| name | reference-dataset-curation |
| description | Use when an agent must select, filter, deduplicate, and document NCBI/GenBank/reference viral genomes before alignment, primer design, PhyloGuide, or tNGS panel design. |
Reference Dataset Curation
Use this skill after taxonomy curation and before MAFFT or Agent API submission. The goal is a reproducible accession manifest, not just a broad NCBI query.
Workflow
- Start from literature/curated accession lists when possible; use broad NCBI queries only as a fallback.
- Record accession, species, genotype/subtype/clade, host, country, date, completeness, length, ambiguous-base fraction, and source.
- Filter by the user’s scope and quality thresholds.
- Deduplicate exact accession and exact sequence hash.
- Keep rejected records and reasons.
python agentskill/scripts/curate_references.py curate-accessions \
--metadata-csv accession_metadata.csv \
--taxonomy-json taxonomy.json \
--species "Target virus" \
--host "target host" \
--min-length 1000 \
--max-n-fraction 0.05 \
--require-complete \
--dedupe-by-sequence \
--output accession_manifest.json \
--accepted-csv accession_accepted.csv \
--rejected-csv accession_rejected.csv
Acceptance Gate
Proceed only when:
- accepted accessions cover the required groups and time/region scope;
- all filters are recorded in
accession_manifest.json;
- rejected records are available for audit;
- accession versions are pinned where available.