skill-research-process

Skill Research Process

Systematic, scalable approach for building comprehensive Claude Code skills using parallel research agents. Use this when a skill requires extensive documentation gathering from official sources.

Process Overview

Stage 1: Initialize → Categorization agent creates TODO checklist
    ↓
Gate 1: Verify categories are distinct and complete
    ↓
Stage 2: Research → Parallel agents populate references/{category}/
    ↓
Gate 2: Anti-hallucination checkpoint (verify all claims cited)
    ↓
Stage 3: Integrate → Update SKILL.md, validate structure
    ↓
Gate 3: Final validation (links work, quality standards met)

Pre-Requisites

1. Activate skill-creator for structure guidance:

   Skill(skill: "plugin-creator:skill-creator")
   

2. Read CLAUDE.md for verification requirements

Stage 1: Initialize Skill Structure

Objective: Create base skill directory and identify documentation categories.

Steps

1. Initialize skill directory:

   plugins/plugin-creator/skills/skill-creator/scripts/init_skill.py <skill-name> --path <output-directory>
   

2. Launch categorization agent - see Agent Prompts

3. Output: {skill-name}.TODO.md with categorized checklist

Quality Gate 1: Category Verification

Before proceeding, verify:

• Categories are distinct (no overlap)
• Each category is specific enough to guide focused research
• 5-10 categories total (fewer for simple tools, more for complex)
• Categories cover the tool's full scope

If categories overlap: Merge or redefine boundaries before Stage 2.

Stage 2: Parallel Category Research

Objective: Launch concurrent research agents to build reference documentation.

Steps

1. Read TODO categories from {skill-name}.TODO.md
2. Launch concurrent Task agents (one per category) with run_in_background: true
3. Each agent outputs to ./references/{category}/