| name | org-bootstrap |
| slug | org-bootstrap |
| version | 1.0.0 |
| description | Bootstrap a multi-agent OpenClaw organization from a GitHub repo manifest |
| author | VelvetClaw |
| license | MIT |
| tags | ["bootstrap","organization","multi-agent","manifest","setup","automation"] |
| requirements | [{"name":"git","install":{"darwin":"xcode-select --install","linux":"sudo apt install git"}},{"name":"clawhub","install":{"all":"clawhub is included with OpenClaw"}}] |
org-bootstrap
Bootstrap a complete multi-agent OpenClaw organization from a single GitHub repository.
What This Skill Does
When you tell your agent to "bootstrap from [repo URL]", it will:
- Clone the repository containing the organization manifest
- Parse
manifest.yaml to understand the org hierarchy
- Create agent workspaces for every agent defined in the manifest
- Install skills from
skill-requirements.yaml via ClawHub
- Copy custom skills from the repo's
skills/ directory
- Initialize ClawVault templates for task/project/decision/lesson primitives
- Seed agent memory from the repo's
memory/ directory and per-agent files
- Wire workflows from the repo's
workflows/ directory
- Apply department rules from the repo's
departments/ directory
- Validate the entire deployment
Usage
Tell your agent:
Bootstrap my organization from https://github.com/username/my-org-config
Or with a specific model:
Bootstrap from https://github.com/username/my-org-config using openai/gpt-5.3-codex as the default model
Repository Structure
Your config repo must follow this structure:
my-org-config/
โโโ manifest.yaml # REQUIRED: Org hierarchy and coordination rules
โโโ agents/ # REQUIRED: One directory per agent
โ โโโ {agent-name}/
โ โโโ MANIFEST.yaml # REQUIRED: Agent role, model, skills, triggers
โ โโโ IDENTITY.md # REQUIRED: Agent persona and voice
โ โโโ SOUL.md # REQUIRED: Core directives
โ โโโ MEMORY.md # OPTIONAL: Seed memory
โโโ skill-requirements.yaml # OPTIONAL: ClawHub skills to auto-install
โโโ skills/ # OPTIONAL: Custom repo-local skills
โ โโโ {skill-name}/
โ โโโ SKILL.md
โโโ workflows/ # OPTIONAL: Trigger-based automation definitions
โ โโโ {workflow-name}.yaml
โโโ templates/ # OPTIONAL: ClawVault primitive schemas
โ โโโ task.md
โ โโโ project.md
โ โโโ decision.md
โ โโโ lesson.md
โโโ departments/ # OPTIONAL: Department coordination rules
โ โโโ {department-name}.yaml
โโโ memory/ # OPTIONAL: Shared organizational memory
โ โโโ {topic}.md
โโโ bootstrap.yaml # OPTIONAL: Custom validation checks
Bootstrap Procedure
When bootstrapping, follow this exact sequence:
Step 1: Clone the Repository
git clone {repo_url} /tmp/org-bootstrap-{timestamp}
cd /tmp/org-bootstrap-{timestamp}
Step 2: Validate the Manifest
Read manifest.yaml and verify:
- All agents referenced in the hierarchy have corresponding directories in
agents/
- All department leads are defined agents
- No circular delegation references
- The coordination section has valid values
If validation fails, report the specific errors and stop.
Step 3: Create Agent Workspaces
For each agent defined in manifest.yaml:
mkdir -p ~/.openclaw/workspace-{agent-id}
cp agents/{agent-id}/MANIFEST.yaml ~/.openclaw/workspace-{agent-id}/
cp agents/{agent-id}/IDENTITY.md ~/.openclaw/workspace-{agent-id}/
cp agents/{agent-id}/SOUL.md ~/.openclaw/workspace-{agent-id}/
if [ -f agents/{agent-id}/MEMORY.md ]; then
cp agents/{agent-id}/MEMORY.md ~/.openclaw/workspace-{agent-id}/
fi
Register the agent in OpenClaw's configuration:
- Add to the agents list in
~/.openclaw/openclaw.json
- Set the model from the agent's MANIFEST.yaml (or use the manifest default)
- Configure channels from the agent's MANIFEST.yaml
Step 4: Install Global Skills
If skill-requirements.yaml exists, install global skills:
clawhub install {skill-name}
Step 5: Install Per-Agent Skills
For each agent in skill-requirements.yaml's per_agent section:
cd ~/.openclaw/workspace-{agent-id}
clawhub install {skill-name}
Step 6: Copy Custom Skills
If skills/ directory exists in the repo:
cp -r skills/{skill-name} ~/.openclaw/workspace-{agent-id}/skills/
Step 7: Initialize Templates
If templates/ directory exists:
for workspace in ~/.openclaw/workspace-*/; do
mkdir -p "$workspace/templates"
cp templates/*.md "$workspace/templates/"
done
Step 8: Seed Memory
Copy shared memory files to each workspace:
for workspace in ~/.openclaw/workspace-*/; do
mkdir -p "$workspace/memory"
cp memory/*.md "$workspace/memory/"
done
Step 9: Wire Workflows
If workflows/ directory exists, copy workflow definitions:
for workspace in ~/.openclaw/workspace-*/; do
mkdir -p "$workspace/workflows"
cp workflows/*.yaml "$workspace/workflows/"
done
Step 10: Apply Department Rules
If departments/ directory exists:
for workspace in ~/.openclaw/workspace-*/; do
mkdir -p "$workspace/departments"
cp departments/*.yaml "$workspace/departments/"
done
Step 11: Validate
Run validation checks:
- Each agent workspace exists and has required files
- All referenced skills are installed
- Heartbeat is responding for each agent
- At least one workflow can be triggered
Report the results as a structured summary.
Step 12: Cleanup
rm -rf /tmp/org-bootstrap-{timestamp}
Sync (Re-Bootstrap)
To sync changes from the repo without recreating everything:
Sync my organization from https://github.com/username/my-org-config
This will:
- Pull latest changes from the repo
- Diff against current workspace state
- Apply only the changes (new agents, updated configs, new skills)
- Report what changed
Troubleshooting
- "Agent workspace already exists": The bootstrap will skip existing workspaces by default. Use "force bootstrap" to overwrite.
- "Skill not found on ClawHub": Check the skill name in
skill-requirements.yaml. Use clawhub search {name} to find the correct slug.
- "Manifest validation failed": Check that all agent names in the hierarchy match directory names in
agents/.
Stage 2: Loop Infrastructure
Stage 2 adds autonomous loop infrastructure to each agent. When bootstrapping a repo with Stage 2 files, the bootstrap process extends with these additional steps.
Stage 2 Agent Files
Each agent directory may contain these additional files (all optional โ Stage 2 features activate when present):
agents/{agent-name}/
โโโ MANIFEST.yaml # Stage 1 (existing)
โโโ IDENTITY.md # Stage 1 (existing)
โโโ SOUL.md # Stage 1 (existing)
โโโ MEMORY.md # Stage 1 (existing)
โโโ TASKS.md # Stage 2: Step-by-step work queue for loop cycles
โโโ CONTEXT.md # Stage 2: Stack, constraints, and known pitfalls
โโโ TOOLS.md # Stage 2: Available skills and capabilities
โโโ USER.md # Stage 2: Owner context and preferences
โโโ HEARTBEAT.md # Stage 2: Loop cycle health and status log
โโโ AGENTS.md # Stage 2: Hierarchy awareness โ subordinates and peers
โโโ EVOLVE.md # Stage 2: Weekly self-evolution with 20 questions
โโโ SELF.md # Stage 2.1: Wipe-recovery identity preservation document
Stage 2 Templates
templates/
โโโ task.md # Stage 1 (existing)
โโโ project.md # Stage 1 (existing)
โโโ decision.md # Stage 1 (existing)
โโโ lesson.md # Stage 1 (existing)
โโโ loop-config.md # Stage 2: Loop cycle configuration primitive
โโโ pressure-prompts.md # Stage 2: 25 strategic audit questions
โโโ evolve-cycle.md # Stage 2: Weekly evolution cycle schema
โโโ self-recovery.md # Stage 2.1: SELF.md agent recovery template
Stage 2 Workflows
workflows/
โโโ content-pipeline.yaml # Stage 1 (existing)
โโโ research-loop.yaml # Stage 1 (existing)
โโโ qa-monitoring.yaml # Stage 1 (existing)
โโโ agent-loop.yaml # Stage 2: Autonomous cron-triggered loop cycle
โโโ weekly-evolve.yaml # Stage 2: Weekly self-evolution trigger
Stage 2 Bootstrap Steps
After Step 10 (Apply Department Rules), add:
Step 10a: Copy Stage 2 Agent Files
for agent_dir in agents/*/; do
agent_id=$(basename "$agent_dir")
workspace="$HOME/.openclaw/workspace-$agent_id"
for stage2_file in TASKS.md CONTEXT.md TOOLS.md USER.md HEARTBEAT.md AGENTS.md EVOLVE.md SELF.md; do
if [ -f "$agent_dir/$stage2_file" ]; then
cp "$agent_dir/$stage2_file" "$workspace/"
fi
done
done
Step 10b: Configure Loop Crons
Read each agent's MANIFEST.yaml loop: section and configure the cron schedule:
- Tier 1 agents (chiefs): every 5 minutes
- Tier 2 leads: every 10 minutes
- Tier 2 members: every 15 minutes
Register the cron trigger in OpenClaw's scheduler for each agent.
Step 10c: Register Weekly Evolution
Set up the weekly evolution cron (Sunday midnight) for all agents:
for workspace in ~/.openclaw/workspace-*/; do
agent_id=$(basename "$workspace" | sed 's/workspace-//')
openclaw schedule add --agent "$agent_id" --cron "0 0 * * 0" --workflow weekly-evolve
done
Stage 2 Validation
Add these checks to Step 11:
- Each agent workspace has TASKS.md, CONTEXT.md, HEARTBEAT.md (minimum viable loop)
- Loop cron is registered and active for each agent
- Weekly evolution cron is registered
- Agent HEARTBEAT.md is writable
- AGENTS.md hierarchy matches manifest.yaml hierarchy