| name | workbench-reference-workflows |
| description | Use when working on NPA reference SkyPilot YAMLs, runner scripts, cookbooks, or customer-adaptable pipeline implementations. |
Workbench Reference Workflows
The supported, customer-facing catalog is the npa.workflow spec set under
npa/workflows/workbench/npa-workflows/. The raw SkyPilot task YAMLs below are
internal runtime templates relocated to npa/src/npa/workflows/skypilot/; they
back the run_*.py wrappers and SkyPilot-only capabilities, and must not be
re-added to the shown npa/workflows/workbench/ catalog (guardrail-enforced).
When To Use
Use this skill for repository workflow YAMLs, runner scripts, cookbooks,
artifact contracts, and customer-adaptable pipeline implementations.
Procedure
- Start from the checked-in SkyPilot YAML under
npa/src/npa/workflows/skypilot/.
- Keep the runner thin. Python runners should materialize config, call the
workflow submission helper, and report artifacts; they should not duplicate
YAML orchestration logic.
- Keep all input and output paths configurable and run-scoped through S3.
- Validate YAML parsing and command help locally before live submission.
Current Reference YAMLs
bdd100k-pipeline.yaml: BDD100K ingest, backfill, CLIP embedding,
materialized views, training, and evaluation.cosmos3-ea-fetch.yaml: Cosmos3 source/checkpoint fetch.cosmos3-text-to-image-inference.yaml: H100 text-to-image smoke inference.isaac-lab-rl-train.yaml: single Isaac Lab RL job.isaac-lab-rl-sweep.yaml: Isaac Lab parameter sweep.mjlab-eval.yaml: MJLab evaluation.retargeting.yaml: motion retargeting.sim-to-real-loop.yaml: iterative sim-to-real loop.sim-to-real-pipeline.yaml: full sim-to-real pipeline.sim-to-real-trigger.yaml: trigger wrapper for sim-to-real work.sonic-train-standalone.yaml: standalone SONIC training.sonic-export.yaml: SONIC export.sonic-eval.yaml: SONIC evaluation.sonic-export-eval.yaml: export plus evaluation.sonic-locomotion-finetuning.yaml: retargeting, SONIC, and MJLab flow.vlm-eval.yaml and vlm-eval-benchmark.yaml: VLM evaluation loops.
Three-Tier Contract
- CLI: use
npa workbench workflow ... and tool-specific workflow commands
such as npa workbench mjlab workflow or npa workbench retargeting workflow.
- SDK: route through shared workflow submission helpers rather than shelling out
from business logic.
- YAML: SkyPilot YAML is the executable source of truth for workflow order,
resources, environment, and artifact paths.
Gotchas
- SkyPilot
envs does not support self-referencing interpolation. Use explicit
values and comments for alternatives.
sky jobs launch has no dry-run flag in the pinned path. Use local YAML
parsing, command help, and mock-endpoint tests before live submission.- Keep orchestration in YAML for SONIC locomotion; do not add a Python runner
that re-implements the DAG.
Verify
npa/.venv/bin/python -m pytest npa/tests/guardrails/test_skills_index.py -q
The smoke test parses the listed workflow YAMLs and invokes workflow CLI help.
