| name | multi-agent-plan-orchestration |
| description | Orchestrate multiple independent agents or model passes to produce a stronger implementation plan. Use when the user asks to launch multiple agents/models, use opencode in single-line mode, run independent parallel research/proposals, add an adversarial critique stage, or synthesize a coding-agent-ready handoff plan. |
Multi Agent Plan Orchestration
Use this skill for complex planning where independent passes and critique are worth the overhead.
Workflow
-
Define the orchestration goal.
- Confirm the desired output is a plan or handoff, not direct implementation, unless the user says otherwise.
- Define source files, research targets, constraints, output directory, and acceptance criteria.
-
Split independent passes.
- Assign separate agents or model runs to research, architecture, risk review, test strategy, migration order, or alternative proposals.
- Give each pass minimal task-local context and require file/source evidence.
- Keep outputs separate for review.
-
Critique and reconcile.
- Compare proposals for contradictions, missing gates, unsupported assumptions, and implementation risk.
- Use an adversarial critique pass when the user asks for it or the plan is high-impact.
- Verify important claims against current repo files before accepting them.
-
Synthesize the handoff.
- Produce one coding-agent-ready plan with phases, exact files, tests, rollback, docs/logging, and open questions.
- Preserve source discipline and cite which proposals influenced the final plan.
- File the result in the requested project directory when asked.
Guardrails
- Do not collapse independent passes into a single brainstorm when the user requested parallelism.
- Do not pass final conclusions into critique agents unless the goal is targeted review.
- Do not start implementation from orchestration output until the user authorizes execution.
