research_skill

Overview

This skill orchestrates the intelligence-gathering phase by spawning parallel sub-agents to analyze a codebase, git history, and web resources. It produces or updates a task file at ./apex/tasks/[ID].md and provides a consolidated research section to feed planning and implementation. Use it as the first step to create a dependable knowledge foundation before changes are designed or coded.

How this skill works

It parses the input (text, ticket ID, file path, or existing task ID), clarifies vague requirements, and runs a low-cost triage scan to detect ambiguity. If the scope is clear, it launches parallel agents that extract implementation patterns, mine git history, search docs and the web, trace system dependencies, and surface risks. All findings are synthesized into an intelligence report and appended to the task file’s research section.

When to use it

• Starting a new feature, bugfix, refactor, or test where code and history matter
• You need a reproducible research artifact for planning or handoff
• Requirements are vague and must be clarified before design work
• You want repeatable pattern extraction and git-based context for a ticket
• Prior to architecting changes that touch multiple subsystems

Best practices

• Provide a concise task description, ticket ID, or path to the task file to ensure files are read fully before analysis
• Answer any ambiguity questions promptly—only one clarification round is allowed before deep agents spawn
• Specify performance, security, compatibility, or acceptance criteria to avoid blocking ambiguity checks
• Limit mentioning extremely large files unless necessary; the triage scan avoids opening huge files by default
• Treat the generated research file as living documentation and update it if requirements change

Example use cases

• Implement a new UI feature: supply the ticket and file path so agents map entrypoints, patterns, and tests
• Investigate intermittent production regressions: run git-historian and systems tracing to find recent related commits and owners
• Plan a cross-service change: surface integration points, dependencies, and risk mitigations before design
• Standardize a coding pattern: extract implementation snippets and tests used across the repo for reuse
• Convert a vague request into a testable enhanced prompt and acceptance criteria for engineering work