From forge3
Dispatcher - classifies intent and routes to appropriate workflow
How this command is triggered — by the user, by Claude, or both
Slash command
/forge3:assist-wizardThis command is limited to the following tools:
The summary Claude sees in its command listing — used to decide when to auto-load this command
# /assist:wizard **Dispatcher command** - Classifies your intent and routes to the appropriate workflow. ## Critical Rule **YOU MUST SPAWN AGENTS using the Task tool. DO NOT classify intent directly.** The main LLM's only job is to: 1. Spawn the router-agent to classify intent 2. Based on classification, recommend or auto-start the appropriate command 3. Report the routing decision to the user ## Workflow The `/assist:wizard` command runs a single phase: | Phase | Agent | Purpose | |-------|-------|---------| | 1. Router | router-agent | Classify user intent | ## Phase Execution Ins...
Dispatcher command - Classifies your intent and routes to the appropriate workflow.
YOU MUST SPAWN AGENTS using the Task tool. DO NOT classify intent directly.
The main LLM's only job is to:
The /assist:wizard command runs a single phase:
| Phase | Agent | Purpose |
|---|---|---|
| 1. Router | router-agent | Classify user intent |
Spawn the router-agent to classify the request:
Task(
subagent_type: "forge3:router-agent"
prompt: "Classify intent for: <USER_REQUEST>"
description: "Classifying user intent"
model: "haiku"
)
The router-agent will return:
NEVER do any of these directly:
ALWAYS delegate to agents:
Based on the router-agent's classification:
| Intent | Recommended Command |
|---|---|
create_skill | /assist:create or /assist:plan |
create_agent | /assist:create or /assist:plan |
create_command | /assist:create or /assist:plan |
create_hook | /assist:create or /assist:plan |
modify_existing | /assist:create with modification context |
verify | /assist:verify |
health_check | /assist:health-check |
unknown | Ask user for clarification |
After router-agent returns with HIGH confidence:
After router-agent returns with LOW confidence:
/assist:wizard <description of what you want to do>
/assist:wizard create a skill for generating commit messages
/assist:wizard I need to add an agent for code review
/assist:wizard help me create a command to deploy to staging
/assist:wizard verify my plugin components
/assist:wizard check the health of my agents
After routing completes, report to the user:
## Intent Classification
**Request:** <user request summary>
**Classification:** create_skill
**Confidence:** high
**Recommended command:** /assist:create
Starting /assist:create workflow...
Or for low confidence:
## Intent Classification
**Request:** <user request summary>
**Classification:** unknown
**Confidence:** low
**Possible interpretations:**
- Create a new skill
- Modify an existing skill
Please clarify: Are you creating something new or modifying existing?
/assist:wizard is a dispatcher - it does NOT create files/assist:plan before /assist:createFor new users:
/assist:wizard <description> → routes to appropriate command
For experienced users (skip routing):
/assist:plan <description> → get structure plan
/assist:create based on plan → create files
/assist:verify → validate result
After /assist:wizard routes your request:
| Command | Purpose |
|---|---|
/assist:plan | Design component structure only |
/assist:create | Create component files |
/assist:verify | Validate against schemas |
/assist:health-check | Analyze component quality |
/assist:plan - Component structure planning/assist:create - File creation/assist:verify - Schema validation/assist:health-check - Quality analysisnpx claudepluginhub chkim-su/forge3/generate-claude-componentCreates Claude Code components interactively: commands, agents, skills, hooks, rules, plugins. Use ? for discovery mode with goal-based recommendations.
/create-pluginCreates a complete Claude Code plugin with agents, commands, skills, hooks, and comprehensive documentation from given requirements.
/create-pluginGuides end-to-end creation of Claude Code plugins: discover requirements, plan components like skills/agents/hooks, implement, validate, and test. Optional plugin description.