[blessley-dev]

Summary

Updates the Parabricks skill for Parabricks 4.8.0 (4.8.0-1).

The following changes were made:

Retarget to 4.8.0

• Document tool option and behavior changes from the 4.8.0 release notes
• Add 4.8.0 runtime requirements; the default CUDA probe image is now nvidia/cuda:13.3.1-base-ubuntu22.04
• Point the baseline references (tool index, getting started, performance guidance, installation requirements, release notes) at the 4.8.0 manual; tool links stay unversioned
• Accept 4.8.0 in the STAR translator
• Bumped the skill version to 1.4.0

Evals

• Retarget existing evals to 4.8.0
• Add parabricks-019 to parabricks-024 for the 4.8.0 changes
• Add parabricks-025 to test the 4.7.1 previous-release notes

Previous-release (v4.7.1) notes

Labeled only where a 4.7.1 command would otherwise come out wrong or fail:

• WES also needs --disable-small-model
• --chaining-on-gpu is a bare switch
• Giraffe indexes use vg 1.70.0
• Custom engines use TensorRT 10.14.1.48
• Pairing two minimap2 --in-fq files is new in 4.8.0

Fixes to existing tool references

• deepsomatic and the pangenome tools have no --gvcf option
• ont_germline and pacbio_germline have no --out-qc-metrics-dir option
• somatic added to the shared BWA and memory rows
• indexgvcf writes .idx for an uncompressed .g.vcf

[greptile-apps]

Confidence Score: 5/5

[Medium impact] The PR appears safe to merge; no actionable defects were established.

Summary

Updates the Parabricks skill to 1.4.0 for Parabricks 4.8.0 (4.8.0-1).

• Parabricks guidance now matches the 4.8.0 tools.
• The runtime helper checks hosts with 4.8.0's CUDA probe image.
• The STAR translator accepts Parabricks 4.8.0 requests.
• Parabricks evaluations now cover 4.8.0 behavior and a 4.7.1 case.

No actionable issues were established. Tests and Parabricks workloads were not run.

_{Reviews (1) · Last reviewed commit: "Adding new evals for v4.8.0 Parabricks c..." · Reviewed by Greptile}

[blessley-dev]

@trvachov @ohadmo @delagoya could you give this a review? Thanks

[delagoya]

Manual review of edits LGTM. Local run of skillevaluator tiers 1 & 2 pass; having issues with tier3 evals running to completion. @ohadmo can you run these via CI?

[ohadmo]

Thanks @blessley-dev! looking into it

[ohadmo]

/nvskills-ci