skills/ai-pair/examples/dev-team.md

82 lines
1.9 KiB
Markdown
Raw Permalink Normal View History

# Dev Team Example
## Scenario: Review a Pull Request
You're working on a project and want multiple AI perspectives on your code changes.
### Start the team
```bash
/ai-pair dev-team my-web-app
```
Output:
```
Team ready.
Team: my-web-app-dev
Type: Dev Team
Members:
- developer: ready
- codex-reviewer: ready
- gemini-reviewer: ready
Awaiting your first task.
```
### Assign a task
```
Implement rate limiting for the /api/login endpoint.
Max 5 attempts per IP per 15 minutes.
```
The developer agent reads the codebase, implements the feature, and reports back.
### Review results
After you approve sending for review, both reviewers analyze in parallel:
```
## Codex Review
### WARNING
- Rate limit key uses only IP address. Behind a proxy, all users share one IP.
Consider combining IP + user-agent or using X-Forwarded-For.
- No cleanup mechanism for expired entries in the rate limit store.
### SUGGESTION
- Add a header (X-RateLimit-Remaining) so clients know their status.
### Summary
Functional implementation, but proxy scenario needs attention.
## Gemini Review
### Architecture Issues
- In-memory store won't work with multiple server instances.
Consider Redis or a shared store.
### Design Patterns
- Rate limiter is tightly coupled to the route handler.
Extract as middleware for reuse on other endpoints.
### Summary
Works for single-instance deployment. Needs abstraction for production scale.
```
Notice how Codex found a **security edge case** (proxy IP sharing) while Gemini found an **architecture limitation** (single-instance only). Different dimensions, both valuable.
### Decide
You can say:
- **"Revise"** — feedback gets sent to developer, loop continues
- **"Pass"** — move to next task or end team
- **"Revise, but only fix the proxy issue and Redis. Skip the middleware refactor for now."** — you control the scope
### End the session
```bash
/ai-pair team-stop
```