---
name: multi-review
description: >
  Use MCO (Multi-CLI Orchestrator) to run parallel content review across multiple AI agents.
  Output: structured findings report with severity, consensus/divergence, and actionable fixes.

  USE WHEN: User wants to review/critique an existing draft or topic for quality,
  accuracy, or readability using multiple AI perspectives.
  Trigger phrases: 审查、review、critique、多角度审查、cross-check、质量检查。

  DON'T USE WHEN:
  - User wants to CREATE new content from a topic → use write-article instead.
  - User wants to deeply ANALYZE content structure (dao framework) → use dao-analysis instead.
  - User wants to IMPORT an external file → use import-draft instead.
  - User wants to see vault status → use my-world instead.
  - User says "写文章"、"生成草稿"、"分析为什么火" → not this skill.
---

# multi-review — Multi-Agent Content Review

Use MCO (Multi-CLI Orchestrator) to run parallel content review across multiple AI agents, combining perspectives for comprehensive feedback.

## 路由规则（Use when / Don't use when）

Use this skill when:
- Reviewing drafts before publication (`03-review/` stage)
- Evaluating topic quality and angles (`01-topics/` stage)
- Cross-checking technical accuracy in tech blogs
- Getting diverse feedback on content structure
- 用户说"审查我的草稿"、"review this"、"帮我检查质量"

Do NOT use this skill when:
- 用户想**写新文章**（从选题到草稿） → use `write-article`
- 用户想**深度分析**内容（道法术器框架） → use `dao-analysis`
- 用户想**导入**外部文件到 vault → use `import-draft`
- 用户想**加载 vault 上下文** → use `my-world`

Edge cases:
- "审查我的草稿" → **this skill**（审查 = multi-review）
- "分析这篇文章为什么火" → **not this skill**（深度分析 = dao-analysis）
- "帮我把草稿改好" → **this skill**（改好需要先审查发现问题）
- "写一篇新文章" → **not this skill**（创作 = write-article）

## Prerequisites

```bash
mco doctor
# Should show at least 2 providers READY
```

## Workflow

### Step 1: Identify Target Content

```bash
cd /home/kang/apps/content-forge/content-forge
obsidian files folder="02-drafts"
obsidian files folder="03-review"
```

### Step 2: Read Content to Review

```bash
obsidian read path="<file-path>"
```

### Step 3: Run Multi-Agent Review

Use MCO to get feedback from multiple agents:

```bash
mco review --repo /home/kang/apps/content-forge \
  --prompt "Review the content in <file-path> for:
1. Clarity and readability
2. Technical accuracy
3. Structure and flow
4. Missing information or gaps
5. Suggestions for improvement

Provide specific, actionable feedback." \
  --providers claude,codex \
  --target-paths content-forge/<file-path> \
  --synthesize \
  --format report
```

### Step 4: Synthesize Results

When `--synthesize` is enabled, MCO will:
1. Collect findings from all agents
2. Deduplicate similar issues
3. Track which agent found what (`detected_by`)
4. Generate a consensus/divergence summary

### Step 5: Apply Feedback

Review the synthesized output and decide which suggestions to incorporate. Update the content manually (Golden Rule: Agents Read, Humans Write).

## Claude Provider Configuration

MCO-Plus supports custom configuration for the Claude provider via `--provider-permissions-json`:

### Switch cc Environment

Use a different cc-managed API endpoint:

```bash
mco run --prompt "..." --providers claude \
  --provider-permissions-json '{"claude":{"cc_env":"bigmodel"}}'
```

Available cc environments (check with `cc ls`):
- `88` — 88code.org
- `bigmodel` — 智谱 BigModel
- `cclaude` — cclaude.codes
- `relay` — relay.nf.video
- `vibe` — vibe.codesuc.top

### Switch Model

Specify Claude model:

```bash
mco run --prompt "..." --providers claude \
  --provider-permissions-json '{"claude":{"model":"opus"}}'
```

Model options: `opus`, `sonnet`, `haiku`, or full model name (e.g., `claude-sonnet-4-6`)

### Combined Usage

```bash
mco review --repo /home/kang/apps/content-forge \
  --prompt "Review this content" \
  --providers claude,codex \
  --provider-permissions-json '{"claude":{"cc_env":"bigmodel","model":"sonnet"}}' \
  --target-paths content-forge/02-drafts/<file>.md \
  --synthesize
```

## Output Format

MCO returns structured findings:

```
=== Multi-Agent Review ===
Task: Review <file-path>
Providers: claude, codex

## Findings (deduplicated)

### Issue 1: [description]
- Severity: High/Medium/Low
- Detected by: claude, codex
- Location: [specific section]
- Suggestion: [actionable fix]

### Issue 2: [description]
...

## Consensus
- Points all agents agreed on

## Divergence
- Points where agents disagreed

## Summary
- Total issues: X
- Critical: Y
- Suggestions: Z
```

## Integration with Content Pipeline

| Stage | MCO Use Case |
|-------|--------------|
| `01-topics` | Evaluate topic quality, identify gaps |
| `02-drafts` | Pre-review feedback before human review |
| `03-review` | Parallel expert review |
| `04-published` | Post-mortem analysis for improvement |

## Example Prompts

### Draft Review
```
Review this draft for clarity, accuracy, and engagement.
Focus on: technical correctness, logical flow, reader value.
```

### Topic Evaluation
```
Evaluate this topic idea for:
1. Reader interest potential
2. Uniqueness vs existing content
3. Suggested angles to explore
```

### Technical Accuracy Check
```
Verify all technical claims in this content.
Flag any inaccuracies or outdated information.
Suggest corrections with sources.
```

## Configuration

Default providers: `claude,codex` (both ready on this system)

To add more providers:
- `gemini` — requires Gemini CLI installed
- `opencode` — requires OpenCode installed
- `qwen` — requires Qwen Code installed

## Reference

- MCO-Plus Fork: https://github.com/CSZHK/mco-plus
- Original MCO: https://github.com/mco-org/mco
- cc Skill: `~/.claude/skills/cc/` (Claude Multi-Environment Manager)