Multi-Agent Orchestrator Skill
Triggers: "spawn agent team", "multi-agent workflow", "coordinate specialists", "researcher coder writer reviewer", "parallel agent execution"
Produces: Configured multi-agent system with 16 specialized roles, domain toolsets, and 7 inter-agent communication primitives.
Architecture Overview
16 specialized agents, each with:
- Domain tools (4–17 specific tools)
- 7 shared communication tools:
consult_subagent,ask_subagent,post_to_blackboard,read_blackboard,send_agent_message,read_agent_inbox,broadcast_to_agents
All agents inherit a persistent Blackboard for shared state, P2P mailboxes for direct messages, and Broadcast for team-wide notifications.
Agent Registry
| Agent | Role | Domain Tools | Count |
|---|---|---|---|
analyze |
Analiză & Audit Date | read_file, list_directory, run_command, get_system_info, file_stats, parse_csv, parse_json, grep_file, count_lines | 9 |
research |
Cercetare Web | search_web, fetch_webpage, deep_research, extract_links | 4 |
coder |
Programare & Arhitectură | read_file, write_file, list_directory, run_command, run_code, run_python, check_python_syntax, grep_file, count_lines, file_stats | 10 |
writer |
Redactare & Conținut | read_file, write_file, list_directory, create_document, search_web, fetch_webpage, file_stats | 7 |
data |
Analiză Date & Vizualizări | read_file, write_file, parse_csv, parse_json, run_python, run_command, list_directory, file_stats | 8 |
security |
Audit Securitate & Privacy | read_file, list_directory, grep_file, file_stats, run_command, check_python_syntax | 6 |
planner |
Planificare & Strategie | read_file, list_directory, search_web, file_stats | 4 |
verify |
Verificare & Audit Livrabile | read_file, list_directory, run_command, search_web, check_python_syntax, file_stats, grep_file, count_lines | 8 |
repair |
Reparare & Depanare Erori | read_file, write_file, list_directory, run_command, run_code, move_or_copy, check_python_syntax, run_python | 8 |
devops |
DevOps & Automatizare Sistem | run_command, run_code, get_system_info, read_file, write_file, list_directory, file_stats, grep_file | 8 |
designer |
Frontend & Design UI/UX | read_file, write_file, list_directory, run_command, file_stats, grep_file | 6 |
translator |
Traducere & Localizare | read_file, write_file, create_document, search_web, fetch_webpage | 5 |
qa |
Testare & QA Automatizat | read_file, write_file, run_command, run_code, check_code_syntax, check_python_syntax, scan_codebase, search_code, list_directory, file_stats, grep_file, count_lines | 12 |
docs |
Documentare Tehnică & API Specs | read_file, write_file, create_document, read_document, list_directory, scan_codebase, search_code, grep_file, count_lines, file_stats, search_web, fetch_webpage | 12 |
git_refactor |
Git & Refactoring Arhitectural | git_status, git_log, git_diff, git_branch, run_command, read_file, write_file, scan_codebase, search_code, check_code_syntax, check_python_syntax, list_directory, file_stats, grep_file, github_list_repos, github_list_issues, move_or_copy | 17 |
scraper_api |
Web Scraping & Documentare API | scrape_page, fetch_webpage, search_web, rss_fetch, extract_entities, text_to_csv, extract_table_from_html, browser_open, browser_fetch, browser_screenshot, read_file, write_file, parse_json, parse_csv, run_command, run_code, list_directory | 17 |
Communication Primitives (7 tools, available to ALL agents)
| Tool | Purpose |
|---|---|
post_to_blackboard(key, value) |
Persistent shared key-value store |
read_blackboard(key) |
Read shared state |
send_agent_message(to_agent, message) |
Direct P2P message to agent inbox |
read_agent_inbox() |
Read own message queue |
broadcast_to_agents(message) |
Fan-out to all team members |
consult_subagent(agent, task) |
Synchronous sub-task delegation, returns result |
ask_subagent(agent, question) |
Async question, returns when answered |
Coordination Protocol: Coordinator-Workers
Pattern: Coordinator decomposes task → spawns specialized workers → aggregates results → delivers final.
Worker Contracts
Researcher — Input: { query, focus, output_format, limits } → Output: structured brief + sources + uncertainty note
Coder — Input: { feature, context, requirements, tests, output_path } → Output: code files + test results + docs
Writer — Input: { doc_type, sources, structure, tone, length } → Output: polished document + references
Reviewer — Input: { artifact_path, audit_type, criteria, min_severity } → Output: findings list + verdict (PASS/CONDITIONAL/FAIL) + blockers
Spawning (via sessions_spawn)
{
"runtime": "subagent",
"mode": "run",
"agentId": "<agent-name>",
"context": "isolated",
"cleanup": "keep"
}
Each worker receives a structured task matching its contract. Results returned via completion path.
Usage Patterns
Pattern 1: Research → Code → Write → Review
Coordinator spawns: research → coder → writer → verify
Each passes artifact to next via blackboard/P2P
Pattern 2: Parallel Fan-out
Coordinator broadcasts task → [analyze, security, qa] run in parallel
Results aggregated via blackboard
Pattern 3: Iterative Refinement
Coordinator → coder (implement) → verify (audit) → repair (fix) → verify (re-audit)
Quick Start (in a session)
- Spawn Coordinator (this agent) with task
- Coordinator decomposes, spawns workers via
sessions_spawn - Workers execute with domain tools + communication tools
- Coordinator aggregates, verifies, delivers
Memory & Persistence
- Blackboard: Survives agent restarts, shared namespace
- Agent Inboxes: Per-agent message queues
- Broadcast Log: Append-only team notification history
Verification Checklist
- All 16 agents spawn successfully
- Blackboard read/write works across agents
- P2P messages deliver to correct inbox
- Broadcast reaches all active agents
- consult_subagent returns structured result
- ask_subagent resolves async
- Coordinator-Workers contracts honored
Related Skills
coordinator-workers-protocol— core delegation patternclawhub— for publishing/installing this skill
Skill authored for OpenClaw Workshop. Apply to activate in this gateway.