Agents¶
Agents are Markdown system prompts. They define the role, the skill lookup policy, and the workflow decision tree for a broad class of tasks. They do not execute anything by themselves; Claude Code or Codex loads the prompt, then uses the router and the referenced skills.
Agent Overview¶
| Agent | Use it for | Typical first skills |
|---|---|---|
omics-scientist |
Reads, assemblies, MAGs, annotations, taxonomy, phylogenomics, viral discovery, JGI data, and final biological interpretation. | bio-reads-qc-mapping, bio-assembly-qc, tracking-taxonomy-updates, bio-annotation, bio-fasta-database-curator |
literature-expert |
Literature discovery, preprint scans, DOI lookup, citation metadata cleanup, impact checks, structured claim/evidence extraction, and current API docs. | polars-dovmed, arxiv-search, biorxiv-search, crossref-lookup, csag-extraction, get-api-docs |
science-writer |
Manuscript drafting, section rewrites, rebuttals, proposal critique, methods documentation, multi-reviewer evaluation, and argument-graph extraction. | scientific-writing, manuscript-review-council, proposal-review, bio-workflow-methods-docwriter, csag-extraction |
dataviz-artist |
Scientific data inspection, reproducible notebooks, exploratory plots, publication figures, and dashboards. | exploratory-data-analysis, notebooks, beautiful-data-viz, plotly-dashboard-skill |
Choosing an Agent¶
Use the router first when the task is not obvious:
Constrain the router to one agent when you already know the domain:
python3 scripts/skill_index.py route --agent omics-scientist "annotate viral contigs and compare relatives"
The output gives the selected agent, primary skills, supporting skills, suggested order, and file paths. Open the returned files before doing substantial work.
Source Files¶
| Agent | Source |
|---|---|
omics-scientist |
agents/omics-scientist.md |
literature-expert |
agents/literature-expert.md |
science-writer |
agents/science-writer.md |
dataviz-artist |
agents/dataviz-artist.md |