Oma Coordination

Multi-Agent Workflow Guide

Scheduling

Goal

Guide manual multi-agent coordination for complex work that spans PM, frontend, backend, mobile, and QA responsibilities.

Intent signature

• User wants step-by-step coordination, manual agent spawning, or multi-domain work planning without full automation.
• Task spans multiple specialist agents and requires contract alignment.

When to use

• Complex feature spanning multiple domains (full-stack, mobile)
• Coordination needed between frontend, backend, mobile, and QA
• User wants step-by-step guidance for multi-agent coordination

When NOT to use

• Simple single-domain task -> use the specific agent directly
• User wants automated execution -> use orchestrator
• Quick bug fixes or minor changes

Expected inputs

• Complex feature or project goal
• Required domains and priority tiers
• Workspace/session constraints and API/data contract needs

Expected outputs

• Manual coordination sequence
• PM task decomposition, agent spawn order, monitoring guidance, and QA review step
• API/data contract alignment checkpoints

Dependencies

• PM, frontend, backend, mobile, QA, and orchestrator skills
• CLI oma agent spawn and progress/result memory conventions

Control-flow features

• Branches by task complexity, priority tiers, dependency ordering, and whether automation is desired
• Spawns independent same-priority tasks in parallel when appropriate
• Monitors progress files and contract alignment

Structural Flow

Entry

1. Confirm the task is complex enough for multi-agent coordination.
2. Start with PM task decomposition.
3. Identify priority tiers and shared contracts.

Scenes

1. PREPARE: Define session, domains, and task decomposition needs.
2. ACT: Spawn agents by priority with separate workspaces.
3. VERIFY: Monitor progress and API/data contract alignment.
4. FINALIZE: Run QA review and coordinate remediation.

Transitions

• If task is simple, route to one specialist.
• If user wants automated execution, use orchestrator.
• If QA finds CRITICAL issues, re-spawn responsible agents.

Failure and recovery

• If contracts diverge, pause downstream frontend/mobile work until backend/API contract is reconciled.
• If agent workspaces conflict, split ownership boundaries.
• If progress stalls, inspect progress files and reissue focused instructions.

Exit

• Success: specialist outputs are coordinated and QA-reviewed.
• Partial success: blocked agents, contract conflicts, or QA failures are explicit.

Logical Operations

Actions

Tools and instruments

• oma agent spawn, PM/frontend/backend/mobile/QA agents
• Memory/progress/result files
• Configured code intelligence, with native fallback per ../_shared/core/code-intelligence.md

Canonical command path

bashCopy code
oma agent spawn pm <pm-prompt-file> <session-id> --task-id <plan.pm_task.id> -w ./pm
oma agent spawn backend <backend-prompt-file> <session-id> --task-id <plan.backend_task.id> -w ./backend &
oma agent spawn frontend <frontend-prompt-file> <session-id> --task-id <plan.frontend_task.id> -w ./frontend &
wait

When native runtime dispatch is available (per-agent target vendor equals the current runtime vendor), prefer the runtime's native subagent path and use oma agent spawn as the cross-vendor fallback — same resolution rule as oma-orchestration.

Useful agent spawn options: -m/--model <vendor> (CLI vendor override), --isolation worktree (git worktree per spawn, prevents file conflicts), --read-only (non-destructive tools only, e.g. for review/QA passes).

Resource scope

Preconditions

• Task requires multiple domains.
• PM decomposition can identify independent priority tiers.

Effects and side effects

• Spawns or guides multiple agents.
• Coordinates workspace ownership and QA feedback.

Guardrails

1. Always start with PM Agent for task decomposition
2. Spawn independent tasks in parallel (same priority tier)
3. Define API contracts before frontend/mobile tasks
4. QA review is always the final step
5. Assign separate workspaces to avoid file conflicts (or use --isolation worktree for a git worktree per spawn)
6. Follow ../_shared/core/code-intelligence.md: discover configured tools, do not auto-install or track, and use native scoped search when unavailable or timed out
7. Never skip steps in the workflow; follow each step sequentially without omission

Workflow

Step 1: Plan with PM Agent

PM Agent analyzes requirements, selects tech stack, creates task breakdown with priorities.

Step 2: Spawn Agents by Priority

Resolve the dispatch path per agent, then spawn:

1. Resolve the per-agent target vendor from oma-config.yaml (agents: override, else model_preset)
2. If the target vendor equals the current runtime vendor and a native subagent path exists, use native dispatch
3. Otherwise use oma agent spawn for that agent
4. Spawn all same-priority tasks in parallel using background processes

bashCopy code
# Example: spawn backend and frontend in parallel
oma agent spawn backend backend-prompt.md session-id --task-id plan.backend_task.id -w ./backend &
oma agent spawn frontend frontend-prompt.md session-id --task-id plan.frontend_task.id -w ./frontend &
wait

Step 3: Monitor & Coordinate

• Use memory read tool to poll progress-{agent}[-{sessionId}].md files (spawned agents write the session-suffixed form)
• Verify API contracts align between agents
• Ensure shared data models are consistent

Step 4: QA Review

Spawn QA Agent last to review all deliverables. Address CRITICAL issues by re-spawning agents.

Automated Alternative

For fully automated execution without manual spawning, use the orchestrator skill instead.

References