A full team in a single Claude Code skill. You're the CEO — OPC is everyone else.

11 specialist agents (PM, Designer, Security, Tester, and more) that review, analyze, build, and brainstorm your code — so you don't have to context-switch between hats.

You can — and for code-level bugs, a single Claude prompt is often more thorough. We tested both on this repo:

Single Claude found more code issues. OPC found different types of issues — things that require thinking from a specific persona's perspective. A security engineer looks for content exposure. A new user tries the install flow and gets confused. A DevOps engineer checks npm packaging. Claude won't switch to these mindsets unless you explicitly ask.

OPC's value isn't finding more bugs. It's finding bugs you wouldn't think to look for.

1. Adaptive triage — OPC picks from 4 modes (Review, Analysis, Execute, Brainstorm) based on what you're asking for.

2. Parallel specialists — 2-5 agents run in parallel, each with domain-specific expertise and anti-patterns (what NOT to flag). They don't see each other's output.

3. Coordinator verification — a coordinator checks agent findings: verifies facts, questions severity, dismisses false positives. You get a curated report, not a dump.

npm install -g @touchskyer/opc

Skill files are automatically copied to ~/.claude/skills/opc/. If the postinstall fails, run opc install manually.

git clone https://github.com/iamtouchskyer/opc.git
ln -s $(pwd)/opc ~/.claude/skills/opc

Note: Symlink tracks the repo — git pull updates the skill immediately. Use cp -r instead if you want a stable snapshot.

# Review a PR
/opc review the changes in this PR

# Analyze an architecture problem
/opc analyze why the API is slow

# Execute with a plan
/opc implement the migration plan in PLAN.md

# Brainstorm approaches
/opc what are our options for auth?

# Interactive mode — agents ask you questions first
/opc -i review the payment flow

# Explicit roles
/opc security compliance

# Open the report viewer
/opc replay

Add a .md file to roles/ following this format:

# Role Name

## Identity
One sentence: who you are and what you care about.

## Expertise
- **Area** — what you know about it
- **Area** — what you know about it
...

## When to Include
- Condition that triggers this role
- Condition that triggers this role
...

The coordinator reads When to Include to decide whether to dispatch your role. It's available immediately — no configuration needed.

If a task needs expertise not covered by any role file, the coordinator creates a temporary role on-the-fly.

You: /opc review this PR

1. Triage     → Mode A: Review
2. Roles      → Frontend, Backend, Security (auto-selected from changed files)
3. Brief      → Coordinator builds context from git log, CLAUDE.md, specs
4. Dispatch   → 3 agents run in parallel, each with role expertise + context
5. Verify     → Mechanical checks auto-reject incomplete outputs (no file:line, no VERDICT).
                 Coordinator spot-checks facts, challenges severity, deduplicates.
6. Report     → Curated findings with severity, file:line references, and fix suggestions

• Claude Code (CLI, desktop app, or IDE extension)
• Node.js ≥ 18 (for npm install only — not needed if you install manually)
• That's it. No runtime dependencies, no build step, no MCP server. Just markdown files.

OPC works standalone — but pair it with memex and it learns across sessions. Memex remembers which roles were useful, which findings were false positives, and your project-specific context. OPC doesn't need to know how memex works — memex drives itself.

npm install -g @touchskyer/memex

OPC saves structured reports to ~/.opc/reports/ after every run. Browse them in a web UI:

npx @touchskyer/opc-viewer

Or use /opc replay in Claude Code to open the viewer automatically.

The viewer shows a Slack-like replay of your review team's conversation, plus a filterable summary of findings by severity.

MIT