researching_skill

Overview

This skill performs comprehensive, codebase-focused research by spawning parallel sub-agents and synthesizing their findings into structured research documents. It documents the codebase as it currently exists, producing traceable references to files and line numbers. The skill adapts interaction level based on an autonomy mode and integrates optional inline review tooling when available.

How this skill works

On invocation, the skill analyzes the research question, reads any directly referenced files first, and decomposes the work into concurrent subtasks. It launches specialized sub-agents (locator, analyzer, pattern finder, etc.), waits for all results, and synthesizes a single self-contained research document with file paths, line references, and metadata. If an inline file-review tool is available and the user opts in, it triggers review automation after document creation.

When to use it

• You need a detailed, factual map of how a code area currently works
• You want consolidated findings across many files with precise file:line references
• You need parallelized analysis to speed up large codebase investigations
• You want a persistent research document stored in a thoughts/research path
• You need a non-evaluative description of existing implementation only

Best practices

• Provide specific files or directories in your initial query so those are read first
• Set autonomy mode explicitly: Autopilot, Critical (default), or Verbose
• When not using Autopilot, answer clarification prompts via the AskUserQuestion tool
• Opt into the file-review integration if you want inline feedback hooks
• Expect the skill to avoid suggestions, critiques, or root-cause analysis unless requested

Example use cases

• Document the flow and responsibilities of a microservice by collecting references across src/ and config/
• Produce a research report listing where a particular type or interface is defined and used with line numbers
• Run a pattern scan to find all implementations of a shared utility and consolidate examples
• Create a dated research note that teams can reference for onboarding or historical context
• Kick off nested deep dives by spawning follow-up research tasks for unclear subsystems