Get from idea to working code in 5 minutes.
Ideate is a structured SDLC workflow for Claude Code. You describe what you want, and ideate plans it, builds it, reviews it, and refines it in cycles until it converges.
Workspace — your project directory (contains .ideate/)
└── Project — a strategic initiative with success criteria and appetite
└── Phase — a thin slice of work (3-6 items) with a specific goal
└── Cycle — one pass through execute → review → refine
└── Work Item — a single atomic task with acceptance criteria
| Ideate | Agile/Scrum | Notes |
|---|---|---|
| Project | Epic / Initiative | Has success criteria and an appetite (max phases) |
| Phase | Sprint | A time-boxed (or scope-boxed) slice of work |
| Cycle | Iteration | One execute → review → refine loop within a phase |
| Work Item | Story / Task | Atomic, has acceptance criteria, non-overlapping scope |
| Andon Cord | Sprint Impediment / Blocker | Issues that require human intervention |
| Capstone Review | Sprint Review + Retrospective | Multi-perspective evaluation after execution |
Ideate also introduces concepts without direct Scrum analogues. Guiding principles are high-level requirements that enforce alignment to the project's goals across all decisions. Domain policies, decisions, and questions borrow from Domain-Driven Design — they capture durable knowledge extracted from review cycles so later cycles get faster, not slower.
Key difference: ideate automates the sprint ceremony. The review is run by specialized agents (code-reviewer, spec-reviewer, gap-analyst). The retrospective produces domain policies automatically. You intervene only when the Andon cord is pulled.
Add ideate to your Claude Code project:
claude install-plugin /path/to/ideateOr manually add the MCP server to your .mcp.json:
{
"mcpServers": {
"ideate-artifact-server": {
"command": "sh",
"args": ["/path/to/ideate/mcp/artifact-server/start.sh"]
}
}
}The start.sh wrapper auto-installs dependencies and builds on first run.
For an existing codebase:
/ideate:init
Ideate surveys your code, asks a few questions, and produces an architecture document, work items, and an execution strategy.
For a new project (no code yet):
/ideate:init
Same command — ideate detects there's no existing code and runs a full planning interview with research.
/ideate:execute
Builds each work item, runs incremental code review after each one, and reports status.
/ideate:review
Runs a capstone review with three specialized agents (code quality, spec adherence, gap analysis) plus a journal keeper. Produces a summary with findings by severity.
If the review found issues:
/ideate:refine
Plans the fixes as new work items. Then /ideate:execute again. Repeat until convergence (zero critical/significant findings).
/ideate:autopilot
Runs the full loop autonomously: execute → review → refine → repeat until convergence or cycle limit.
| Command | What it does |
|---|---|
/ideate:init |
Initialize — survey codebase, interview, produce plan |
/ideate:execute |
Build work items with per-item review |
/ideate:review |
Multi-perspective capstone review |
/ideate:refine |
Plan the next cycle of changes |
/ideate:autopilot |
Autonomous execute → review → refine loop |
/ideate:project |
Manage projects and phases — create, view, switch, pause, complete, archive |
/ideate:triage |
Quick work item intake — bugs, features, chores |
/ideate:status |
Project status views — workspace, project, or phase |
/ideate:settings |
Configure agent budgets, models, spawn mode |
-
Be specific in the init interview. The more precise your answers, the better the plan. Vague answers produce vague specs.
-
Start with
/ideate:init, not/ideate:refine. Init handles both new and existing codebases. Refine is for changing an existing plan. -
Let autopilot run. For well-specified work,
/ideate:autopilothandles everything. You only intervene when the Andon cord fires. -
Phase your work. Large projects benefit from thin-slice phases (3-6 items each). Ideate auto-chunks when you have more than 5 work items.
-
Check the review findings. Critical and significant findings need attention. Minor findings are auto-fixed. Suggestions are informational.
-
Use
/ideate:settingsto tune. Adjust agent budgets if agents run out of turns. Switch to teammate mode for better visibility.
All artifacts live in .ideate/ in your project root:
- Work items — atomic tasks with acceptance criteria
- Guiding principles — durable rules governing decisions
- Domain policies — specific rules from review findings
- Cycle summaries — review results per cycle
- Journal entries — chronological project history
- Metrics — token usage and agent performance data
- GUIDE.md — Complete reference for all 9 skills, projects, phases, and the SDLC loop
- README.md — Full documentation with worked examples
- ARCHITECTURE.md — Technical reference for the MCP server and agent system
/ideate:settings— Configure agent budgets and spawn mode