[FEATURE]: Built-in "Ask" Agent for Codebase Discovery and Brainstorming

[iTiPo]

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

1. The Problem / Motivation

I am recently migrating to OpenCode from Cursor, and while I am really enjoying the OpenCode experience, there is one critical workflow feature I deeply miss: a dedicated "Ask" Agent.

I know OpenCode already has an excellent "Plan" agent, but I often find myself needing to deeply analyze the existing code and brainstorm approaches before I actually know what to ask the Plan agent to do. Currently, this initial discovery phase feels fragmented. I need a safe, read-only environment to explore the codebase so I can formulate a clear, accurate request for the Plan agent.

2. Feature Description

I propose implementing a new built-in primary agent called "Ask".

Unlike agents designed to mutate code or write step-by-step implementation plans, the "Ask" agent would be tailored strictly for read-only exploration, Q&A, and high-level project reasoning.

It establishes a robust, 3-step workflow: Ask (Discover) -> Plan (Outline) -> Build (Execute).

It would leverage full codebase context to help developers:

• Deep Dive: Ask complex questions about how different parts of the repository interact or where specific logic lives.
• Brainstorm: Discuss potential architectural decisions, design patterns, or refactoring strategies without immediately generating actionable steps.
• Pre-Plan Formulation: Build up context and understanding so the user can then switch to the "Plan" agent with a highly detailed and accurate prompt.

3. Key Use Cases

• Onboarding / Discovery: A new developer drops into a huge project and asks, "How does the authentication flow work in this app? Point me to the relevant files."
• Pre-Refactor Analysis: "I want to migrate UserService to use the new database schema. What files rely on UserService, and what edge cases should I be aware of before I construct a plan for this?"
• Formulating Requests: Using the Ask agent to trace a bug or understand a complex module, and then using that gathered intelligence to tell the Plan agent exactly what architectural changes need to be mapped out.

4. Proposed Solution & Contribution Offer

I would love to contribute this feature to OpenCode myself!

If this aligns with the OpenCode product strategy and roadmap, I am ready to implement the "Ask" agent and submit a Pull Request for review.

Before I begin drafting the code, I would love to get feedback from the maintainers on strategic alignment. Does a dedicated "Ask" agent fit smoothly into your vision alongside the existing "Plan" and coding agents?

Looking forward to hearing your thoughts and hopefully collaborating on this!

[iTiPo]

Just as a quick follow-up: I went ahead and created a custom "Ask" agent locally in OpenCode to test this exact workflow, and I've been using it for my daily tasks. It works incredibly well! Using it for codebase discovery and brainstorming before handing things off to the Plan agent has completely smoothed out my workflow.

Seeing how seamlessly it fits, it really convinced me that this should be a first-class, built-in agent out of the box.

I'm happy to share my custom agent prompts/configuration if it helps evaluate the idea, or just let me know if you're on board and I can start working on the PR for the core implementation!

[rakotomandimby]

@iTiPo , can you share your custom "Ask" agent definition?
I think the PR would be as simple as adding it into https://github.com/anomalyco/opencode/blob/0b6fd5f6123e51387d2f9fc5d04a1295f820d561/packages/opencode/src/agent/agent.ts

[iTiPo]

@rakotomandimby thanks for the idea! I opened pull request #22196

You can find the definition in the pull request.