categories":241,"keywords":251,"publishedAt":44},"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",[242,245,246],{"categoryNo":243,"categoryName":244,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000009","前端开发",{"categoryNo":199,"categoryName":200,"taxonomyType":21,"taxonomyTypeDesc":22},{"categoryNo":247,"categoryName":248,"taxonomyType":249,"taxonomyTypeDesc":250},"01JBSEEDCAT000000000000032","设计师","PROFESSION","适用职业",[252,253,254,255,256,257,258,259,260,261,262,263,264,265,266,267,268,269,163,270],"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":272,"name":273,"nameZh":6,"description":274,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":146,"licenseFileUrl":6,"sourceHtmlUrl":275,"categories":276,"keywords":279,"publishedAt":44},"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",[277,278],{"categoryNo":199,"categoryName":200,"taxonomyType":21,"taxonomyTypeDesc":22},{"categoryNo":247,"categoryName":248,"taxonomyType":249,"taxonomyTypeDesc":250},[252,254,280,261,264,161,281,282,187,269,283,163,188,284,190,285,286],"watchos design guidelines","workout","human","watch","interface","platform-design","features"]