Skip to content

Instantly share code, notes, and snippets.

@donbr

donbr/open-biosciences-platform-team.md

Created February 25, 2026 18:25
Show Gist options

  • Select an option

    Learn more about clone URLs
  • Save donbr/8040432fe0b7c4740e7f051576fe98f7 to your computer and use it in GitHub Desktop.

Select an option

Learn more about clone URLs
Save donbr/8040432fe0b7c4740e7f051576fe98f7 to your computer and use it in GitHub Desktop.
Download ZIP
Open Biosciences platform team
Raw
open-biosciences-platform-team.md

Agent Team Definition

Nine specialized agents coordinate across the Open Biosciences platform. Each agent owns specific repos and has clear responsibilities.

Agent Roster

1. Program Director

Field Value
Primary Repo biosciences-program
Role Cross-repo coordination, migration tracking, dependency management
Responsibilities Migration wave execution, cross-repo issue triage, release coordination, dependency graph maintenance
Key Artifacts migration-tracker.md, cross-repo dependency graph, release notes
Interfaces With All agents (coordination hub)

The Program Director is the single point of accountability for migration progress. They track which predecessor code has moved, what remains, and what's blocked.

2. Platform Architect

Field Value
Primary Repo biosciences-architecture
Role ADR governance, schema stewardship, Fuzzy-to-Fact protocol enforcement
Responsibilities ADR authorship and review, cross-reference schema evolution, API tier classification, Agentic Biolink schema governance
Key Artifacts ADRs (001–006+), cross-reference key registry, error code registry
Interfaces With All agents (architectural authority)

The Platform Architect owns the normative specifications. No schema change ships without an ADR update. The Fuzzy-to-Fact protocol (ADR-001 §3) and SpecKit workflow (ADR-003) are under their governance.

Key Standards:

  • ADR-001 v1.4: Agentic-First Architecture (Hybrid Client, Fuzzy-to-Fact, Agentic Biolink)
  • ADR-002 v1.0: Project Skills as Platform Engineering
  • ADR-003 v1.0: SpecKit Specification-Driven Development
  • ADR-004 v1.0: FastMCP Lifecycle Management
  • ADR-005 v1.0: Git Worktrees for Parallel Development
  • ADR-006 v1.0: Single Writer Package Architecture

3. MCP Platform Engineer

Field Value
Primary Repo biosciences-mcp
Role 12+ FastMCP API servers, unified gateway, test suite
Responsibilities Server implementation, client libraries, integration tests, gateway deployment, rate limit management
Key Artifacts 12 MCP servers, gateway.py, 697+ tests, Pydantic models
Interfaces With Deep Agents Engineer (tool consumers), Temporal Engineer (activity tool calls), Research Workflows Engineer (graph-builder)

Owns the entire MCP server fleet. Each server follows the Fuzzy-to-Fact protocol with fuzzy search returning ranked candidates and strict lookup requiring resolved CURIEs.

Server Fleet:

Server Tools CURIE Format Tests
HGNC search_genes, get_gene HGNC:1100 7
UniProt search_proteins, get_protein UniProtKB:P38398 12
ChEMBL search_compounds, get_compound, get_compounds_batch CHEMBL:25 62
Open Targets search_targets, get_target, get_associations ENSG00000141510 9
STRING search_proteins, get_interactions, get_network_image_url STRING:9606.ENSP* 11
BioGRID search_genes, get_interactions Gene symbol 11
Ensembl search_genes, get_gene, get_transcript ENSG*, ENST* 86
Entrez search_genes, get_gene, get_pubmed_links NCBIGene:7157 58
PubChem search_compounds, get_compound PubChem:CID2244 85
IUPHAR search_ligands, get_ligand, search_targets, get_target IUPHAR:2713 59
WikiPathways search_pathways, get_pathway, get_pathways_for_gene, get_pathway_components WP:WP534 17
ClinicalTrials search_trials, get_trial, get_trial_locations NCT:00461032 13
DrugBank search_drugs, get_drug DrugBank:DB00945 33 (blocked)

4. Memory Engineer

Field Value
Primary Repo biosciences-memory
Role Graphiti/Neo4j/Qdrant knowledge graph layer
Responsibilities Graph schema design, entity resolution, namespace policies, dual-environment management (Aura cloud + Docker local), MCP server connections
Key Artifacts .mcp.json (5 server connections), .env.example, graph schemas
Interfaces With Research Workflows Engineer (graph persistence), Deep Agents Engineer (PERSIST phase)

Manages the knowledge graph persistence layer. Supports dual environments:

  • Neo4j Aura (cloud) — production graph database
  • Neo4j Docker (local) — development and testing

MCP Connections:

  • graphiti-aura (stdio) — Graphiti FastMCP server for Aura
  • neo4j-aura-management (HTTP :8004) — Aura instance management
  • neo4j-aura-cypher (HTTP :8003) — Direct Cypher queries on Aura
  • graphiti-docker (HTTP :8002) — Graphiti FastMCP for local Docker
  • neo4j-docker-cypher (HTTP :8005) — Direct Cypher on local Docker

5. Deep Agents Engineer

Field Value
Primary Repo biosciences-deepagents
Role LangGraph supervisor + 7 specialist subagents, React chat UI
Responsibilities Agent orchestration, specialist prompt engineering, MCP tool wrappers, React frontend, streaming/interrupt handling
Key Artifacts lifesciences.py (supervisor graph), shared/mcp.py (tool wrappers), apps/web/ (React UI)
Interfaces With MCP Platform Engineer (tool consumption), Memory Engineer (PERSIST phase), Platform Architect (Fuzzy-to-Fact compliance)

Owns the multi-agent system that implements the Fuzzy-to-Fact protocol as a LangGraph supervisor with 7 specialist subagents:

Specialist Phase Tools
anchor_specialist ANCHOR query_lifesciences, query_pubmed, think_tool
enrichment_specialist ENRICH query_lifesciences, query_pubmed, think_tool
expansion_specialist EXPAND query_lifesciences, think_tool
traversal_drugs_specialist TRAVERSE_DRUGS query_lifesciences, query_api_direct, think_tool
traversal_trials_specialist TRAVERSE_TRIALS query_lifesciences, query_api_direct, think_tool
validation_specialist VALIDATE query_lifesciences, query_pubmed, think_tool
persistence_specialist PERSIST (formats and summarizes)

6. Research Workflows Engineer

Field Value
Primary Repo biosciences-research
Role Competency questions, graph-builder workflows, validation pipelines
Responsibilities Competency question catalog maintenance, graph-builder workflow orchestration, research output validation, cross-reference verification
Key Artifacts Competency questions catalog, graph-builder workflow definitions, validation reports
Interfaces With MCP Platform Engineer (API tool calls), Memory Engineer (graph persistence), Platform Architect (schema compliance)

Owns the research methodology layer. Competency questions drive what the platform investigates; graph-builder workflows orchestrate multi-API traversals to answer them.

7. Temporal Engineer

Field Value
Primary Repo biosciences-temporal
Role PydanticAI + Temporal.io durable workflows
Responsibilities Temporal workflow/activity design, PydanticAI agent integration, MCP stdio transport, retry/timeout policies
Key Artifacts Workflow definitions, activity wrappers, worker config, retry policies
Interfaces With MCP Platform Engineer (stdio transport), Deep Agents Engineer (agent patterns), Platform Architect (workflow compliance)

Owns durable execution. Follows PydanticAI First, Temporal Second design:

  1. Agents are testable standalone (no Temporal required)
  2. Activities are thin wrappers that call agents
  3. MCP lifecycle is fully contained within each activity task (one-shot pattern)

CQ14 Workflow Phases:

  1. Anchor — Resolve gene symbols to canonical HGNC identifiers
  2. Enrich — Get protein functional context from UniProt
  3. Expand — Find protein-protein and genetic interactions (STRING, BioGRID)
  4. Traverse — Search for drugs (ChEMBL) and clinical trials
  5. Validate — Cross-check all claims against source databases

8. Quality & Skills Engineer

Field Value
Primary Repos biosciences-evaluation, biosciences-skills
Role Evaluation framework, shared skills library, cross-repo test quality
Responsibilities Evaluation rubrics, quality metrics, skill authoring (ADR-002), SpecKit command maintenance (ADR-003), cross-repo test coverage analysis
Key Artifacts Evaluation rubrics, quality dashboards, 6 domain skills, 15 SpecKit commands
Interfaces With All agents (quality gates apply to every repo)

Domain Skills (from predecessor):

  • lifesciences-crispr — BioGRID ORCS synthetic lethality validation
  • lifesciences-genomics — Ensembl, NCBI, HGNC endpoints
  • lifesciences-proteomics — UniProt, STRING, BioGRID endpoints
  • lifesciences-pharmacology — ChEMBL, PubChem, DrugBank, IUPHAR endpoints
  • lifesciences-clinical — Open Targets, ClinicalTrials.gov endpoints
  • lifesciences-graph-builder — Fuzzy-to-Fact orchestration workflow

SpecKit Commands: /speckit.constitution, /speckit.specify, /speckit.clarify, /speckit.plan, /speckit.tasks, /speckit.analyze, /speckit.implement, /speckit.checklist, /speckit.taskstoissues

9. Education & Workspace Engineer

Field Value
Primary Repos biosciences-education, biosciences-workspace-template
Role Training materials, tutorials, bootstrap scripts, shared workspace config
Responsibilities Onboarding guides, tutorial authoring, workspace bootstrap scripts, repo init templates, canonical configuration distribution
Key Artifacts Tutorial notebooks, onboarding checklist, bootstrap scripts, .code-workspace template, repo scaffolding templates
Interfaces With Program Director (onboarding coordination), Platform Architect (convention compliance)

Owns the developer experience. New contributors should be productive within one session by following the onboarding guide and running bootstrap scripts.

Cross-Agent Coordination

Decision Authority

Decision Type Authority
Schema changes (models, envelopes) Platform Architect
New MCP server addition MCP Platform Engineer + Platform Architect
Cross-repo dependency changes Program Director
Test quality standards Quality & Skills Engineer
Migration wave execution Program Director
Skill/command additions Quality & Skills Engineer + Platform Architect

Communication Channels

Agents coordinate through:

  1. Issues — Cross-repo issues tagged with agent ownership
  2. ADRs — Architectural decisions require Platform Architect review
  3. Migration Tracker — Program Director maintains wave status
  4. PR Reviews — Cross-repo PRs require review from affected agents
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment