Multi-Agent Coordination Protocol
Rules for role-based agent teams: brief-only messages, independent answers first, round and depth caps, loop detection, one synthesis.
Before & after
Same prompt, same agent — the only difference is the skill.
Running multiple agents requires coordination to avoid conflicts and ensure they work together effectively. You need to define what each agent's role is and how they communicate with each other. Consider creating a shared context that all agents can access. Use standardized message formats so agents understand each other. Set up error handling for when an agent fails or gives conflicting advice. You might want to run agents sequentially or in parallel depending on the task. Testing with a few scenarios helps ensure the setup works correctly.
AGENT PROTOCOL SETUP: C-SUITE BOARD MEETING
COMPOSITION
Board meeting analysis: 3 agents
• CFO Agent (Financial analysis, modeling, risk)
• COO Agent (Operations, execution, timeline)
• CMO Agent (Go-to-market, revenue, brand)
INVOCATION SYNTAX
Step 1: Initialize multi-agent session
/board-sync --agenda "Q3 spending strategy" --date 2026-08-15
Triggers CFO, COO, CMO sequentially
Step 2: Each agent receives context block
agent_id: cfo-001
task: Analyze proposed Q3 budget of $2M, assess ROI assumptions
constraints: Max response time 5 min, JSON output, refer to CMO output
Step 3: Agent loops execute in order
Loop 1 - CFO first (financial foundation)
Input: Business context, Q3 targets
Output: Budget allocation, risk flags, assumptions
Loop 2 - COO after CFO
Input: CFO output, operational constraints
Output: Execution plan, resource needs, bottlenecks
Loop 3 - CMO after COO
Input: CFO budget, COO timeline
Output: Go-to-market plan, revenue model, risks
LOOP PREVENTION
Isolation rule: Agents read prior outputs but cannot re-invoke agents
Circular dependency check: If CMO asks for COO re-run, request rejected
Escalation: Human decides trade-offs
RESPONSE FORMAT (All agents follow)
json
{
agent_id: cfo-001,
timestamp: 2026-08-15T10:30:00Z,
task: Analyze Q3 budget and ROI,
recommendation: Approve $2M with conditions...,
assumptions_made: [Revenue 15% YoY, CAC stable $120],
dependencies_on_other_agents: [{agent: CMO, field: revenue_target}],
conflict_flags: [{agent: COO, issue: Timeline conflict}],
next_steps: [Human review, CMO can proceed]
}
CONFLICT RESOLUTION
Scenario A: CFO and CMO disagree on CAC budget
Identify in JSON conflict_flags
Capture both positions
Flag for human decision
Provide trade-off analysis
Scenario B: COO timeline infeasible for CMO
CFO notes dependency
CMO flags impossibility
Protocol stops; human resolves
Options: Extend timeline, increase resources, reduce scope
ERROR HANDLING
• Agent timeout (>5 min): Return partial analysis + flag
• Missing dependency: Retry once, escalate
• Conflicting facts: Log both, flag for human review
OUTPUT STRUCTURE (After all agents complete)
1. Executive summary (recommendation + key trade-offs)
2. Agent analyses (CFO, COO, CMO sections)
3. Conflict resolution (if any disputes)
4. Next steps (decision needed, follow-up actions)
5. Audit trail (who said what, when, assumptions)About this skill
Make a team of role agents, like CFO, CTO, and CMO advisors or specialist subagents, produce a better answer than one agent would, without looping or all agreeing. Or find out that one agent is the better choice.
What it does
- Tests whether the job really needs multiple agents before building it
- Defines non-overlapping role scopes and a fixed request and response format
- Collects independent answers before any sharing, then one cross-examination round
- Caps depth and rounds, routes through one orchestrator, and detects loops
- Keeps disagreements visible and ends with one recommendation and its deciding assumption
What you get
Either a run of the session with a clear synthesis, or a protocol spec you can implement in your agent framework, plus a failure-mode table for debugging.
Best for
Builders of agent systems and anyone running advisor-panel or board-meeting style sessions with AI roles.
SKILL.md preview
---
name: agent-protocol
… (sign up to view the full skill)More ai & agents skills
View all AI & Agents skills →New Skill Scaffolder
Create a new Claude Code skill correctly: right mechanism, valid name and folder, a triggering description, and a smoke test.
Skill Library Lint and Overlap Report
Lint every skill in a repo at once: frontmatter, name-folder match, weak descriptions, overlapping triggers, broken refs, plus a CI script.
Self-Verifying Agent Loop: Plan, Check, Retry, Close
Turns a goal into a task plan with machine-run checks, executes with retry caps, escalates, and refuses to close until verified.
Skill Test Runner (Trigger and A/B)
Test a skill in fresh sessions: ready-to-run prompts, commands, and rubric for trigger recall, false triggers, and blind A/B.