Model gemini-3.6-flashMeasured 8/13/2026v15 of 22 cases passed with skill · 6 comparable
Tokens, turns and time describe resource use, not a dollar price.
Lift withheld: 6 comparable cases; this report requires 8.
See it in action
A test case the skill fixed. Same model, same prompt, with and without the skill.
Prompt
We observed mutation p.V600E in BRAF in a melanoma patient sample. Search COSMIC database evidence for mutation frequency, sample count, and histology distribution.
What was checked, without and with the skill
Checked
Without skill
With skill
The response reports the COSMIC mutation sample count for BRAF p.V600E.
Passed
Passed
The response explicitly states licensing or authentication restrictions for COSMIC.
Missed
Passed
case-06 · gemini-3.6-flash · 8/13/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md38 lines2 files
Bio Database Evidence
Use This Skill For
Use this skill when the main task is biological database lookup, annotation, or evidence gathering across one or more biological sources:
Pathway mapping, ID conversion, KEGG pathways, Reactome enrichment, disease pathways, and pathway evidence.
Target-disease association evidence, tractability, safety, known drugs, and Open Targets evidence.
Protein structure evidence from AlphaFold DB or RCSB PDB, including UniProt IDs, mmCIF/PDB downloads, pLDDT, PAE, and structure metadata.
Protein-protein interaction evidence, STRING networks, hub proteins, and enrichment evidence.
Reference single-cell data lookup from CELLxGENE Census when the user asks for census metadata or expression data, not full downstream analysis.
Cross-database biological ID mapping and evidence tables across multiple resources.
Do Not Use This Skill For
Single-cell RNA-seq analysis, clustering, UMAP, marker genes, cell annotation, AnnData/h5ad container editing, or scVI/scANVI batch-correction planning. Use scanpy.
Bulk RNA-seq differential expression. Use pydeseq2.
BAM, SAM, CRAM, VCF, pileup, coverage, or region extraction as a primary file-processing task.
deepTools signal-track processing and heatmaps.
Protein language models, embeddings, inverse folding, or protein-design workflows.
Constraint-based metabolic modeling, FBA, or metabolic-engineering simulation.
BED/genomic interval embeddings, genomic-region ML, or gene regulatory network inference.
FCS or flow-cytometry file parsing.
Workflow
Identify the biological entity type: gene, transcript, variant, pathway, target, protein structure, protein interaction, trait association, or reference cell population.
Pick the narrowest source that answers the evidence question.
Preserve source names, query terms, access dates, identifiers, and API caveats in the result.
Return evidence in a table when comparing multiple sources.
State when authentication, license, rate limits, or non-public access restricts a source.
Source Guide
See references/database-evidence-sources.md for source-specific boundaries and query notes.