Design and release reproducible Nextflow pipelines with controlled sample metadata, pinned software and references, test evidence, recovery, and a reviewable handoff.
A Nextflow run can finish cleanly and still use the wrong samples, a moving reference, or an unsafe resume. This skill starts with those failure modes.
It turns a bioinformatics pipeline requirement into explicit input and channel contracts, a DSL2 workflow plan, pinned containers and references, bounded resource rules, and a test ladder that includes partial and recovery paths. The final handoff separates proposed code from executed evidence and uses the narrowest honest verdict: blocked, ready for implementation, ready for a pilot, or ready for release.
It covers pipeline engineering and reproducibility. Scientific interpretation, clinical validation, and permission to move sensitive data stay with the qualified owners.
Input
We have three Bash scripts for FASTQ validation, alignment, and per-sample QC. Samples may have several lanes. Design a Nextflow pipeline for Slurm with Apptainer, but do not run it.
Output
A samplesheet and channel contract, keyed lane-gathering design, module and file plan, pinned container and reference ledger, Slurm profile boundaries, negative and integration tests, and a ready-for-implementation verdict with open owners.
Input
Review this DSL2 workflow and configuration. It uses a latest container tag, joins tumor and normal files by list position, retries every failure, and publishes directly to the accepted results bucket.
Output
A blocked pilot verdict, specific identity and pairing findings, corrected keyed channel contracts, digest and reference requirements, failure-class retry rules, a staged publish design, and the evidence needed to authorize a bounded pilot.
Provide the analysis purpose, sample and file relationships, reference assets, current scripts or workflow files, target executor, data-handling limits, expected outputs, and release authority. Ask the agent to design or review the Nextflow workflow and return the contracts, file plan, tests, release evidence, recovery decision, and verdict.
The analysis purpose, input types, pseudonymous sample relationships, samplesheet rules, expected outputs, and acceptance checks.
Current workflow or scripts, Nextflow target version, tool and container requirements, reference assembly and annotation, and immutable revisions when available.
Target executor, storage, quotas, privacy or residency constraints, allowed actions, reviewers, and the person authorized to approve a pilot or release.
SKILL.md; pipeline and handoff contract; release and recovery guide; OpenAI interface metadata
No reviews yet.
Input
A run failed after 63 of 80 samples. We changed one process script and its container tag, and someone already copied partial reports into the final folder. Can we use -resume?
Output
A corrected-rerun or clean-restart decision based on cache compatibility, quarantined partial outputs, preserved failure evidence, per-sample reconciliation, a versioned output target, and explicit recovery and release gates.
Creator
Oosvelo1
