RPI Research Phase - Systematic codebase exploration with parallel agents and chunked output

To explore an unfamiliar codebase before feature development, this Skill systematically spawns 3-5 parallel agents to concurrently analyze different domains like API entry points, business logic, and database layers, outputting self-contained research chunks.

Analyzing API entry points and database layers concurrently is achieved by deploying parallel agents that divide the codebase exploration task, ensuring comprehensive domain coverage and generating structured, chunked output for research and planning phases.

Parallel agent codebase exploration works by assigning per-agent budgets to concurrently analyze specific domains such as external integrations and test coverage. It targets a total context budget of 50k tokens and compacts the final output into structured chunks.

Yes, you can use this codebase exploration output directly for planning phases. The Skill creates an inter-phase contract designed specifically for a subsequent planning phase, defining exactly how the generated research chunks will be processed sequentially.

To manage context limits when researching extensive external integrations, the parallel agents operate under strict per-agent budgets. The system targets a 50k token context budget and applies final output compaction to keep the research chunks concise.

You should use parallel agents instead of manual codebase exploration when starting new feature development and needing to quickly identify dependencies, map relevant code areas, and pinpoint testing zones across multiple domains like logic and integrations.