skills/ai-pair/examples/dev-team.md
wangyitong a65adcc2e5 Initial commit: merged, deduplicated, and vetted skill collection
Sources: extracted from two upstream archives (skill-repo, skills-main),
merged with the following policy:

- 15 broken symlinks (pointing to /Users/jameslee/.cc-switch/skills or
  ../../.agents/skills on a foreign machine) discarded
- 3 real name collisions with identical content (ai-pair, ifind-http-api,
  zhipu-websearch) kept as one copy
- Functional overlaps deduped keeping the strongest variant:
  - docx family: kept docx (official, full toolchain) + docx-cn
    (GB/T 9704 Chinese official-document constants),
    dropped docx_writer (no scripts, name collided with docx)
  - humanizer family: kept humanizer-zh (6 zh reference docs),
    dropped humanizer (en, redundant for CN workflow)
- Skills that only ran in a foreign environment removed:
  ablemind-ops, app-publish, hlb-design-system, openclaw-adj-skill,
  claude-driver
- alphapai excluded from this public repo because its SKILL.md hard-coded
  live credentials

Result: 25 skills, 572 files, ~7.5 MB.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-13 14:47:12 +08:00

82 lines
1.9 KiB
Markdown

# 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
```