research-adhoc-feature

You can research a codebase by spawning parallel sub-agents that map components and fetch design context. This approach synthesizes findings into a self-contained research artifact for ad-hoc feature exploration.

The best way to map components for design context is using an orchestrated agent protocol. It coordinates parallel subtasks across the repository to capture visual baselines and document findings without producing implementation plans.

No, you cannot use this agent coordination approach to generate implementation plans. It follows a strict research-only protocol that writes results to thoughts/research, explicitly avoiding the creation of implementation plans.

Yes, parallel sub-agent research does store visual baselines and design context. It orchestrates tasks across the codebase to collect this information and saves the synthesized findings as a self-contained artifact in thoughts/research.

To start a guided research session for an ad-hoc feature, simply specify your target research question. The system will then orchestrate parallel sub-agents to analyze the codebase and document the findings.

A key limitation of using sub-agents for codebase research is the strict research-only protocol. It synthesizes findings into thoughts/research but does not produce actionable implementation plans for the identified components.