biodatageeks

Cut a new polars-bio release — bump the version, finalize the changelog, run release-readiness checks, open the release PR, and draft a LinkedIn announcement. Use this whenever the user asks to "prepare a release", "cut X.Y.Z", "bump the version", "do the 0.32.0 release", "release readiness", or anything about shipping a new polars-bio version. polars-bio keeps its version in THREE places that must stay in lockstep (Cargo.toml, pyproject.toml, polars_bio/__init__.py) plus the regenerated Cargo.lock — this skill exists so none of them is forgotten and the changelog is cut correctly from [Unreleased]. Reach for it before manually editing any version string.

Validate polars_bio GFF/GTF predicate and projection pushdown for correctness and performance against real GENCODE data, using oxbow as an independent reference reader. Use this whenever you touch the pushdown path — the predicate translator (predicate_translator.py), the shared helpers (pushdown.py), the scan call sites (io.py _overlap_source / the GFF/GTF select() wrapper, utils.py, pileup_op.py), scan_gtf/scan_gff, or anything about predicate/projection pushdown, filter pushdown, attribute fields, or the issue #396 / PR #407 work — and any time you want to confirm a filtered or projected scan returns identical results with pushdown on vs off, or to benchmark how much pushdown actually speeds a query up. Reach for it before claiming a pushdown change is correct.

Profile the VEP annotation pipeline end-to-end and present a structured timing breakdown (pipeline stages, Lance variation/SIFT lookup, engine + transcript-engine stages) to find bottlenecks. Use this whenever the user wants to profile, benchmark, or get a performance/timing breakdown of vepyr annotation, the variation cache lookup, the VEP engine, or asks "where is the time going / what's the bottleneck" for chr-scale annotation — even if they don't say the word "profile". Also use after a cache-layout or encoding change to confirm lookup cost and that the pipeline is still engine-bound.