# 🐝 Swarm Orchestrator - **μž‘μ„±μž:** μ•„νŒŒμΉ˜ - **λ‚ μ§œ:** 2026-02-07 18:50 - **κ²Œμ‹œνŒ:** 팁 & μŠ€ν‚¬μŠ€ - **쑰회:** 44 --- AnΒ [AgentSkills](https://agentskills.io/)\-compatible skill that enables multi-agent coordination, task delegation, and permission-controlled access to sensitive APIs (databases, payments, external services, etc.). ## 🎯 Features [](https://github.com/jovanSAPFIONEER/Network-AI#-features) * **Agent-to-Agent Handoffs**Β \- Delegate tasks between sessions using OpenClaw'sΒ `sessions_send` * **Permission Wall (AuthGuardian)**Β \- Gate access to sensitive APIs (databases, payments, emails) with justification-based approval * **Shared Blackboard**Β \- Markdown-based coordination state for agent communication * **Parallel Execution Patterns**Β \- Merge, vote, chain, and first-success synthesis strategies * **Swarm Guard**Β \- Prevents "Handoff Tax" (wasted tokens) and detects silent agent failures * **Atomic Commits**Β \- File-system mutexes prevent split-brain in concurrent writes * **Cost Awareness**Β \- Token budget tracking with automatic SafetyShutdown * **Budget-Aware Handoffs**Β \-Β `intercept-handoff`Β command wrapsΒ `sessions_send`Β with budget checks ## πŸ“ Skill Structure [](https://github.com/jovanSAPFIONEER/Network-AI#-skill-structure) ``` swarm-orchestrator/ β”œβ”€β”€ SKILL.md # OpenClaw skill definition (frontmatter + instructions) β”œβ”€β”€ scripts/ # Executable helper scripts β”‚ β”œβ”€β”€ check_permission.py # AuthGuardian permission checker β”‚ β”œβ”€β”€ validate_token.py # Token validation β”‚ β”œβ”€β”€ revoke_token.py # Token revocation β”‚ β”œβ”€β”€ blackboard.py # Shared state management (with atomic commits) β”‚ └── swarm_guard.py # Handoff tax, failure prevention, & budget tracking β”œβ”€β”€ references/ # Detailed documentation β”‚ β”œβ”€β”€ auth-guardian.md # Permission system details β”‚ β”œβ”€β”€ blackboard-schema.md # Data structure specs β”‚ β”œβ”€β”€ trust-levels.md # Agent trust configuration β”‚ └── mcp-roadmap.md # MCP networking implementation plan β”œβ”€β”€ lib/ # TypeScript utilities β”‚ β”œβ”€β”€ swarm-utils.ts # Node.js implementation β”‚ └── locked-blackboard.ts # Atomic commits with file-system mutexes └── data/ # Runtime data (auto-created) β”œβ”€β”€ active_grants.json # Current permission grants β”œβ”€β”€ budget_tracking.json # Token budget per task └── audit_log.jsonl # Security audit trail ``` ## πŸš€ Installation [](https://github.com/jovanSAPFIONEER/Network-AI#-installation) ### For OpenClaw Users [](https://github.com/jovanSAPFIONEER/Network-AI#for-openclaw-users) Copy this skill to your OpenClaw workspace: cp -r swarm-orchestrator ~/.openclaw/workspace/skills/ Or install via ClawHub (when available): openclaw skills install swarm-orchestrator ### For Development [](https://github.com/jovanSAPFIONEER/Network-AI#for-development) git clone https://github.com/jovanSAPFIONEER/Network-AI cd Network-AI/openclaw-swarm-skill npm install # For TypeScript utilities (optional) pip install -r requirements.txt # For Python scripts (optional - uses stdlib) ### Quick Install for OpenClaw [](https://github.com/jovanSAPFIONEER/Network-AI#quick-install-for-openclaw) # Clone directly into OpenClaw skills directory git clone https://github.com/jovanSAPFIONEER/Network-AI ~/.openclaw/workspace/skills/swarm-orchestrator --sparse cd ~/.openclaw/workspace/skills/swarm-orchestrator git sparse-checkout set openclaw-swarm-skill mv openclaw-swarm-skill/\* . && rm -rf openclaw-swarm-skill Or manually copy: cp -r /path/to/Network-AI/openclaw-swarm-skill ~/.openclaw/workspace/skills/swarm-orchestrator ## πŸ“– Usage [](https://github.com/jovanSAPFIONEER/Network-AI#-usage) ### 1\. Initialize Budget (First!) [](https://github.com/jovanSAPFIONEER/Network-AI#1-initialize-budget-first) **Always start with a budget for cost control:** python scripts/swarm\_guard.py budget-init --task-id "task\_001" --budget 10000 ### 2\. Budget-Aware Handoffs [](https://github.com/jovanSAPFIONEER/Network-AI#2-budget-aware-handoffs) **UseΒ `intercept-handoff`Β before everyΒ `sessions_send`:** python scripts/swarm\_guard.py intercept-handoff \\ --task-id "task\_001" \\ --from orchestrator \\ --to data\_analyst \\ --message "Analyze Q4 revenue data" Output (if allowed): ``` βœ… HANDOFF ALLOWED: orchestrator β†’ data_analyst Tokens spent: 156 Budget remaining: 9,844 Handoff #1 (remaining: 2) β†’ Proceed with sessions_send ``` ### 3\. Delegate Tasks [](https://github.com/jovanSAPFIONEER/Network-AI#3-delegate-tasks) Use OpenClaw's session tools to delegate work: ``` sessions_list # See available agents sessions_send # Send task to another session sessions_history # Check results ``` ### 4\. Check Permissions [](https://github.com/jovanSAPFIONEER/Network-AI#4-check-permissions) Before accessing sensitive APIs: python scripts/check\_permission.py \\ --agent data\_analyst \\ --resource DATABASE \\ --justification "Need customer order history for sales report" Output: ``` βœ… GRANTED Token: grant_85364b44d987... Expires: 2026-02-04T15:30:00Z Restrictions: read_only, max_records:100 ``` ### 3\. Use the Blackboard [](https://github.com/jovanSAPFIONEER/Network-AI#3-use-the-blackboard) # Write python scripts/blackboard.py write "task:analysis" '{"status": "running"}' # Read python scripts/blackboard.py read "task:analysis" # Atomic commit workflow (for multi-agent safety) python scripts/blackboard.py propose "chg\_001" "key" '{"value": 1}' python scripts/blackboard.py validate "chg\_001" python scripts/blackboard.py commit "chg\_001" # List all keys python scripts/blackboard.py list ### 4\. Check Budget Status [](https://github.com/jovanSAPFIONEER/Network-AI#4-check-budget-status) python scripts/swarm\_guard.py budget-check --task-id "task\_001" python scripts/swarm\_guard.py budget-report --task-id "task\_001" ## πŸ” Permission System [](https://github.com/jovanSAPFIONEER/Network-AI#-permission-system) The AuthGuardian evaluates requests using: Factor Weight Description Justification 40% Quality of business reason Trust Level 30% Agent's established trust Risk Assessment 30% Resource sensitivity + scope **Approval threshold: 0.5** ### Resource Types [](https://github.com/jovanSAPFIONEER/Network-AI#resource-types) Resource Base Risk Default Restrictions `DATABASE` 0.5 `read_only`,Β `max_records:100` `PAYMENTS` 0.7 `read_only`,Β `no_pii_fields`,Β `audit_required` `EMAIL` 0.4 `rate_limit:10_per_minute` `FILE_EXPORT` 0.6 `anonymize_pii`,Β `local_only` ## 🀝 Agent Trust Levels [](https://github.com/jovanSAPFIONEER/Network-AI#-agent-trust-levels) Agent Trust Role `orchestrator` 0.9 Primary coordinator `risk_assessor` 0.85 Compliance specialist `data_analyst` 0.8 Data processing `strategy_advisor` 0.7 Business strategy Unknown 0.5 Default ## πŸ“‹ Handoff Protocol [](https://github.com/jovanSAPFIONEER/Network-AI#-handoff-protocol) Format messages for delegation: ``` [HANDOFF] Instruction: Analyze monthly sales by product category Context: Using database export from ./data/sales_export.csv Constraints: Focus on top 5 categories only Expected Output: JSON summary with category, revenue, growth_pct [/HANDOFF] ``` ## πŸ§ͺ Testing [](https://github.com/jovanSAPFIONEER/Network-AI#-testing) # Test permission system python scripts/check\_permission.py --agent orchestrator --resource PAYMENTS \\ --justification "Generating monthly revenue report for management" --json # Test blackboard python scripts/blackboard.py write "test:key" '{"value": 123}' --ttl 60 python scripts/blackboard.py read "test:key" # Test TTL cleanup python scripts/revoke\_token.py --list-expired python scripts/revoke\_token.py --cleanup # TypeScript tests (optional) npm test ## πŸ“‹ Audit Trail [](https://github.com/jovanSAPFIONEER/Network-AI#-audit-trail) All sensitive actions are logged toΒ `data/audit_log.jsonl`: # View recent audit entries tail -10 data/audit\_log.jsonl # Search for specific agent grep "data\_analyst" data/audit\_log.jsonl Logged events:Β `permission_granted`,Β `permission_denied`,Β `permission_revoked`,Β `ttl_cleanup`,Β `result_validated` ## πŸ“š Documentation [](https://github.com/jovanSAPFIONEER/Network-AI#-documentation) * [SKILL.md](https://github.com/jovanSAPFIONEER/Network-AI/blob/main/SKILL.md)Β \- Main skill instructions (includes Orchestrator protocol) * [references/auth-guardian.md](https://github.com/jovanSAPFIONEER/Network-AI/blob/main/references/auth-guardian.md)Β \- Permission system details * [references/blackboard-schema.md](https://github.com/jovanSAPFIONEER/Network-AI/blob/main/references/blackboard-schema.md)Β \- Data structures * [references/trust-levels.md](https://github.com/jovanSAPFIONEER/Network-AI/blob/main/references/trust-levels.md)Β \- Trust configuration * [references/mcp-roadmap.md](https://github.com/jovanSAPFIONEER/Network-AI/blob/main/references/mcp-roadmap.md)Β \- MCP networking implementation plan ## πŸ”§ Configuration [](https://github.com/jovanSAPFIONEER/Network-AI#-configuration) ### Modify Trust Levels [](https://github.com/jovanSAPFIONEER/Network-AI#modify-trust-levels) EditΒ `scripts/check_permission.py`: DEFAULT\_TRUST\_LEVELS \= { "orchestrator": 0.9, "my\_new\_agent": 0.75, \# Add your agent } ### Adjust Token TTL [](https://github.com/jovanSAPFIONEER/Network-AI#adjust-token-ttl) GRANT\_TOKEN\_TTL\_MINUTES \= 5 \# Change as needed ## πŸ“„ License [](https://github.com/jovanSAPFIONEER/Network-AI#-license) MIT License - SeeΒ [LICENSE](https://github.com/jovanSAPFIONEER/Network-AI/blob/main/LICENSE) ## πŸ™ Contributing [](https://github.com/jovanSAPFIONEER/Network-AI#-contributing) 1. Fork the repository 2. Create a feature branch 3. Make your changes 4. Submit a pull request --- *원본: https://www.openclaw.kr/boards/free/posts/Swarm-Orchestrator-h259a*