Multi-Agent Coordination

ido4 works with single agents and multi-agent teams alike. The governance, context, and intelligence capabilities work the same whether you have one agent or ten.

For teams deploying multiple AI agents on the same codebase, ido4 adds coordination: task intelligence that recommends the highest-leverage work for each agent, exclusive locking that prevents duplicate effort, and atomic handoff that keeps work flowing when one agent finishes. Without this, multiple agents produce duplicate work, conflicting changes, and cascade failures.

Multi-Agent Coordination — architecture diagram

Agent Workflow — daily agent loop

Agent Registration

Each AI agent registers with a unique identity, role, and capability profile:

register_agent("agent-alpha", "Alpha", role: "coding", capabilities: ["backend", "data", "etl"])
register_agent("agent-beta", "Beta", role: "coding", capabilities: ["frontend", "auth", "security"])

Agent profiles are used for:

Audit trail — Every action is attributed to a specific agent
Task locking — Exclusive access to prevent conflicts
Work distribution — Capability-matched recommendations
Coordination — Awareness of what other agents are doing

Task Locking

Before an agent starts working on a task, it acquires an exclusive lock:

lock_task(#42, agentId: "agent-alpha")
→ Lock acquired. Expires in 30 minutes.

Locks prevent duplicate effort:

Only one agent can hold a lock on a task
Locks expire after 30 minutes (configurable)
Agents can heartbeat to extend locks
The TaskLockValidation BRE step warns when a task is locked by another agent

Intelligent Work Distribution

The get_next_task tool recommends the highest-leverage task for each agent. It scores candidates across 4 dimensions:

Cascade Value (0-40 points)

How many downstream tasks does completing this task unblock? A task that unblocks 3 others has higher cascade value than a task with no dependents.

#273 Data export → unblocks #274 Batch processing
cascade = 15 (1 dependent at depth 1)

Container Momentum (0-25 points)

Is the task’s grouping container (epic/bet) close to completion? Finishing in-progress epics or bets is more valuable than starting new ones.

Epic "Data Pipeline": 2 of 7 tasks done (28%)
momentum = 7

Capability Match (0-20 points)

Does the agent’s declared role and capabilities match the task?

agent-beta capabilities: [auth, security]
Task: "OAuth integration"
capability match = 13 (title contains "OAuth" → auth match)

Dependency Freshness (0-15 points)

Was a dependency recently completed? Tasks whose dependencies just finished have momentum — the context is fresh, the patterns are established.

Task #273 depends on #266 (Data ingestion)
#266 was approved 6 hours ago
freshness = 15 (within 24-hour window)

Scoring Example

get_next_task("agent-alpha")

#273 Data export       score:50  cascade:15 epic:7  cap:13 fresh:15
#274 Batch processing  score:17  cascade:0  epic:7  cap:10 fresh:0
#271 OAuth integration score:10  cascade:0  epic:0  cap:10 fresh:0

Every score is deterministic — computed from dependency graphs, epic completion ratios, agent capability vectors, and audit trail timestamps. No LLM reasoning involved.

Complete-and-Handoff

When an agent finishes a task, complete_and_handoff performs an atomic operation:

Approves the task through full BRE validation
Releases the agent’s lock
Identifies newly-unblocked downstream tasks
Recommends which agent should pick up each unblocked task
Suggests the completing agent’s next task

This eliminates the gap between completion and next assignment.

Coordination State

get_coordination_state provides the full multi-agent picture:

{
  "agents": [
    { "id": "agent-alpha", "status": "working", "currentTask": "#267" },
    { "id": "agent-beta", "status": "available", "currentTask": null }
  ],
  "recentEvents": [
    { "type": "task.transition", "issueNumber": 266, "transition": "approve", "ago": "6h" }
  ],
  "recommendations": {
    "agent-beta": { "task": "#273", "score": 47, "reasoning": "..." }
  }
}

Agents can poll this endpoint to stay aware of what other agents are doing and coordinate effectively.

Governance Hooks

Two PostToolUse hooks monitor governance impact:

State Transition Hook — After any task transition, checks: Did this unblock something? Did this create a new blocker? Is this a wave milestone?
Container Assignment Hook — After assigning a task to a container (wave/sprint/cycle), checks: Does this create an integrity issue? Does this introduce dependency problems?

Anti-Patterns ido4 Prevents

Anti-Pattern	How ido4 Prevents It
Two agents working on the same task	Task locking with exclusive access
Agent starts a task with incomplete dependencies	DependencyValidation BRE step
Agent bypasses code review	StatusTransitionValidation enforces the workflow
Agent ignores container integrity	ContainerIntegrityValidation blocks split assignments
No visibility into agent activity	Full audit trail with per-agent attribution
Agents working on low-value tasks	Work distribution recommends highest-leverage work