菜单
Use when authoring, reviewing, running, or debugging NPA SkyPilot workflows and runner scripts.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | skypilot-workflows |
| description | Use when authoring, reviewing, running, or debugging NPA SkyPilot workflows and runner scripts. |
SkyPilot is the sole workflow orchestrator in this repo. Argo is deprecated; do not add or revive Argo workflows.
SkyPilot lives in an isolated virtualenv outside NPA's main Python environment. Invoke it through NPA_SKYPILOT_BIN; never rely on sky from PATH.
Use npa skypilot bootstrap to create or reuse the pinned SkyPilot 0.12.2
venv, then set NPA_SKYPILOT_BIN="$(npa skypilot status --bin-path)".
The Kubernetes controller is the default path (W9-skypilot-k8s-controller). The VM controller exists only as a fallback.
envs does not support self-referencing variable interpolation. Use explicit comment blocks for alternatives, following the BDD100K_LABEL_MAP pattern in npa/src/npa/workflows/skypilot/bdd100k-pipeline.yaml.sky jobs launch has no dry-run flag. Use mock-endpoint mode for validation before live submission.Dag support is effectively single-task for this repo's burst path. Use npa burst submit-yaml only for rendered single-task SkyPilot YAMLs; use npa workbench workflow submit for multi-stage workbench YAMLs.resources.image_id before YAML setup runs. For private cr.*.nebius.cloud images, the submitter must inject SkyPilot Docker login config (SKYPILOT_DOCKER_SERVER, SKYPILOT_DOCKER_USERNAME, SKYPILOT_DOCKER_PASSWORD) into task secrets before launch. npa burst submit-yaml does this by minting a short-lived Nebius IAM token when the submitter has Nebius credentials.npa/src/npa/workflows/skypilot/bdd100k-pipeline.yaml.npa/scripts/run_bdd100k_pipeline.py, a thin wrapper around npa.orchestration.skypilot.submit_workflow.npa/scripts/run_isaac_lab_rl.py.Acquire /tmp/npa-commit-lock/workflows-skypilot before committing workflow files in parallel-run contexts.
Cleanup is best-effort and must not raise. also_teardown_controller=False is the safe default; only opt into controller teardown when no other run can be using it.
Use when ingesting, validating, curating, or querying production sensor data as a versioned dataset-of-record, or wiring the dataset-ingest-curate workflow.
Use when generating adversarial scenarios via RL, ranking mined failures of a policy-under-test, or wiring the adversarial-scenario-hardening workflow.
Use when navigating the Cosmos3 integration in NPA or locating upstream Cosmos3 framework files, defaults, scripts, configs, recipes, and docs.
Use when Cosmos3 setup, fetch, inference, CUDA, uv, Docker, Hugging Face, GitHub, NGC, or checkpoint staging fails in NPA or in an upstream Cosmos framework checkout.
Use when setting up Cosmos3 access through NPA, checking source or Hugging Face reachability, staging the public Cosmos3 framework and checkpoint cache, or deciding which NPA workflow to use before inference.
Use for Claude Code reviews that need robotics, simulation, GPU-routing, sim-to-real, or BDD100K pipeline domain context.