meta-cognition-parallel

Instead of sequential analysis, this skill launches three parallel subagents - one for each cognitive layer - then synthesizes their results.

```

User Question

│

▼

┌─────────────────────────────────────────────────────┐

│ meta-cognition-parallel │

│ (Coordinator) │

└─────────────────────────────────────────────────────┘

│

├─── Task(fork) ──► layer1-analyzer ──► L1 Result

│ (Language Mechanics)

│

├─── Task(fork) ──► layer2-analyzer ──► L2 Result

│ (Design Choices) ├── Parallel

│ │

└─── Task(fork) ──► layer3-analyzer ──► L3 Result

(Domain Constraints)

│

▼

┌─────────────────────────────────────────────────────┐

│ Cross-Layer Synthesis │

│ (In main context with all results) │

└─────────────────────────────────────────────────────┘

│

▼

Domain-Correct Architectural Solution

```

```

/meta-parallel

```

Example:

```

/meta-parallel 我的交易系统报 E0382 错误，应该用 clone 吗？

```

Step 1: Parse User Query

Extract from $ARGUMENTS:

• The original question
• Any code snippets
• Domain hints (trading, web, embedded, etc.)

Step 2: Launch Three Parallel Agents

CRITICAL: Launch all three Tasks in a SINGLE message to enable parallel execution.

```

Read agent files, then launch in parallel:

Task(

subagent_type: "general-purpose",

run_in_background: true,

prompt:

+ "\n\n## User Query\n" + $ARGUMENTS

)

Task(

subagent_type: "general-purpose",

run_in_background: true,

prompt:

+ "\n\n## User Query\n" + $ARGUMENTS

)

Task(

subagent_type: "general-purpose",

run_in_background: true,

prompt:

+ "\n\n## User Query\n" + $ARGUMENTS

)

```

Step 3: Collect Results

Wait for all three agents to complete. Each returns structured analysis.

Step 4: Cross-Layer Synthesis

With all three results, perform synthesis:

```markdown

Layer Results Summary

| Layer | Key Finding | Confidence |

|-------|-------------|------------|

| L1 (Mechanics) | [Summary] | [Level] |

| L2 (Design) | [Summary] | [Level] |

| L3 (Domain) | [Summary] | [Level] |

Cross-Layer Reasoning

1. L3 → L2: [How domain constraints affect design choice]
2. L2 → L1: [How design choice determines mechanism]
3. L1 ← L3: [Direct domain impact on language features]

Synthesized Recommendation

Problem: [Restated with full context]

Solution: [Domain-correct architectural solution]

Rationale:

• Domain requires: [L3 constraint]
• Design pattern: [L2 pattern]
• Mechanism: [L1 implementation]

Confidence Assessment

• Overall: HIGH | MEDIUM | LOW
• Limiting Factor: [Which layer had lowest confidence]

```

```markdown

# Three-Layer Meta-Cognition Analysis

> Query: [User's question]

---

[L1 agent result]

---

[L2 agent result]

---

[L3 agent result]

---

Reasoning Chain

```

L3 Domain: [Constraint]

↓ implies

L2 Design: [Pattern]

↓ implemented via

L1 Mechanism: [Feature]

```

Final Recommendation

Do: [Recommended approach]

Don't: [What to avoid]

Code Pattern:

```rust

// Recommended implementation

```

---

Analysis performed by meta-cognition-parallel v0.1.0 (experimental)

```

Test 1: Trading System E0382

```

/meta-parallel 交易系统报 E0382，trade record 被 move 了

```

Expected: L3 identifies FinTech constraints → L2 suggests shared immutable → L1 recommends Arc

Test 2: Web API Concurrency

```

/meta-parallel Web API 中多个 handler 需要共享数据库连接池

```

Expected: L3 identifies Web constraints → L2 suggests connection pooling → L1 recommends Arc

Test 3: CLI Tool Config

```

/meta-parallel CLI 工具如何处理配置文件和命令行参数的优先级

```

Expected: L3 identifies CLI constraints → L2 suggests config precedence pattern → L1 recommends builder pattern

• Subagent results are summarized, may lose detail
• Parallel execution depends on Claude Code version
• Cross-layer synthesis quality depends on result structure
• May have higher latency than sequential approach

This is experimental. Please report issues and suggestions to improve the three-layer parallel analysis approach.