Life Science Research

Life Science Research Plugin

This plugin is a general life-sciences research layer for Codex. It packages a broad set of modular skills that can be composed to answer questions across human genetics, functional genomics, expression, pathway biology, protein structure, chemistry, clinical evidence, and public study discovery.

The goal is not to force every request through one fixed workflow. The goal is to help Codex understand the user's research question, normalize the relevant entities, choose the smallest useful set of skills, and synthesize a concise evidence-backed answer.

The plugin now includes a research-router-skill that should be treated as the default entrypoint for broad, ambiguous, or multi-step life-sciences research tasks.

What This Plugin Should Do

When a user invokes this plugin, treat it as a general research copilot for life sciences:

1. Understand the research task. Determine whether the user is asking for gene or target background, variant interpretation, locus-to-gene prioritization, pathway context, expression profiling, structure lookup, chemistry or ligand evidence, clinical-trial landscape, literature discovery, or dataset discovery.
2. Normalize the core entities. Resolve the gene, protein, disease, phenotype, variant, compound, tissue, cell type, species, accession, or pathway identifiers before branching into downstream lookups.
3. Route to the right skills. Prefer the minimum number of skills needed to answer the question well. Use single-source lookups for focused questions and multi-skill chains only when the question requires synthesis.
4. Parallelize only when it helps. If the work breaks into independent evidence lanes and Codex subagents are available, use them for bounded parallel retrieval and analysis. Keep initial scoping, entity normalization, and final synthesis with the coordinating agent.
5. Cross-check evidence across sources. Where the answer matters, compare orthogonal evidence types instead of over-indexing on one source.
6. Synthesize for the user. Return a concise research answer with the key evidence, important caveats, and clear next steps. Save raw payloads only when the user asks for them.

Research Patterns