Teng-bio Bioinformatics Evidence Orchestrator
Use this as a router and evidence-pack builder for bioinformatics work. It does not replace specialist database, omics, literature, visualization, or reproduction skills.
Boundary
This skill does:
- classify the user's bioinformatics task and route to existing specialist skills
- collect confirmed evidence, inferred interpretation, risks, and missing information
- produce writing-ready structured artifacts such as
EVIDENCE_PACK.md,RISK_TABLE.md,FIGURE_INVENTORY.md, andMISSING_INFO.md - prepare the handoff to manuscript-writing skills without drafting the manuscript itself
This skill does not:
- write abstracts, introductions, Results prose, Discussions, reviewer responses, cover letters, or PPT text
- run specialist analyses itself when a dedicated skill exists
- invent sample sizes, accessions, p values, FDR values, software versions, genome builds, mechanisms, or citations
- make exploratory evidence sound causal or validated
If the user asks for both analysis and writing, first complete the evidence pack, then route to a writing skill in a separate step.
Routing map
| User intent | Route to |
|---|---|
| PubMed, MeSH, PMID, biomedical literature search | pubmed-database; add auto-deep-research for broader web/literature discovery |
| Life-science database facts, genes, proteins, diseases, drugs, pathways, variants | tooluniverse |
| Literature deep research, mechanism evidence, evidence grading | tooluniverse-literature-deep-research |
| “文献是怎么做的”, method/data extraction from papers or supplements | literature-method-data-miner |
| RNA-seq count matrix, DESeq2, DEG, design formula, batch effects | tooluniverse-rnaseq-deseq2; add scientific-critical-thinking for rigor audit |
| GO/KEGG/Reactome/GSEA/ORA/gene-set analysis | tooluniverse-gene-enrichment |
| DNA/RNA/protein sequence, FASTA, GenBank, RefSeq, ENA | tooluniverse-sequence-retrieval |
| Protein structures, PDB, PDBe, AlphaFold | tooluniverse-protein-structure-retrieval |
| Phylogeny, alignments, Newick, orthologs, molecular evolution | research-project-os first in .project_os projects; then execute with project scripts or general tools |
| Code repository reproduction or README-first paper repo intake | repo-intake-and-plan -> env-and-assets-bootstrap -> minimal-run-and-audit |
| Methodology, statistics, bias, confounding, evidence quality | scientific-critical-thinking |
| Publication figures from existing data | scientific-visualization or nature-figure; keep figure prose separate |
| Multi-step project with durable state | research-project-os for branch/task/run/result/asset provenance; project-flow-guard only for extra output/version guardrails |
Workflow
- Clarify the task type only if necessary. If enough context exists, infer the route.
- Route to specialist skills instead of reimplementing their methods.
- Capture provenance: inputs, accessions, software/tools, versions, parameters, references, dates, and output paths when available.
- Assess evidence with labels:
confirmed: directly supported by supplied data, logs, tables, or verified recordsinferred: reasonable interpretation but not directly shownmissing: required for a strong claim but absentunsupported: stated by the user but not supported by current evidence
- Build structured artifacts using
references/evidence-pack-template.mdwhen preparing a writing handoff. - Stop before prose writing. Recommend the appropriate writing skill only after evidence is organized.
Evidence discipline
- For datasets, preserve original accession/path and state source, organism, assay, sample count, groups, and metadata status.
- For workflows, record command/config/profile, engine, version, container/environment, reference genome, annotation, database versions, seeds, and outputs when known.
- For differential analysis, record normalization, design formula or contrast, covariates, filtering, multiple-testing method, thresholds, and effect-size fields.
- For enrichment, record species, ID namespace, gene universe/background, database/library, method, correction, and threshold.
- For figures, separate the figure's message from the evidence actually shown.
- Always mark missing or ambiguous items instead of filling them in.
Default output
For short tasks, return:
Route chosen:specialist skill(s) and why.Evidence summary:confirmed facts and key caveats.Risks / missing information:items blocking stronger claims.Next step:analysis, validation, or writing handoff.
For writing handoff or multi-part tasks, create or draft:
EVIDENCE_PACK.mdRISK_TABLE.mdFIGURE_INVENTORY.mdMISSING_INFO.md
Use the template in references/evidence-pack-template.md.