BioResearch Agent — Agent Router Skill
Capability
A rule-based intent classifier that translates a free-text research request into one of the framework's workflows/skills:
| Route | Workflow / skill |
|---|---|
literature | Literature review + co-occurrence graph (bioresearch-literature-analysis) |
biomarker | Biomarker discovery — DEG + enrichment (bioresearch-biomarker-discovery) |
causal | Mendelian randomization (bioresearch-causal-inference) |
case-study | Disease case study / validation run (bioresearch-disease-case-study) |
doctor | Environment & reproducibility check (bioresearch-environment-check) |
It is intent translation, not reasoning: it maps a request to an existing workflow and emits the matching CLI command. It does not generate, rank, or evaluate hypotheses, and uses no model and no network — the same input always yields the same route (deterministic).
Run
bioresearch route "find biomarkers for Parkinson's disease using GSE7621"
bioresearch route "test the causal effect of BMI on type 2 diabetes with Mendelian randomization"
Output is JSON: {route, confidence, matched_keywords, scores, rationale, suggestion} plus a
suggested CLI command. confidence is the share of all matched keywords belonging to the winning
route; an unmatched intent returns route: "unknown" with a suggestion list of available routes.
When to invoke
An agent should call this router first when it gets an ambiguous or free-form biomedical request, to decide which specialized skill/workflow to dispatch to. It is the entry funnel of the skill system — not a replacement for the analysis skills themselves.
Note
Honest scope: keyword-substring matching only. It will not understand paraphrases that contain
none of the known keywords (it then returns unknown rather than guessing). Extend
bioresearch/router.py:ROUTE_KEYWORDS to cover new phrasing.