P20260415-cv-pipeline

This skill should be used when the user asks to "prepare training data", "build a training dataset", "merge datasets for a CV model", "audit label quality", "check dataset labels", "curate training set", "create 00_data_preparation.yaml", "run annotation QA", or wants to turn raw/ folders into a training_ready/ split in this cv-pipeline repo. Also triggers on mentions of p00_data_prep, p02_annotation_qa, p01_auto_annotate, SAM3 label verification, or label grading.

Use when the user asks to "test the full pipeline", "run an end-to-end pipeline test", "smoke test the whole pipeline", "pipeline regression check", "verify every step p00..p10 works", or mentions exercising the cv-pipeline against raw unlabeled images. Orchestrates the canonical tests/test_p12_raw_pipeline.py (which covers p01→p02→p04 LS roundtrip→p00 merge+split→p05→p06→p07 HPO→p08→p09→p10 in order) plus a Playwright MCP sanity pass against Label Studio. Never skips stages, enforces services up-front, honors short Bash timeouts with background polling, and emits a markdown report.