1/6
I gave myself an 11-person review team with one slash command.

/opc review — Security Engineer, PM, New User, DevOps, and more review your code in parallel.

Built it as a Claude Code skill. Zero dependencies. Just markdown files.

github.com/iamtouchskyer/opc

2/6
Honest test: I ran both on the same repo.

Single Claude prompt: 14 bugs found
OPC (3 agents): 9 bugs found

Claude won on code bugs. Variable shadowing, DRY violations, exit codes — more thorough in a single pass.

So why bother?

3/6
Because OPC caught 5 things Claude completely missed:

- New user runs "opc review" in terminal, confused it's not a CLI command
- README symlink breaks silently if you cd into the repo first
- Claude Code link goes to marketing page, not install docs

Perspective bugs. Not code bugs.

4/6
How it works:

11 markdown files, each defining a specialist role with anti-patterns ("don't flag missing auth on local tools")

2-5 agents run in parallel with different system prompts

A coordinator verifies facts, questions severity, dismisses false positives

No magic. Structured prompt management.

5/6
The anti-patterns are the key design choice.

Each role knows what NOT to flag. Security agent won't cry about "no auth" on a CLI tool. New-user agent won't demand tutorials in a dev tool.

This prevents the #1 problem with AI review: generic checklist filling.

6/6
npm install -g @touchskyer/opc

Then in Claude Code: /opc review

30 seconds. Zero deps. Finds bugs you wouldn't think to look for.

Star it if it catches something Claude alone wouldn't.
