Plans

Plans are collaborative documents for scoping and breaking down work before creating tickets.

What plans are for

Plans sit between an idea and actionable tickets. Use them to:

• Scope a feature before implementation
• Break a large initiative into discrete tickets
• Get AI feedback on your technical approach
• Extract structured tickets from a written plan

Creating a plan

Go to Project → Plans and click New Plan. Plans start as blank markdown documents. Write your thoughts, requirements, and technical approach.

AI suggestions

Click Suggest to get AI feedback on your plan. The AI:

• Reads your plan content and the project codebase
• Suggests additions, structure, and technical details
• Inserts suggestions as ghost text (dimmed) that you can accept or dismiss

The AI uses the project's configured model. It has access to the same tools as the agent (file reading, grep, directory listing, memory) so it can reference actual code in its suggestions.

Iterating

Edit your plan after receiving suggestions and trigger another suggestion round. The AI sees what changed since the last suggestion and builds on your edits. Each suggestion round is stateless — the AI does not remember previous suggestion conversations, but it can see the current state of the document.

Readiness check

Click Check Readiness to assess whether a plan is detailed enough to generate tickets from. The readiness check uses a fast, lightweight model to make a quick judgment. It errs on the side of ready — any plan with substantive content typically passes.

The check returns a boolean (ready/not ready) and a brief explanation.

Extracting tickets

Once a plan is ready, click Extract Tickets. The AI reads the plan and generates structured tickets with:

• Title
• Description
• Kind (bug, feature, task, chore)
• Priority
• Suggested order

Review the extracted tickets. Click Approve to create them as real tickets in the project's tracker. You can edit extracted tickets before approving.

Plan lifecycle

Plans have these states:

Auto-generated titles

When you first request suggestions on an untitled plan, the AI generates a title based on the content. This uses the fast model for speed.

Tips

• Start with a rough outline. The AI is good at fleshing out structure.
• Include technical constraints and requirements. The AI references your actual codebase.
• Use plans for features that span multiple tickets. Single-ticket work can go straight to a ticket.
• The extract step is not destructive. You can modify the plan and extract again.