AI Agent Workflow Collection

Critically review, score, compare, and rank one or more AI scientist outputs for biology, bioinformatics, computational life science, or adjacent research tasks. Trigger when the user asks to evaluate notebooks, code, figures, analyses, manuscripts, software, or final reports produced by AI scientists; compare multiple AI scientists on the same task; judge publication readiness; or audit rigor, reproducibility, novelty, and task completion. Do not use this skill to perform the original research task itself unless the user is explicitly asking for a reviewer-style audit of already produced outputs.

Generate anywidget components for marimo notebooks.

Create publication-quality, aesthetically refined charts in Python/Jupyter (matplotlib/seaborn): readable axes, tight whitespace, and curated palettes (categorical/sequential/diverging).

Assemble genomes/metagenomes and produce assembly QC artifacts.

Perform metagenomic binning, refinement, and QC with completeness/contamination checks.

Curate, validate, and standardize FASTA/FAA sequence databases: standardize headers, merge databases, remove duplicates, convert GenBank to FASTA, and generate statistics. Use when preparing reference databases for HMM searches, MMseqs2, BLAST, and other bioinformatics workflows.

Initialize a bioinformatics project scaffold with reproducible environments, schemas, and data cataloging. Use for new projects or repo setup.

Call genes and annotate basic features for prokaryotes, viruses, and eukaryotes.

Detect and polarize interdomain horizontal gene transfer (virus eukaryote/bacteria/archaea, and cross-cellular HGT) using reciprocal best hits, transfer-direction inference, a genomic-context contamination guard, and per-gene phylogenetic confirmation. Use when asked about HGT, lateral gene transfer, gene flow between domains, virus-host gene exchange, endogenous viral elements, or host-derived / host-acquired genes.

Evaluates scientific research rigor using systematic frameworks. Assesses methodology, statistics, biases, and evidence quality. Use when reviewing papers, critiquing claims, designing studies, rating evidence strength (GRADE/Cochrane ROB), checking study design, statistical critique, or risk of bias assessment.

Build marker gene alignments and phylogenetic trees.

Designs and scaffolds bioinformatics pipelines using Prefect (Python) with Dask for local/distributed task execution and Nextflow for HPC scheduler-native execution. Use when an agent must choose between Prefect+Dask vs Nextflow, generate runnable project skeletons, or adapt workflows for laptops, workstations, and HPC clusters (e.g., Slurm/PBS) with reproducibility, caching/resume, and resource-aware configuration.

Ingest, QC, and map reads with reproducible outputs. Use for raw read processing and coverage stats.

Aggregate results, train ML models, and produce reports with validated references.

Detect, classify, and QC viral contigs.

Generate reproducible bioinformatics/data-science Methods + run documentation from workflow run artifacts (Nextflow/Snakemake/CWL), including a top-of-doc workflow summary, exact executed commands, tool versions, parameters, QC, and outputs.

Search bioRxiv preprints through the official bioRxiv API and locally filter titles, abstracts, and authors for keyword queries. Use when you need recent biology preprints, bioRxiv-native metadata, date-range scans, DOI lookups, or author shortlists that may not yet appear in peer-reviewed literature indexes.

Run a Codex-native planning and implementation loop with docs/plans progress tracking, MEMORY.md failure memory, per-task git worktrees, and iterative verification until tasks pass or need human intervention. Use when you want Mycelium-like orchestration for Codex without the Mycelium stack.

Query the Crossref REST API for DOI validation, title search, citation metadata, and bibliography audits. Use when you need DOI lookup, title-to-DOI matching, or reference metadata cleanup.

Extract a CSAG (Conditional Scientific Argumentation Graph) from a manuscript using the canonical argumentation spine (Assertions, EvidenceItems, EvidenceLinks, InferenceSteps) while enforcing CSAG conditionality (no assertion without at least one Context). Also generates paper-grounded Q&A items using bundled QA templates.

Implement a research paper in a marimo notebook fully automatically without extra user input.

Queries JGI Lakehouse (Dremio) for genomics metadata from GOLD, IMG, Mycocosm, Phytozome. Downloads genome files from JGI filesystem using IMG taxon OIDs. Use when working with JGI data, GOLD projects, IMG annotations, or downloading microbial genomes.

Run a multi-agent scientific manuscript review with parallel specialist reviewers, disagreement checks, and an editor meta-review. Use when reviewing a manuscript, preprint, revision, or rebuttal in Codex or Claude Code.

Create/refactor reproducible analysis notebooks with Marimo (preferred) or Jupyter (minimal support). Use for interactive, narrative-first analyses.