React Expert Research Skill

Overview

This skill produces exhaustive documentation research on any React API or concept by searching authoritative sources (tests, source code, PRs, issues) rather than relying on LLM training knowledge.

<CRITICAL> **Skepticism Mandate:** You must be skeptical of your own knowledge. Claude is often trained on outdated or incorrect React patterns. Treat source material as the sole authority. If findings contradict your prior understanding, explicitly flag this discrepancy.

Red Flags - STOP if you catch yourself thinking:

"I know this API does X" → Find source evidence first
"Common pattern is Y" → Verify in test files
Generating example code → Must have source file reference </CRITICAL>

Invocation

/react-expert useTransition
/react-expert suspense boundaries
/react-expert startTransition

Sources (Priority Order)

React Repo Tests - Most authoritative for actual behavior
React Source Code - Warnings, errors, implementation details
Git History - Commit messages with context
GitHub PRs & Comments - Design rationale (via
```
gh
```
CLI)
GitHub Issues - Confusion/questions (facebook/react + reactjs/react.dev)
React Working Group - Design discussions for newer APIs
Flow Types - Source of truth for type signatures
TypeScript Types - Note discrepancies with Flow
Current react.dev docs - Baseline (not trusted as complete)

No web search - No Stack Overflow, blog posts, or web searches. GitHub API via

gh

CLI is allowed.

Workflow

Step 1: Setup React Repo

First, ensure the React repo is available locally:

bash

# Check if React repo exists, clone or update
if [ -d ".claude/react" ]; then
  cd .claude/react && git pull origin main
else
  git clone --depth=100 https://github.com/facebook/react.git .claude/react
fi

Get the current commit hash for the research document:

bash

cd .claude/react && git rev-parse --short HEAD

Step 2: Dispatch 6 Parallel Research Agents

Spawn these agents IN PARALLEL using the Task tool. Each agent receives the skepticism preamble:

"You are researching React's
<TOPIC>
. CRITICAL: Do NOT rely on your prior knowledge about this API. Your training may contain outdated or incorrect patterns. Only report what you find in the source files. If your findings contradict common understanding, explicitly highlight this discrepancy."

Agent	subagent_type	Focus	Instructions
test-explorer	Explore	Test files for usage patterns	Search `.claude/react/packages/*/src/__tests__/` for test files mentioning the topic. Extract actual usage examples WITH file paths and line numbers.
source-explorer	Explore	Warnings/errors in source	Search `.claude/react/packages/*/src/` for console.error, console.warn, and error messages mentioning the topic. Document trigger conditions.
git-historian	Explore	Commit messages	Run `git log --all --grep="<topic>" --oneline -50` in `.claude/react` . Read full commit messages for context.
pr-researcher	Explore	PRs introducing/modifying API	Run `gh pr list -R facebook/react --search "<topic>" --state all --limit 20` . Read key PR descriptions and comments.
issue-hunter	Explore	Issues showing confusion	Search issues in both `facebook/react` and `reactjs/react.dev` repos. Look for common questions and misunderstandings.
types-inspector	Explore	Flow + TypeScript signatures	Find Flow types in `.claude/react/packages//src/.js` (look for `@flow` annotations). Find TS types in `.claude/react/packages/*/index.d.ts` . Note discrepancies.

Step 3: Agent Prompts

Use these exact prompts when spawning agents:

test-explorer

You are researching React's <TOPIC>.

CRITICAL: Do NOT rely on your prior knowledge about this API. Your training may contain outdated or incorrect patterns. Only report what you find in the source files.

Your task: Find test files in .claude/react that demonstrate <TOPIC> usage.

1. Search for test files: Glob for `**/__tests__/**/*<topic>*` and `**/__tests__/**/*.js` then grep for <topic>
2. For each relevant test file, extract:
   - The test description (describe/it blocks)
   - The actual usage code
   - Any assertions about behavior
   - Edge cases being tested
3. Report findings with exact file paths and line numbers

Format your output as:
## Test File: <path>
### Test: "<test description>"
```javascript
<exact code from test>

Behavior: <what the test asserts>


#### source-explorer

You are researching React's <TOPIC>.

CRITICAL: Do NOT rely on your prior knowledge about this API. Only report what you find in the source files.

Your task: Find warnings, errors, and implementation details for <TOPIC>.

Search .claude/react/packages/*/src/ for:
- console.error mentions of <topic>
- console.warn mentions of <topic>
- Error messages mentioning <topic>
- The main implementation file
For each warning/error, document:
- The exact message text
- The condition that triggers it
- The source file and line number

Format your output as:

Warnings & Errors

Message	Trigger Condition	Source
"<exact message>"	<condition>	file:line

Implementation Notes

<key details from source code> ```

git-historian

You are researching React's <TOPIC>.

CRITICAL: Do NOT rely on your prior knowledge. Only report what you find in git history.

Your task: Find commit messages that explain <TOPIC> design decisions.

1. Run: cd .claude/react && git log --all --grep="<topic>" --oneline -50
2. For significant commits, read full message: git show <hash> --stat
3. Look for:
   - Initial introduction of the API
   - Bug fixes (reveal edge cases)
   - Behavior changes
   - Deprecation notices

Format your output as:
## Key Commits
### <short hash> - <subject>
**Date:** <date>
**Context:** <why this change was made>
**Impact:** <what behavior changed>

pr-researcher

You are researching React's <TOPIC>.

CRITICAL: Do NOT rely on your prior knowledge. Only report what you find in PRs.

Your task: Find PRs that introduced or modified <TOPIC>.

1. Run: gh pr list -R facebook/react --search "<topic>" --state all --limit 20 --json number,title,url
2. For promising PRs, read details: gh pr view <number> -R facebook/react
3. Look for:
   - The original RFC/motivation
   - Design discussions in comments
   - Alternative approaches considered
   - Breaking changes

Format your output as:
## Key PRs
### PR #<number>: <title>
**URL:** <url>
**Summary:** <what it introduced/changed>
**Design Rationale:** <why this approach>
**Discussion Highlights:** <key points from comments>

issue-hunter

You are researching React's <TOPIC>.

CRITICAL: Do NOT rely on your prior knowledge. Only report what you find in issues.

Your task: Find issues that reveal common confusion about <TOPIC>.

1. Search facebook/react: gh issue list -R facebook/react --search "<topic>" --state all --limit 20 --json number,title,url
2. Search reactjs/react.dev: gh issue list -R reactjs/react.dev --search "<topic>" --state all --limit 20 --json number,title,url
3. For each issue, identify:
   - What the user was confused about
   - What the resolution was
   - Any gotchas revealed

Format your output as:
## Common Confusion
### Issue #<number>: <title>
**Repo:** <facebook/react or reactjs/react.dev>
**Confusion:** <what they misunderstood>
**Resolution:** <correct understanding>
**Gotcha:** <if applicable>

types-inspector

You are researching React's <TOPIC>.

CRITICAL: Do NOT rely on your prior knowledge. Only report what you find in type definitions.

Your task: Find and compare Flow and TypeScript type signatures for <TOPIC>.

1. Flow types (source of truth): Search .claude/react/packages/*/src/*.js for @flow annotations related to <topic>
2. TypeScript types: Search .claude/react/packages/*/index.d.ts and @types/react
3. Compare and note any discrepancies

Format your output as:
## Flow Types (Source of Truth)
**File:** <path>
```flow
<exact type definition>

TypeScript Types

File: <path>

typescript

<exact type definition>

Discrepancies

<any differences between Flow and TS definitions> ```

Step 4: Synthesize Results

After all agents complete, combine their findings into a single research document.

DO NOT add information from your own knowledge. Only include what agents found in sources.

Step 5: Save Output

Write the final document to

.claude/research/<topic>.md

Replace spaces in topic with hyphens (e.g., "suspense boundaries" → "suspense-boundaries.md")

Output Document Template

markdown

# React Research: <topic>

> Generated by /react-expert on YYYY-MM-DD
> Sources: React repo (commit <hash>), N PRs, M issues

## Summary

[Brief summary based SOLELY on source findings, not prior knowledge]

## API Signature

### Flow Types (Source of Truth)

[From types-inspector agent]

### TypeScript Types

[From types-inspector agent]

### Discrepancies

[Any differences between Flow and TS]

## Usage Examples

### From Tests

[From test-explorer agent - with file:line references]

### From PRs/Issues

[Real-world patterns from discussions]

## Caveats & Gotchas

[Each with source link]

- **<gotcha>** - Source: <link>

## Warnings & Errors

| Message | Trigger Condition | Source File |
|---------|------------------|-------------|
[From source-explorer agent]

## Common Confusion

[From issue-hunter agent]

## Design Decisions

[From git-historian and pr-researcher agents]

## Source Links

### Commits
- <hash>: <description>

### Pull Requests
- PR #<number>: <title> - <url>

### Issues
- Issue #<number>: <title> - <url>

Common Mistakes to Avoid

Trusting prior knowledge - If you "know" something about the API, find the source evidence anyway
Generating example code - Every code example must come from an actual source file
Skipping agents - All 6 agents must run; each provides unique perspective
Summarizing without sources - Every claim needs a file:line or PR/issue reference
Using web search - No Stack Overflow, no blog posts, no social media

Verification Checklist

Before finalizing the research document:

React repo is at
```
.claude/react
```
with known commit hash
All 6 agents were spawned in parallel
Every code example has a source file reference
Warnings/errors table has source locations
No claims made without source evidence
Discrepancies between Flow/TS types documented
Source links section is complete

react-expert

NPX Install

Tags

SKILL.md Content