| name | workflow-router |
| model | sonnet |
| description | Routes user goals to the appropriate specialist agent workflow with resource allocation confirmation. Use when: the user wants to start a task but has not specified a workflow, asks 'how should I approach this?', or needs orchestration across research → plan → build → fix stages. |
| category | agent-orchestration |
| triggers | ["how should I approach this","route this to the right workflow","orchestrate across research plan build fix","start a task but not sure which workflow"] |
| tier | 1 |
| agents | ["primary"] |
| tool_dependencies | ["meta_skill","file_system"] |
| inputs | [{"name":"goal","type":"string","description":"User's stated goal or task description — the router will classify and route it","required":false}] |
| outputs | [{"name":"routed_workflow","type":"string","description":"Confirmation of the agent(s) dispatched and their assigned tasks, with handoff paths"}] |
Workflow Router
Routes user goals to the appropriate specialist agent workflow with explicit resource allocation and confirmation before dispatch.
Description
Classifies an incoming goal into one of four categories (Research, Plan, Build, Fix), maps it to the appropriate specialist agent(s), confirms resource allocation with the user, and dispatches. Handles the Fix workflow as a two-stage process (investigate first, then implement). Preserves context across agent handoffs using shared handoff documents. This skill is the entry point for multi-agent orchestration when the user knows what they want to accomplish but not which agent or workflow to use.
When to Use
- User wants to start a non-trivial task without specifying a workflow
- User asks "how should I approach this?" or "what's the best way to tackle X?"
- A task requires multiple sequential agents (Research → Plan → Build)
- You need explicit resource allocation agreement before spawning agents
Workflow
Step 1: Classify the Goal
If the goal is not provided or is ambiguous, ask:
What's your primary goal for this task?
- Research — Understand/explore something (unfamiliar code, libraries, concepts)
- Plan — Design/architect a solution (implementation plans, problem decomposition)
- Build — Implement/code something (new features, components, implementation from a plan)
- Fix — Debug/fix an issue (investigate and resolve bugs, failing tests)
If intent is clear from context (e.g., "this test is failing" → Fix), infer without asking.
Step 2: Check for Existing Plans
ls thoughts/shared/plans/*.md 2>/dev/null || ls docs/plans/*.md 2>/dev/null
- Build goal: Ask if implementing an existing plan; load it to avoid duplication
- Plan goal: Surface existing plans to prevent redundant work
- Research / Fix: Proceed without plan check
Step 3: Confirm Resource Allocation
Present options and confirm before spawning:
How would you like to allocate resources?
- Conservative — 1-2 agents, sequential (minimal context, simple tasks)
- Balanced (recommended) — appropriate agents for the task, some parallelism
- Aggressive — max parallel agents (time-critical tasks)
- Auto — system decides based on complexity
Default to Balanced for unspecified or Auto selection.
Step 4: Map Goal to Agent
