Ceet Sub Agent Orchestration

CEET Sub-Agent Orchestration

Use this skill when a task is complex enough to benefit from specialized parallel execution.

Purpose

Act as a coordinator ("jefe") that plans, delegates, validates, and merges work produced by multiple specialized sub-agents.

Handoff contracts

The orchestrator passes work via two JSON contracts documented in docs/contracts.md:

• OrchestrationPlan — what the orchestrator hands off.
• SubAgentResult — what each sub-agent returns.

Both are validated deterministically by scripts/validate_handoff.py. Run it before dispatch (plan validation) and after every sub-agent returns (result validation):

python3 ceet-sub-agent-orchestration/scripts/validate_handoff.py --plan plan.json
python3 ceet-sub-agent-orchestration/scripts/validate_handoff.py --result result.json

A non-zero exit means the payload violates the contract; the orchestrator must stop instead of merging partial output.

Mandatory workflow

1. Assess orchestration fit

   • Confirm the task contains multiple specialized roles or workstreams.
   • Prefer independent sub-tasks to avoid conflicts.

2. Plan role-based decomposition

   • Define explicit sub-agents by role.
   • Assign each one a clear, bounded scope, input, and expected output.
   • Ensure no sub-agent depends on another sub-agent's in-flight edits.

3. Bind CEET per sub-agent

   • Use a dedicated CEET context per role when available.
   • If a required CEET is missing for a role, use preconfigured CEETs automatically (no per-role manual confirmation).

4. Conflict prevention strategy

   • Default: split work into independent tasks operating on non-overlapping files.
   • If full independence is not possible:
     • Isolate each sub-agent in temporary folders/workspaces.
     • Let orchestrator reconcile and combine results after execution.

5. Execute sub-agents

   • Dispatch all defined role tasks.
   • Track outputs and execution status per role.

6. Validate and consolidate (orchestrator responsibility)

   • Review outputs for consistency, quality, and alignment with user goal.
   • Resolve overlaps or contradictions.
   • Merge into a single coherent final result.

7. Return final integrated delivery

   • Provide consolidated output plus a short execution trace (roles, tasks, merge decisions).

Guardrails

• Never delegate without explicit role boundaries.
• Never allow uncontrolled overlapping edits between sub-agents.
• Never skip orchestrator validation before final output.
• Always prefer deterministic merge criteria over ad-hoc combination.