Medical & Healthcare Agents

Drop-in OpenClaw / NemoClaw agent configuration for the ibl.ai Medical & Healthcare solution segment — one Care Assistant orchestrator plus 12 specialist subagents, packaged so a NemoClaw host can import the whole roster with zero conversion.

---

This directory contains the complete NemoClaw/OpenClaw configuration for the Medical / Healthcare solution segment. It is designed as a drop-in configuration: copy the contents of this directory into /sandbox/.openclaw/ on a NemoClaw sandbox, replace credentials, recompute the hash, and the segment is live.

The segment consists of 1 parent agent and 12 specialist subagents (13 agents total). All PHI handling follows HIPAA minimum necessary standards. All clinical agents include explicit disclaimers that AI output supports but does not replace licensed clinician judgment.

---

Agent Roster

Agent ID	Name	Role
medical-healthcare-assistant	Care Assistant	PARENT — entry point; interprets intent and delegates to specialist subagents via sessions_spawn
clinical-support-agent	Clinical Support	Evidence-based protocol recommendations and clinical decision support
patient-education-agent	Patient Education	Condition explanations and discharge instruction drafting
medical-coding-agent	Medical Coding	ICD-10-CM/PCS and CPT code selection assistance
compliance-training-agent	Compliance Training	HIPAA training delivery and certification tracking
research-agent	Research	Literature review and clinical trial matching
provider-onboarding-agent	Provider Onboarding	Credentialing, privileging, and new provider orientation
prior-authorization-agent	Prior Authorization	Insurance PA requests, status tracking, and denial appeals
care-coordination-agent	Care Coordination	Referral management, care transitions, and follow-up scheduling
documentation-agent	Documentation	Clinical note drafting assistance and CDI quality review
quality-improvement-agent	Quality Improvement	HEDIS/CAHPS metrics, care gap analysis, and accreditation readiness
it-help-desk-agent	IT Help Desk	EHR support, system access, and IT ticket management
knowledge-management-agent	Knowledge Management	Clinical protocol search, formulary guidance, and policy retrieval

---

Import Instructions

Follow these steps to deploy this configuration onto a NemoClaw sandbox.

1. Copy directory contents

git clone https://github.com/iblai/medical-healthcare-agents.git
cp -r medical-healthcare-agents/. /sandbox/.openclaw/

This copies openclaw.json, .config-hash, workspace/, skills/, .env.example, and the entire agents/ tree into the sandbox root.

2. Set ownership

Ensure the sandbox user owns all agent directories:

chown -R sandbox:sandbox /sandbox/.openclaw/agents/
chown -R sandbox:sandbox /sandbox/.openclaw/workspace/

3. Set up third-party platform credentials

Copy the included sample env file and fill in real API keys for every platform the segment connects to:

cp /sandbox/.openclaw/.env.example ~/.openclaw/.env
vi ~/.openclaw/.env

.env.example is the committed sample template representing the OpenClaw daemon env file ~/.openclaw/.env; all values in it are non-functional placeholders. The live ~/.openclaw/.env file (never committed) is sourced by the OpenClaw gateway at startup and injected into each agent's sandbox environment. See Tool Skills below for the full list of platforms and the variables each one requires.

4. Replace sample LLM credentials

Every agent directory contains agents/<agent-id>/agent/auth-profiles.json with a clearly marked sample key. Replace the apiKey value with a real Anthropic API key (and any other provider credentials your deployment requires) before going live:

# Example — repeat for each agent directory
vi /sandbox/.openclaw/agents/medical-healthcare-assistant/agent/auth-profiles.json

The file structure is:

{
  "profiles": {
    "anthropic": {
      "provider": "anthropic",
      "apiKey": "<YOUR-REAL-KEY-HERE>"
    }
  }
}

5. Recompute the config hash

After any modification to openclaw.json (required if you edit the file at all), recompute the hash from inside the openclaw root directory:

cd /sandbox/.openclaw
sha256sum openclaw.json > .config-hash

On macOS:

cd /sandbox/.openclaw
shasum -a 256 openclaw.json > .config-hash

6. Reload the OpenClaw gateway

openclaw reload
# or, if running as a systemd service:
systemctl reload openclaw-gateway

---

Tool Skills

The skills/ directory contains one subdirectory per third-party platform, each with a SKILL.md file. Each skill describes what the platform is, how agents authenticate to it, and which operations are available — giving every agent in the segment the context it needs to invoke that platform correctly. All skills draw their runtime credentials from ~/.openclaw/.env (.env.example is the committed sample template containing non-functional placeholder values; copy it to ~/.openclaw/.env and fill in real API keys before deploying).

Skill directory	Description
availity/SKILL.md	Availity Essentials — lets an agent perform real-time eligibility verification, prior authorization initiation and status checks, and payer enrollment management via the Availity clearinghouse REST API.
cerner-fhir/SKILL.md	Cerner Millennium FHIR R4 API — lets an agent read patient clinical data (demographics, encounters, diagnoses, medications, notes) within a Cerner-hosted health system.
epic-fhir/SKILL.md	Epic EHR FHIR R4 API — lets an agent read and write patient clinical data (demographics, encounters, diagnoses, medications, notes, orders) within an Epic-hosted health system.
healthstream/SKILL.md	HealthStream LMS — lets an agent query employee training transcripts, assign compliance curricula, check credential expiration alerts, and pull completion reports from the HealthStream learning management system.
innovaccer/SKILL.md	Innovaccer — lets an agent query population health analytics, patient risk scores, care gap lists, and SDOH screening data from the Innovaccer unified patient data platform.
micromedex/SKILL.md	Micromedex (Merative/IBM) — lets an agent perform drug interaction screening, dose range checking, toxicology lookups, and IV compatibility queries using the Micromedex drug knowledge base.
nuance-dax/SKILL.md	Nuance DAX (Dragon Ambient eXperience) — lets an agent submit encounter audio or transcripts and retrieve AI-generated structured clinical note drafts for clinician review.
pubmed/SKILL.md	PubMed / NCBI E-utilities — lets an agent search MEDLINE biomedical literature, retrieve structured abstracts, and export citations for clinical research and evidence synthesis tasks.
servicenow/SKILL.md	ServiceNow (Healthcare IT) — lets an agent create, update, and query IT service management tickets, knowledge articles, and configuration items within a ServiceNow instance deployed in a health system.
uptodate/SKILL.md	UpToDate (Wolters Kluwer) — lets an agent query evidence-based clinical decision support content including treatment recommendations, drug monographs, and clinical practice guidelines.

---

Agent Directory Layout

Each agent lives at agents/<agent-id>/agent/ and may contain the following files:

agents/<agent-id>/agent/
├── IDENTITY.md          # Agent persona: Name, Role, Vibe
├── SOUL.md              # Behavioral guidelines and boundaries
├── TOOLS.md             # Available integrations, tool usage notes, and data sources
├── auth-profiles.json   # LLM provider credentials (replace before deploy)
├── AGENTS.md            # Sub-agent delegation map (parent agent only)
├── MEMORY.md            # (optional) Seed long-term facts for rule/code-heavy agents
├── BOOTSTRAP.md         # (optional) One-time first-run setup steps; consumed after first use
├── HEARTBEAT.md         # (optional) Periodic monitoring checklist for proactive-watch agents
└── USER.md              # (optional) User/environment-specific context when genuinely needed

MEMORY.md, BOOTSTRAP.md, HEARTBEAT.md, and USER.md are only present for agents where they are genuinely warranted. Agents without proactive monitoring duties, seed knowledge requirements, or first-run setup do not include these files. Model selection and instance settings (heartbeat schedule, session scope) are configured in openclaw.json — not in per-agent files.

---

Architecture Notes

• The parent agent (medical-healthcare-assistant) is marked "default": true in openclaw.json and is the entry point for all incoming sessions.
• The parent delegates to specialist subagents via the sessions_spawn tool; it does not perform deep clinical, coding, or compliance work itself.
• Concurrent spawning is enabled (maxConcurrent: 8) to support multi-domain requests (e.g., simultaneous referral and PA initiation).
• All agents share a common writable workspace at /sandbox/.openclaw/workspace/ for inter-agent artifact passing.
• maxSpawnDepth: 2 prevents subagents from spawning further subagents, keeping the delegation hierarchy flat and auditable.

---

License

Released under the MIT License.