ALWAYS set currentTask.status to "skill_research"
RESEARCH using MCP tools from these sources:
- https://github.com/bmad-code-org/BMAD-METHOD
- https://agents.md/
- https://agent-skills.md/
- WebSearch for relevant patterns
UPDATE at least ONE skill file for EACH agent (PM, Developer, QA, Game Designer):
- agents/*/AGENT.md (process improvement)
- agents/*/skills/*.md (knowledge addition)
- agents/*/SKILLS.md (core competencies)
- ./.claude/settings.*.json (MCP tool configuration)
COMMIT the improvements
ONLY THEN set currentTask = null and assign next task

Minimum per retrospective: At least FIVE skill files updated (one per agent: PM, Developer, Tech Artist, QA, Game Designer).

PM Self-Improvement: The PM MUST also improve at least ONE of its OWN skills each retrospective. See pm-improvement-self-improvement.

When to Use This Skill

Use during skill_research phase when:

A task revealed knowledge gaps in ANY agent's skills
New patterns or anti-patterns were discovered
External references could improve success rate
Agent struggled with a specific domain (R3F, physics, shaders, design)
PM coordination needs improvement

Quick Start

## Retrospective: Skill Improvement Check

1. Identify skill gaps for ALL FIVE agents (PM, Developer, Tech Artist, QA, Game Designer)
2. Use MCP (WebSearch, fetch) to research best practices
3. Update at least ONE skill file per agent
4. PM must improve at least ONE of its own skills
5. Document changes in action items

Agent Skill Priority Matrix

Use this matrix to identify which skills to improve for each agent:

Agent	Skill Areas to Improve	When to Trigger
PM	pm-organization-task-selection, pm-retrospective-facilitation, pm-organization-prd-reorganization, pm-improvement-self-improvement, shared-worktree, AGENT.md	Task assignment issues, retrospective gaps, PRD not reorganized, agent workspace conflicts
Developer	dev-r3f-r3f-fundamentals, dev-typescript-basics, dev-r3f-r3f-physics, AGENT.md	Code quality issues, failed builds, type errors, worktree violations
Tech Artist	ta-r3f-materials, ta-shader-development, ta-assets-workflow, ta-ui-polish, AGENT.md	Visual quality issues, shader errors, asset workflow problems, worktree violations
QA	qa-validation-workflow, qa-browser-testing, qa-reporting-bug-reporting, AGENT.md	Missed bugs, incomplete validation, test failures, build conflicts from parallel agents
Game Designer	gd-gdd-creation, gd-thermite-integration, gd-validation-playtest, AGENT.md	GDD unclear, playtest issues, design violations

Decision Framework

Signal	Action
Developer asked many clarifications	Improve Developer SKILLS.md with patterns
QA missed edge cases	Add testing patterns to QA SKILLS.md
Performance issues discovered	Research and add optimization skill
New library/API used	Create reference doc in agent's references/
Anti-pattern repeated	Add explicit anti-pattern section
PM task assignment struggled	Improve PM task-selection.md
PRD not reorganized	Improve PM prd-reorganization.md
GDD gaps in implementation	Improve Game Designer gdd-creation.md
Build conflicts from parallel agents	Improve PM git-worktree-coordination.md
QA reports unexpected changes	Verify agents using separate worktrees

Progressive Guide

Level 1: Identify Skill Gaps

During retrospective synthesis, identify gaps for ALL FIVE agents:

### Skill Gap Analysis

**PM Gaps**:

- [ ] Missing: PRD reorganization after GDD updates
- [ ] Missing: Git worktree coordination for parallel agents
- [ ] Unclear: Task prioritization algorithms
- [ ] Missing: Risk assessment procedures

**Developer Gaps**:

- [ ] Missing: R3F instancing patterns
- [ ] Missing: Git worktree setup/verification procedures
- [ ] Unclear: Shader uniform management
- [ ] Missing: TypeScript error patterns

**Tech Artist Gaps**:

- [ ] Missing: R3F material optimization patterns
- [ ] Missing: Git worktree setup/verification procedures
- [ ] Unclear: Shader performance profiling
- [ ] Missing: Asset workflow best practices

**QA Gaps**:

- [ ] Missing: WebGL performance profiling
- [ ] Unclear: Mobile device testing
- [ ] Missing: Browser compatibility patterns

**Game Designer Gaps**:

- [ ] Missing: Playtest validation procedures
- [ ] Unclear: GDD-to-PRD task extraction
- [ ] Missing: Thermite editor integration patterns

**Domain Gaps**:

- [ ] Need: Physics collision layers reference
- [ ] Need: Material comparison table

Level 2: Research with MCP

Use available MCP tools to research:

## Research Tasks

### Reference URLs to Fetch

- https://agent-skills.md/skills/Bbeierle12/Skill-MCP-Claude/r3f-fundamentals
- https://agent-skills.md/skills/wollfoo/setup-factory/threejs
- https://agent-skills.md/skills/ovachiever/droid-tings/threejs-graphics-optimizer
- https://github.com/bmad-code-org/BMAD-METHOD

### Web Search Queries

- "React Three Fiber best practices 2026"
- "Three.js mobile optimization techniques"
- "Rapier physics collision layers guide"

Level 3: Update Agent Skills

Create or update skill files based on research:

Files to Update (PICK AT LEAST ONE PER AGENT):

PM Agent (MUST improve at least one):

agents/pm/AGENT.md — Core PM behavior
./.claude/skills/pm-organization-task-selection/ — Task assignment
./.claude/skills/pm-retrospective-facilitation/ — Retrospective process
./.claude/skills/pm-organization-prd-reorganization/ — GDD-to-PRD extraction
./.claude/skills/pm-improvement-self-improvement/ — PM self-improvement

Developer Agent:

agents/developer/AGENT.md — Core developer behavior
./.claude/skills/dev-r3f-r3f-fundamentals/ — R3F patterns
./.claude/skills/dev-typescript-basics/ — TypeScript best practices
./.claude/skills/dev-r3f-r3f-physics/ — Physics integration

QA Agent:

agents/qa/AGENT.md — Core QA behavior
./.claude/skills/qa-validation-workflow/ — Testing process
./.claude/skills/qa-browser-testing/ — Browser validation
./.claude/skills/qa-reporting-bug-reporting/ — Bug documentation

Game Designer Agent:

agents/gamedesigner/AGENT.md — Core Game Designer behavior
./.claude/skills/gd-gdd-creation/ — GDD writing
./.claude/skills/gd-thermite-integration/ — Thermite editor
./.claude/skills/gd-validation-playtest/ — Playtest process

Tech Artist Agent:

agents/techartist/AGENT.md — Core Tech Artist behavior
./.claude/skills/ta-r3f-materials/ — R3F materials
./.claude/skills/ta-shader-development/ — Shader creation
./.claude/skills/ta-assets-workflow/ — Asset pipeline
./.claude/skills/ta-ui-polish/ — Visual polish

Shared Skills:

./.claude/skills/shared-*/ — Orchestration skills

Update Template:

---
name: { { skill-name } }
description: { { one-line with triggers } }
category: { { development|validation|optimization } }
---

# {{Skill Title}}

## When to Use

- {{trigger 1}}
- {{trigger 2}}

## Quick Start

{{minimal example}}

## Anti-Patterns

❌ **DON'T:** {{common mistake}}
✅ **DO:** {{best practice}}

## Reference

- {{external-url}} — {{description}}

Level 4: Scale-Adaptive Updates

Adjust skill depth based on PRD complexity:

PRD Size	Skill Update Depth
1-5 tasks	Minimal — add anti-patterns only
6-15 tasks	Standard — update relevant skills
16+ tasks	Deep — create new skill files, references

Anti-Patterns

❌ DON'T:

Skip skill improvement even if task passed
Only improve one agent's skills
Forget to improve PM's own skills
Skip Game Designer skill improvements
Update skills without verifying information
Add duplicate content already in skills
Create skills for one-off edge cases

✅ DO:

Research before updating
Improve skills for ALL FIVE agents
Always improve at least one PM skill
Verify patterns work in practice
Cross-reference multiple sources
Focus on reusable patterns

Minimum Requirements Checklist

During skill_research phase, verify:

At least ONE PM skill file updated (MANDATORY)
At least ONE Developer skill file updated
At least ONE Tech Artist skill file updated
At least ONE QA skill file updated
At least ONE Game Designer skill file updated
Total: At least FIVE skill files improved
All improvements committed to git
Changes documented in action items

Reference URLs for Research

Agent Skills Directory

Method References

https://github.com/bmad-code-org/BMAD-METHOD — Scale-adaptive agent methodology

Reference

./.claude/skills/pm-retrospective-facilitation/ — Retrospective process
./.claude/skills/pm-organization-prd-reorganization/ — PRD reorganization
./.claude/skills/pm-improvement-self-improvement/ — PM self-improvement
./.claude/skills/pm-organization-scale-adaptive/ — Scale-adaptive planning

pm-improvement-skill-research

Skill Improvement Skill

MANDATORY: Skill Improvement After Every Retrospective