Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

Sign up
Appearance settings

choism4/mob-review

Folders and files

NameName
Last commit message
Last commit date

Latest commit

History

4 Commits

Repository files navigation

Mob Review

Five experts. Five perspectives. One deliverable. Zero blind spots.

Install

npx skills add choism4/mob-review

Usage

/mob-review

Run it after completing any work — code, paper, report, design doc. The skill analyzes your deliverable, assembles a mob of 5 domain experts, and runs them in parallel. Fixes are applied automatically, and the loop repeats until every expert approves.

What It Does

Mob Review assembles 5 virtual experts to mob your work from different angles simultaneously — catching issues that any single perspective would miss. Inspired by mob programming, where the whole team swarms a problem at once.

/mob-review
 │
 ▼
Phase 1: Analyze Deliverable
├── Code: git diff, identify languages/frameworks/domain
├── Docs: read files, identify type/audience/purpose
│
Phase 2: Assemble Expert Team (automatic)
├── Select 5 experts across different axes
├── Assign specific personas (name, company, experience)
│
Phase 3: Parallel Review ◄── THE CORE
├── Spawn 5 Agents simultaneously
├── Each reviews from their unique perspective
├── Returns structured JSON feedback
│
Phase 4: Synthesize + Auto-Fix
├── Consolidate feedback, prioritize by consensus
├── Auto-fix CRITICAL and MAJOR issues
├── Present summary table
│
Phase 5: Re-Review Loop (automatic)
├── Re-spawn 5 Agents with fix context
├── Repeat until all approve or 3 rounds reached
│
Phase 6: Final Report
└── Expert verdicts, fix summary, statistics

Expert Selection

Experts are chosen to maximize perspective diversity:

For code:

Axis Example
Architecture / Design System design veteran
Domain Expertise Field-specific practitioner
Code Quality Clean code advocate, OSS maintainer
Security / Reliability Security engineer, SRE
Performance / Optimization Low-level specialist
Testing / QA TDD practitioner

For documents:

Axis Example
Subject Expertise Scholar or practitioner
Methodology / Logic Research methods expert
Data / Numbers Statistician, fact-checker
Style / Readability Technical writer, editor
Target Audience Actual reader persona

Example Teams

  • NLE code → ex-Adobe Premiere engineer, Pixar graphics engineer, Netflix streaming architect, FFmpeg contributor, Apple FCP QA lead
  • NeurIPS paper → DeepMind senior researcher, NeurIPS PC veteran, Stanford stats professor, Meta FAIR engineer, former Nature editor
  • Strategy report → McKinsey partner, regional market specialist, Goldman Sachs analyst, The Economist editor, industry C-level

Issue Severity

Level Meaning Action
CRITICAL Must fix — bugs, security holes, factual errors Auto-fixed immediately
MAJOR Strongly recommend — design flaws, weak arguments Auto-fixed immediately
MINOR Improvement — naming, style, phrasing Fixed at coordinator's judgment
NITPICK Optional — personal preference Usually ignored

Termination

Condition Result
All 5 experts approve Success
0 CRITICAL/MAJOR issues Success
3 rounds reached Report remaining issues and stop
No new issues (repeats only) Stop

Deadlock Handling

  • Conflicting expert opinions → majority rules, or higher confidence score wins
  • Same issue flagged 2 rounds straight → escalated to user

Example Output

## Mob Review Complete
### Expert Team
| Expert | Specialty | Verdict |
|--------|----------|---------|
| Jun Park (ex-Cloudflare) | Security | ✅ APPROVED |
| Sujin Kim (ex-Stripe) | API Design | ✅ APPROVED |
| Alex Chen (ex-Google) | Performance | ✅ APPROVED |
| Maria Santos (ex-DataDog) | Observability | ✅ APPROVED |
| Tom Wright (ex-Vercel) | DX / Testing | ✅ APPROVED |
### Statistics
- Rounds: 2
- Issues found: 7 (CRITICAL 1, MAJOR 3, MINOR 3)
- Issues fixed: 6
- Not applied: 1 (NITPICK, style preference)

Philosophy

  • Parallel, not serial — 5 agents at once. No bias from earlier reviews.
  • Diversity > depth — Different axes catch different blind spots.
  • Specific personas > generic roles — "Ex-Cloudflare WAF lead" reviews differently than "security expert."
  • Conservative auto-fix — Clear bugs get fixed. Design calls go to you.
  • Bounded loops — 3 rounds max. No infinite review cycles.

License

MIT

About

Mob review skill for Claude Code. 5 domain experts mob your work in parallel, auto-apply fixes, and loop until all experts approve.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

Contributors

AltStyle によって変換されたページ (->オリジナル) /