Features

The Features workspace is designed for structured feature ideation and specification. It helps teams turn high-level ideas into clearly defined, implementation-ready artifacts before any code is written.

Unlike a Trial, the Features workspace does not execute code, run agents against a repository, or modify files. It is a planning surface focused on clarity, structure, and alignment.

Use Features when you want to:

• Define new product functionality
• Break down requirements into user stories
• Draft acceptance criteria
• Identify edge cases
• Outline subtasks before development begins

It bridges the gap between idea and execution.

What Features Is (and Is Not)

Features is:

• A structured ideation environment
• A collaborative specification builder
• A way to generate well-formed feature documents
• A pre-development planning tool

Features is not:

• A code execution sandbox
• A repository-modifying workflow
• A Trial session
• A branch-based implementation environment

Once a feature is defined, it can be generated as a task and then implemented inside a Trial.

The Feature Ideation Interface

When you open Features, you land on the Feature Ideation screen.

The interface has a few key areas:

Feature/PRD docs from Context shows any context documents currently attached to the session, for example, a code index for your repository. You can click Open next to any listed document to view its contents in a panel on the left. This gives the AI grounding in your actual codebase when generating specifications.

Suggested ideas shows a list of feature ideas generated based on your repository context. These refresh each time you open the page or click Refresh. They are starting points, you can click one to use it as your prompt, or ignore them and describe something entirely different.

At the bottom, a text input lets you describe the feature you want to build in natural language. The model selector shows which model is active (e.g., Claude Sonnet 4.6), and the mode selector lets you switch between Feature Ideation, PRD Creator, and Tasks Planning depending on what kind of output you need.

Modes

The Features workspace supports three modes, selectable from the dropdown next to the model selector:

• Feature Ideation: generate feature descriptions, user stories, acceptance criteria, and edge cases from a natural language prompt

• PRD Creator: produce a more formal Product Requirements Document structure for larger initiatives

• Tasks Planning: focus on breaking a feature down into implementation-ready task lists

Choose the mode that matches what you need at your current stage of planning.

Context Documents

Clicking Context at the bottom of the chat panel opens the Project Context Documents selector.

This shows all available context documents for the project. You can check or uncheck documents to control what the AI reasons over when generating feature specifications. Including the code index gives the AI awareness of your existing architecture, which leads to more relevant and grounded output.

The number next to Feature/PRD docs from Context in the main panel reflects how many documents are currently active.

Starting Feature Ideation

To begin, describe what you want to build in the input at the bottom of the screen. Be as specific or as broad as you like, the AI will ask follow-up questions to refine scope if needed.

Example prompts:

• "Add a searchable command palette with keyboard shortcuts."
• "Implement role-based access control for the admin dashboard."
• "Create a CSV export feature for analytics reports."

The system will guide you through clarifying questions and structured options before generating a formal feature specification.

Guided Clarification

After the initial description, Features may ask targeted follow-up questions to refine scope. For example:

• What search behavior should be used? (fuzzy, prefix, exact)
• Should recent commands be supported?
• Should export functionality be included?

You can select predefined options, add custom inputs, or request alternative approaches. This ensures the resulting specification reflects deliberate product decisions rather than vague intent.

Generated Feature Specification

Once clarified, Features generates a structured specification that typically includes:

Description

A clear summary of what the feature does and why it exists.

User Stories

Structured stories describing expected behavior from different perspectives. For example:

• As a user, I want to open the command palette with a keyboard shortcut so I can navigate quickly.
• As a user, I want grouped search results so I understand what each result represents.

Acceptance Criteria

Concrete conditions used to determine when the feature is complete. For example:

• Opens with Cmd+K / Ctrl+K
• Closes on Escape
• Search auto-focuses on open
• Results update in real time

Edge Cases

Identified boundary conditions such as long labels, empty states, mobile fallback behavior, and keyboard shortcut conflicts.

Subtasks

Implementation-level breakdowns that convert the feature into development-ready units.

Saving and Acting on a Specification

Once a specification has been generated, three actions are available at the bottom of the document panel:

• Update in Context: if the feature was previously saved to context, this updates the existing context document with the latest version of the specification. This button appears in place of Save once a document has already been saved.

• Save as New: saves the current specification as a new context document. This is useful when you want to preserve multiple feature specs as separate reference documents that the AI can draw on in future sessions.

• Generate Tasks: converts the specification into a structured task in your project backlog. The task is created with the relevant title, description, and subtasks pulled from the specification. From there it can be assigned, labeled, prioritized, and implemented inside a Trial.

Feature vs Trial

Use Features to decide what to build. Use Trials to build it.

When to Use Features

Features is especially useful when:

• Scoping complex product work
• Aligning across frontend, backend, and infrastructure teams
• Breaking down large initiatives into clear deliverables
• Preparing tickets for sprint planning
• Exploring alternative implementation approaches

It prevents premature coding by enforcing clarity first.

Example Workflow

1. Navigate to Features
2. Attach relevant context documents if needed
3. Select the appropriate mode (Feature Ideation, PRD Creator, or Tasks Planning)
4. Describe the feature idea in the input
5. Answer clarification prompts
6. Review the generated specification
7. Save to context for future reference, or generate tasks directly
8. Implement inside a Trial