Skills¶
One row per skill. Each skill's SKILL.md is the authoritative description: its frontmatter gives the name and trigger, and its body gives instructions, inputs, outputs and quality gates. The sources are under skills/.
Two kinds of skill exist. A model-visible skill is listed in every session, and the model loads it when its description fits the task. A pipeline stage skill is hidden (disable-model-invocation: true in Claude Code, allow_implicit_invocation: false for Codex), so its description costs no prompt space. The model reaches a stage skill through omics-pipeline. A person can still call one by name, such as /bio-annotation.
Model-visible skills¶
| Skill | Use when | Main result |
|---|---|---|
omics-pipeline |
A genome, metagenome, or viral analysis needs a stage page, a driver, or the shared pipeline rules. | The shared rules (domain routing first, hypothesis register, five comparative axes, discovery and reporting, environment, scheduler), the stage pages below, and the stage skills to read. |
bioinformatics-project |
Starting, reorganizing, or reproducibility-hardening a bioinformatics project. | Project layout, pinned environment, task records, hypothesis register, restartable drivers, provenance, and sharing metadata. |
public-db-lookup |
Fetching a record from UniProt, NCBI, MGnify, InterPro, AlphaFold DB, STRING, or ENA. | A compact JSON envelope with bounded records, counts, and an optional raw-response file. |
bio-logic |
Auditing scientific reasoning, study design, bias, or strength of evidence behind project results. | A hypothesis register and a structured critique with uncertainty, alternative explanations, and follow-up checks. |
polars-dovmed |
Searching PMC Open Access or bioRxiv full text, through the connector or the hosted API or parquet script. | A run directory with query.json, raw responses, and timings, plus a paper list with relevance notes. |
crossref-lookup |
Validating DOIs, matching titles to citation metadata, or comparing papers by citations. | Crossref records, DOI matches, bibliography cleanup evidence, and OpenAlex citation counts. |
notebooks |
Building or converting reproducible marimo or Jupyter notebooks. | A fully executed notebook with embedded figures and clear analysis flow. |
beautiful-data-viz |
Making, restyling, or reviewing any plot, chart, figure, dashboard, or data visual. | Greyscale-first PNG, SVG, PDF, or HTML figures, color only to encode information, reproducible plotting code, and a Plotly Dash guide with an example app. |
Pipeline stage pages¶
A stage without scripts of its own is one decision page in skills/omics-pipeline/references/. A page gives the default tool per case, the rules a model would not guess, command lines, the QC gates to record, and the outputs. The model runs the tools; a project writes its own restartable driver as bioinformatics-project describes.
| Page | Use when |
|---|---|
reads-qc-mapping.md |
Raw FASTQ, adapters, contaminants, mapping and coverage. |
assembly.md |
QC-passed reads to isolate or metagenome contigs, with QUAST or MetaQUAST. |
binning.md |
Binning contigs and scoring completeness, contamination and chimerism of MAGs. |
gene-calling.md |
CDS prediction for prokaryotes, viruses and eukaryotes, and the tRNA and rRNA census. |
phylogenomics.md |
Marker alignments, ML trees, support values and closest relatives. |
structure.md |
Fold or complex prediction and structure search when sequence hits are weak. |
Tool notes sit beside the pages for tools that models know poorly: QuickBin, Autocycler, pyrodigal-gv, Infernal with its Rfam models, Boltz and TM-Vec.
Pipeline stage skills¶
Read these through omics-pipeline. Projects call their scripts by path, so the names and script paths stay stable.
| Skill | Use when | Main result |
|---|---|---|
tracking-taxonomy-updates |
Routing sequences by domain with QuickClade before any domain-specific stage, comparing NCBI, GTDB, ICTV, or eukaryotic taxonomy releases, or assigning taxonomy. | results/taxonomy/domain_routing.tsv, GTDB-Tk, EukCC, vConTACT3, or GVClass assignments, and a versioned release change and conflict report. |
bio-annotation |
Assigning function and taxonomy from sequence homology. | Annotation and taxonomy Parquet tables, marker_census.tsv, a family copy-number matrix, and family_expansion_candidates.tsv. |
bio-viromics |
Detecting, classifying, and QCing viral contigs. | Viral calls, quality summaries, taxonomy evidence, and candidate discovery tables. |
bio-interdomain-hgt |
Testing interdomain horizontal gene transfer and donor direction. | Homology, context, contamination, and per-gene phylogenetic evidence for candidate transfers. |
Skills from other repositories¶
The pack ships no skills for manuscript prose, manuscript review, proposal review, evaluation of AI-scientist output, PDF conversion, or claim-graph extraction. Install these from the repositories that own them, for example academic-writing with stop-slop for prose, nelli-paper-review for review, pdf-to-md for PDF conversion, and csag-extraction for claim graphs.