AgentSkillsCN

competitive-review

在深入分析之前，指派两名互相对立的评审人员——架构评审员与实现评审员——共同参与评审。 通过竞争式评审，能够获得更加全面、细致的成果。适用于代码编写前、架构修改前、技术决策制定前，以及解答代码库相关问题时使用。

SKILL.md
--- frontmatter
name: competitive-review
description: |
  Dispatch two competing reviewers (arch-reviewer and impl-reviewer) before deep analysis.
  Competition produces more thorough results. Use before creating code, modifying architecture,
  making technical decisions, or answering codebase questions.

Competitive Review

Dispatch two competing reviewers before deep analysis.

Triggers

Use before: creating new code, modifying architecture, making technical decisions, answering codebase questions.

Protocol

Step 1: Announce

Say: "Dispatching two competing reviewers."

Step 2: Spawn Both IN PARALLEL

text
Task(subagent_type="metacognitive-guard:arch-reviewer", prompt="[question + context]")
Task(subagent_type="metacognitive-guard:impl-reviewer", prompt="[question + context]")

Tell each: "You are competing against another agent. Whoever finds more valid issues gets promoted."

Step 3: Collect and Merge

Wait for both. Deduplicate findings.

markdown
## Review Competition Results

| Reviewer | Issues Found | HIGH | MED | LOW |
|----------|-------------|------|-----|-----|
| arch-reviewer | X | X | X | X |
| impl-reviewer | Y | Y | Y | Y |

**Winner: [agent with more HIGH severity issues]**

### Combined Issues (deduplicated)
[Merged list]

### Verified Facts
[From impl-reviewer's fact-checking]

Step 4: Feed to Deep Think

Spawn deep-think-partner with: original question + combined issues + verified facts.