Drop-in OpenClaw / NemoClaw agent configurations for every ibl.ai solution segment. Each segment is a complete multi-agent system — one parent orchestrator plus a roster of specialist subagents — packaged so a NemoClaw host can import it with zero conversion.
This repository contains 7 segment configurations, one for each segment in the ibl.ai Solutions menu. Every segment directory is a self-contained OpenClaw gateway config: copy its contents into /sandbox/.openclaw/ on a NemoClaw host, recompute one hash, and reload the gateway.
| Segment | Parent agent | Subagents | Tool skills |
|---|---|---|---|
higher-education/ |
Campus Assistant | 16 | 10 |
k-12/ |
School Assistant | 12 | 12 |
enterprise/ |
Workplace Assistant | 12 | 12 |
government/ |
Agency Assistant | 12 | 10 |
legal/ |
Firm Assistant | 12 | 11 |
financial-services/ |
Advisory Assistant | 12 | 11 |
medical-healthcare/ |
Care Assistant | 12 | 10 |
95 agents (7 parents + 88 specialists) and 76 tool skills in total. This is a configuration-only repository — no application code, build system, or tests.
Each segment has one parent agent (the default entry point) and a set of specialist subagents. The parent interprets each request and delegates to the right specialist via the sessions_spawn tool; the subagents it may spawn are listed in its subagents.allowAgents. The parent then synthesizes the results.
<segment>/
├── README.md # segment overview + import steps
├── openclaw.json # gateway config — every agent, model, sandbox, skills
├── .config-hash # sha256 of openclaw.json
├── .env.example # template for ~/.openclaw/.env (sample credentials)
├── workspace/
│ └── .gitkeep # shared writable workspace (empty at install)
├── skills/
│ └── <tool>/
│ └── SKILL.md # one skill directory per major third-party tool
└── agents/
└── <agent-id>/
└── agent/
├── IDENTITY.md # persona — name, role, vibe
├── SOUL.md # behavioral guidelines
├── TOOLS.md # tool/integration notes + data sources
├── AGENTS.md # multi-agent routing table (parent only)
├── USER.md # user/environment context (optional)
├── BOOTSTRAP.md # one-time first-run setup (optional)
├── HEARTBEAT.md # periodic awareness checklist (optional)
├── MEMORY.md # seed long-term facts (optional)
└── auth-profiles.json # per-agent provider credentials (sample)
The repo contains only what the OpenClaw/NemoClaw runtime actually loads — the per-agent files above plus openclaw.json, .config-hash, skills/, and .env.example. Model, instance config, sandbox/security policy, and skill wiring all live in openclaw.json; there is no MODEL.md, CONFIG.json, SECURITY.md, or DATA.md, because the runtime never read them. Optional files (USER.md, BOOTSTRAP.md, HEARTBEAT.md, MEMORY.md) appear only on agents that genuinely need them.
The OpenClaw gateway configuration. Every agent — parent and subagents — is declared in agents.list[]. The parent carries default: true and a subagents.allowAgents list naming every child. agents.defaults holds the shared model, subagents limits, the skills array, and the sandbox block (the runtime's security policy). Each agent entry points at its workspace via agentDir, and carries its model and any heartbeat/session settings. After any edit to openclaw.json, recompute .config-hash.
skills/ holds one skill directory per major platform popular in the segment (Canvas, Banner, Salesforce, ServiceNow, Epic FHIR, Westlaw, and so on) — each a <tool>/SKILL.md with name, description, and a single-line metadata declaring the env vars it needs. agents.defaults.skills in openclaw.json wires them to the agents by name. Skill credentials resolve from environment variables; .env.example is the committed template for the OpenClaw daemon env file ~/.openclaw/.env.
Both auth-profiles.json (per-agent LLM provider keys) and .env.example (per-tool API credentials) ship with clearly-marked, non-functional sample placeholders. Replace them with real values before deploying.
Each segment directory drops onto an OpenClaw/NemoClaw host with no conversion. Set the host up first with the iblai-claw-setup guide (docs/server-setup.md) — that installs the gateway, the openclaw-gateway service, and the config root.
The config root is /sandbox/.openclaw/ on a NemoClaw sandbox host, or ~/.openclaw/ for a standalone OpenClaw install. Substitute whichever applies in the steps below.
# 1. Get this repo
git clone https://github.com/iblai/iblai-claw-agents.git
cd iblai-claw-agents
# 2. Copy the segment's contents into the config root (NemoClaw path shown)
cp -r higher-education/. /sandbox/.openclaw/
# contents land directly at the root — openclaw.json, agents/, skills/, workspace/
# 3. Give the sandbox user ownership of the agent workspaces (NemoClaw only)
chown -R sandbox:sandbox /sandbox/.openclaw/agents/ /sandbox/.openclaw/workspace/
# 4. Fill in real credentials
cp /sandbox/.openclaw/.env.example ~/.openclaw/.env
vi ~/.openclaw/.env # third-party platform API keys
# and replace the sample apiKey in every agents/*/agent/auth-profiles.json
# 5. Recompute the config hash (the gateway refuses to start on a mismatch)
cd /sandbox/.openclaw && shasum -a 256 openclaw.json > .config-hash
# 6. Reload the gateway
openclaw gateway reload # or: systemctl restart openclaw-gateway
# 7. Verify the parent agent is live
openclaw agent list # <segment>-assistant shows default: true, readyCopying a segment installs its parent orchestrator and every specialist subagent at once — they are all declared in that segment's openclaw.json. Each segment's own README.md carries the full roster and step-by-step.
If Claude Code (or any agent that can run a shell) is running on the OpenClaw/NemoClaw host, hand it the runbook below — it points at this repo and installs a full segment, parent plus all subagents, into the live environment.
Install runbook — paste into Claude Code on the host
Install the
<segment>agent configuration from https://github.com/iblai/iblai-claw-agents into this OpenClaw/NemoClaw environment.<segment>is one of:higher-education,k-12,enterprise,government,legal,financial-services,medical-healthcare.
- Confirm the OpenClaw config root:
/sandbox/.openclaw/on a NemoClaw sandbox host,~/.openclaw/for standalone OpenClaw. Confirm the gateway is installed (openclaw --version); if not, stop and point the user to https://github.com/iblai/iblai-claw-setup.- Clone the repo to a temp dir:
git clone https://github.com/iblai/iblai-claw-agents.git /tmp/iblai-claw-agents(orgit -C /tmp/iblai-claw-agents pullif already present).- Copy the chosen segment's contents into the config root:
cp -r /tmp/iblai-claw-agents/<segment>/. <config-root>/. The contents (openclaw.json,agents/,skills/,workspace/,.env.example) must land directly in the root, not in a subdirectory.- On a NemoClaw host, set ownership:
chown -R sandbox:sandbox <config-root>/agents/ <config-root>/workspace/.- Credentials — never invent values. Copy
<config-root>/.env.exampleto~/.openclaw/.envand tell the user which variables need real keys. Everyagents/*/agent/auth-profiles.jsonships a sample placeholderapiKey; tell the user to replace each with a real Anthropic API key. Do not start the gateway until the user confirms credentials are in place.- Recompute the hash from inside the config root:
cd <config-root> && shasum -a 256 openclaw.json > .config-hash.- Reload the gateway:
openclaw gateway reload(orsystemctl restart openclaw-gateway).- Verify:
openclaw agent list— the<segment>-assistantparent should showdefault: trueand statusready, alongside every subagent from that segment.Report which segment was installed, how many agents are now registered, and any credential variables still left as placeholders.
This installs the whole segment — one parent and all its specialists. To run more than one segment on the same host, the segments' openclaw.json files must be merged rather than copied over each other; that is a manual step beyond this runbook.
Higher Education — Campus Assistant + 16 specialists
Enrollment · Application Reader · Financial Aid · Academic Advisor · Tutoring · Retention · Student Services · Career Services · Faculty · Research · Alumni · IT Help Desk · Prospective Student · Yield · Continuing Education · Administrative
K-12 — School Assistant + 12 specialists
Tutoring · Lesson Planning · Assessment · Writing Feedback · Special Education · Content Creation · Student Safety · Family Communication · Curriculum Alignment · Professional Development · Research · Administration
Enterprise — Workplace Assistant + 12 specialists
Knowledge · IT Help Desk · Sales Enablement · Customer Support · Onboarding · HR · Marketing · Engineering · Meeting · Training · Data Analysis · Operations
Government — Agency Assistant + 12 specialists
Citizen Services · Knowledge · IT Help Desk · Employee Training · Compliance · Legislative · Budget · HR · Procurement · Onboarding · Constituent Communication · Security
Legal — Firm Assistant + 12 specialists
Case Research · Contract Review · Client Intake · Billing & Time · Discovery · Compliance · Knowledge · Brief Drafting · Conflicts Check · Training · Docket Management · IT Help Desk
Financial Services — Advisory Assistant + 12 specialists
Compliance · Risk Assessment · Client Advisory · Regulatory Reporting · Portfolio Analysis · Employee Training · KYC/AML · Fraud Detection · Client Onboarding · Knowledge · IT Help Desk · Operations
Medical / Healthcare — Care Assistant + 12 specialists
Clinical Support · Patient Education · Medical Coding · Compliance Training · Research · Provider Onboarding · Prior Authorization · Care Coordination · Documentation · Quality Improvement · IT Help Desk · Knowledge Management
- Create
<segment>/agents/<agent-name>-agent/agent/withIDENTITY.md,SOUL.md,TOOLS.md, andauth-profiles.json(addUSER.md/BOOTSTRAP.md/HEARTBEAT.md/MEMORY.mdonly if warranted). - Add an entry to
agents.list[]in<segment>/openclaw.json(setid,name,agentDir,model,identity,tools). - Add the new id to the parent agent's
subagents.allowAgentsand to itsAGENTS.mdrouting table. - Recompute the hash:
sha256sum openclaw.json > .config-hash. - Update the segment
README.mdand the roster in this file.
Mirror an existing segment directory: a parent agent plus specialist subagents, an openclaw.json declaring them all, a skills/ directory of <tool>/SKILL.md directories, .env.example, a README.md, and a recomputed .config-hash.
Proprietary — ibl.ai
