arthai-marketplace

/wizard

Guided step-by-step workflows for non-engineers. Role-based flows that walk you through common tasks.

Synopsis

/wizard [role] [step]

When to use it

You’re a PM, founder, designer, or marketer and want to be walked through a full process (idea → plan → launch) step by step, with progress saved between sessions
You want to resume a multi-step process you started earlier, /wizard resume
Not for: a one-off choice menu, that’s /welcome; or a direct feature plan when you already know what you want, that’s /planning

Quickstart

/wizard

What you’ll see: your role is detected (or asked once), then the matching 5-step workflow overview, for a PM, “Plan and Ship a Feature”: describe your idea → understand your users → write the plan → create a launch plan → track progress. You pick a step; each one explains itself, asks before running, and saves its output.

Examples

/wizard               # detect role or ask, then show available workflows
/wizard pm            # show the PM workflow steps
/wizard pm 1          # jump straight to PM step 1 (Describe Your Idea)
/wizard pm next       # continue to the next incomplete step
/wizard founder       # Founder workflow: Build and Ship
/wizard designer      # Designer workflow: Design, Validate, Ship
/wizard marketer      # Marketer workflow: Research, Position, Launch
/wizard status        # progress table across workflows
/wizard resume        # resume where you left off
/wizard reset         # clear saved progress and start fresh

What it does

Detects your role, reads USER_ROLE from .claude/.claude-agents.conf (a plain-text config file in your project’s .claude folder, your Claude Code session can create or edit it for you); if unset, asks via AskUserQuestion (Product Manager / Founder / Designer / Marketer), a user-confirmation checkpoint. The answer is not automatically saved back to the config file, so you’ll be asked again next session unless you (or Claude) add USER_ROLE to it manually.
Presents the role’s 5-step workflow and lets you start anywhere (out-of-order is allowed, it warns but doesn’t block):
- PM. Plan and Ship a Feature: feature brief → user personas + assumptions to validate → implementation plan (via /planning) → launch plan → progress report from real git/PR/issue data
- Founder. Build and Ship: vision → competitive landscape (including how you’re positioned against similar products, using frameworks like Strategy Canvas / 7 Powers) → build plan (via /planning) → build (via /implement) → launch plan + status
- Designer. Design, Validate, Ship: design brief → expert critique → improved iteration → implementation plan → final review against the original intent
- Marketer. Research, Position, Launch: landscape analysis → positioning statement (how you describe the product vs. alternatives) → content drafts (blog, social, announcement, grounded in the real product) → launch plan → results + recommendations
Runs each step with the same protocol, loads previous steps’ outputs as context, explains what the step does in plain language, then asks “Ready to start? Or would you like to skip this step?”, a user-confirmation checkpoint, before bringing in the right specialist. Everything is in everyday language: no agent names, model names, or command syntax.
Saves state after every step to .claude/.wizard-state.json (role, workflow, current step, per-step summaries and file references), then offers: “next”, “revise” (redo this step with changes, overwriting its saved output, no history of prior attempts is kept), “share this”, “save and stop”, or “skip to step N”.
Resumes across sessions, /wizard resume shows completed steps with summaries and continues from where you stopped; /wizard status shows the progress table; /wizard reset deletes the saved state.

Output & artifacts

.claude/.wizard-state.json, saved progress (survives session restarts)
Step deliverables saved as files where applicable (e.g. .claude/plans/<topic>-brief.md; planning/implementation steps produce their own plan files), summaries are kept in state with file references
Briefs, personas, positioning, content drafts, launch plans, and status reports presented in the conversation

Troubleshooting

Problem	Fix
“No workflow in progress” on `/wizard resume`	There’s no saved state yet: run `/wizard` to start a workflow
It asked for my role again	Nothing writes your answer back to config: set `USER_ROLE` in `.claude/.claude-agents.conf` (a config file in your project’s `.claude` folder; your Claude Code session can edit it for you) if you don’t want to be asked each time
Want to redo a step	Run that step again (`/wizard pm 2`) or say “revise” at the checkpoint: this redoes the step and overwrites its output in state; no earlier version is kept
Progress feels stale or wrong	`/wizard reset` clears `.claude/.wizard-state.json` and starts fresh
A step failed mid-run	State was saved before the step ran: it explains what happened in plain English and offers to retry

/welcome, single-choice role menu (wizard is the full guided process)
/planning, the planning engine behind the “write the plan” steps
/implement, the build engine behind the founder workflow’s Step 4
/onboard, engineer-facing session briefing