Skip to main content

AI Planner

The Planner turns plain-language feature descriptions into structured specifications through an AI-guided 6-step wizard. The output is a plan.md document that can be exported, edited, or converted directly into a ticket — without a developer in the loop.

When to use the Planner

  • A PM or stakeholder has an idea but no spec
  • A feature touches multiple areas and needs structured analysis before implementation
  • You want a paper trail of clarifying questions and trade-off decisions

The 6-step wizard

1. Configuration → 2. Understanding → 3. Clarifications →
4. Recommendations → 5. Specifications → 6. Review

1. Configuration

Provide the inputs the AI needs to ground the plan:

  • Workspace — which repo/codebase the plan targets
  • Provider & Model — AI provider for the planning session
  • Branch — base branch for the planning worktree
  • Description — your initial plain-language ask
  • Attachments — screenshots, mockups, prior specs

2. Understanding

The AI restates what it heard and asks clarifying questions to confirm scope.

3. Clarifications

Answer the AI's questions. Multiple rounds are supported until the AI is confident.

4. Recommendations

The AI proposes trade-off options (e.g., "store as JSON vs. dedicated table") and asks you to pick a direction. Multiple rounds supported.

5. Specifications

A structured document is generated, organized by ## headings for readability. Tabs view shows each section; full document view shows everything.

6. Review

Edit the spec in a Monaco editor with AI-assisted edits, undo/redo via version snapshots, and a Create Ticket from Plan button.

Auto-Answer modes

If you don't want to babysit the wizard, pick an auto-answer mode that lets the AI fill in clarifications and recommendations on your behalf:

ModeBehavior
NoneYou answer every question (default)
MVPLeans toward the leanest scope
BalancedLeans toward a balanced scope
Features-RichLeans toward the richest scope
CustomYou define a custom decision policy

Question throttling

To keep the wizard tractable, the Planner automatically asks fewer follow-up questions as the conversation matures, and stops asking once enough context has been gathered.

Plan status

StatusMeaning
DraftWizard in progress
CompletedSpec finalized
CanceledPlan discarded

Worktree isolation

Each plan runs in its own isolated Git worktree so the AI can read the codebase without disturbing other sessions. Concurrent edits to the same plan are prevented automatically.

Draft plans appear in the sidebar with a progress indicator showing wizard step completion. Hide individual plans you don't want cluttering the list.

Live updates

All connected clients stay in sync automatically — wizard step advances, status changes, and AI-assisted edits appear in real time without a manual refresh.

Exports and tickets

From the Review step:

  • Export Plan — download the spec as Markdown
  • Create Ticket from Plan — pre-fills a ticket with the plan as the description and links the plan as the source

Tickets created from a plan show a From a Plan source badge and back-reference the plan.

Attachments

Add reference materials at any step:

  • Screenshots
  • Existing docs / specs
  • Diagrams or mockups

These are passed as context to the planning AI.

Examples

Turn a stakeholder one-liner into a ticket A product owner writes "let users export their dashboard to PDF." Start a plan against the web app workspace, set auto-answer to Balanced, and let the wizard ask its clarifying questions (page size? include filters? which dashboards?). After the spec is generated, click Create Ticket from Plan — the developer picks up a fully-scoped ticket instead of a vague request.

Compare implementation approaches before committing A feature could store settings as JSON or in a dedicated table. Run a plan with auto-answer None so you answer each question yourself. At the Recommendations step the AI lays out the trade-offs of each option; you pick the direction and the rationale is captured in the spec — a paper trail for the decision.

Spec a feature from a mockup Attach a screenshot of the desired screen at the Configuration step. The planning AI grounds its questions and spec in the mockup, so the resulting plan.md reflects the intended UI rather than a guess.

Fast MVP scoping You want the leanest version shipped first. Choose auto-answer MVP so the AI leans toward the leanest scope, generate the spec, then create the ticket — minimal back-and-forth.

Permissions

PermissionCapability
planner.createStart a new plan
planner.view_allView plans created by other users
planner.manageEdit/delete any plan