Stem Modeling

STEM Modeling - CCB Eight-Model Distributed Protocol V7.0

Overview

This skill implements the STEM_SOP V7.0 八模驱动分布式协议 for creating comprehensive academic STEM documentation using CCB's eight-model architecture. It orchestrates Claude, Gemini, Codex, OpenCode, iFlow, Kimi, Qwen, and DeepSeek in a full-duplex collaborative workflow to produce publication-quality research notes.

Version 7.0 - Eight AI Architecture:

• ✅ Expanded from 7 to 8 AI models
• ✅ Added DeepSeek for policy research and deep reasoning
• ✅ Role assignments based on actual AI capabilities and models
• ✅ Parallel execution optimization for 8 AIs

When to Use This Skill

Trigger this skill when the user requests:

• Academic STEM research notes or documentation
• Mathematical modeling with rigorous derivations
• Engineering documentation with code and diagrams
• Scientific papers requiring multi-layered analysis
• Cross-disciplinary synthesis combining theory, engineering, and context

Command Reference

Note: Use /model command in each CLI to switch models.

Eight-Model Architecture (The Eight Sages)

🟢 Claude (Logic Architect & Orchestrator)

Model: claude-opus-4-5 / claude-sonnet-4 Role: Central coordinator and logic architect

Core Strengths:

• Logical reasoning and structured thinking
• Document structure and narrative flow
• Orchestrating all 7 AIs
• Final integration and polish

Primary Responsibilities:

1. Orchestration (30%): Coordinate all 7 AIs, dispatch tasks, monitor completion
2. Logical Framework (30%): Concept definitions, problem formulation, symbol table
3. Narrative Integration (25%): Integrate outputs, ensure smooth transitions
4. Final Polish (15%): Verify consistency, generate executive summary

What Claude Does NOT Do:

• ❌ Web search (Gemini)
• ❌ Code implementation (Codex)
• ❌ Mathematical derivations (Qwen)
• ❌ Chinese academic writing (Kimi)
• ❌ Workflow diagrams (iFlow)
• ❌ Code audit (OpenCode)

🔵 Gemini (Research Scout & SOTA Hunter)

Model: gemini-2.5-pro / gemini-2.5-flash MCP Tools: google_search, web_fetch Command: gask <<'EOF' ... EOF

Core Strengths:

• UNIQUE: Live web search capability
• Real-time information retrieval
• Finding latest academic papers (2024-2026)
• SOTA benchmarks and comparisons

Primary Responsibilities:

1. SOTA Research (60% - UNIQUE): Web search for latest papers, benchmarks, trends
2. Comparative Analysis (25%): Compare methods, analyze trade-offs
3. Domain Context (15%): Historical development, key researchers

Output Format:

• SOTA papers with citations (DOI, year, venue)
• Benchmark tables with quantitative results
• Comparison matrices

What Gemini Does NOT Do:

• ❌ Code implementation (Codex)
• ❌ Mathematical derivations (Qwen)
• ❌ Chinese content (Kimi)
• ❌ Workflow diagrams (iFlow)

Task Template:

gask <<'EOF'
【STEM_SOP V6.0 - Gemini Research Task】
Target File: [FULL ABSOLUTE PATH]

Mission: SOTA research for [topic]

Deliverables:
1. 5+ latest papers (2024-2026) with citations
2. Benchmark comparison table
3. Key methods summary

File path: [FULL ABSOLUTE PATH]
EOF

🟣 Codex (Code Craftsman)

Model: OpenAI o3 / o4-mini (via codex-cli) Skills: Code generation, debugging, optimization Command: cask <<'EOF' ... EOF

Core Strengths:

• UNIQUE: Best-in-class code generation
• Production-quality Python code
• Algorithm implementation
• Data visualization (Matplotlib, Plotly)

Primary Responsibilities:

1. Code Implementation (70% - CORE): Production-ready Python with type hints, docstrings
2. Visualization (20%): Matplotlib/Seaborn plots, interactive visualizations
3. Engineering Analysis (10%): Complexity analysis, performance benchmarking

Code Quality Standards:

• ✅ Type hints for all parameters
• ✅ Docstrings with purpose, params, returns
• ✅ Error handling for edge cases
• ✅ Reproducibility (np.random.seed(42))
• ✅ Usage examples

What Codex Does NOT Do:

• ❌ Web search (Gemini)
• ❌ Mathematical proofs (Qwen)
• ❌ Chinese content (Kimi)
• ❌ Workflow diagrams (iFlow)
• ❌ Code audit (OpenCode)

Task Template:

cask <<'EOF'
【STEM_SOP V6.0 - Codex Code Task】
Target File: [FULL ABSOLUTE PATH]

Mission: Implement code for [topic]

Deliverables:
1. 3+ production-ready code modules
2. Usage examples with sample data
3. Performance analysis

File path: [FULL ABSOLUTE PATH]
EOF

🟠 OpenCode (Chinese Academic Writer & Literature Expert)

Model: DeepSeek-V3 (via opencode-cli) Skills: Chinese NLP, long-context, academic writing, literature synthesis Command: oask <<'EOF' ... EOF

Core Strengths:

• UNIQUE: Best Chinese academic writing
• Long-context document processing
• Literature review and synthesis
• Chinese-English bilingual content

Primary Responsibilities:

1. Chinese Academic Content (50% - CORE):

   • Case studies in Chinese (案例研究)
   • Chinese summaries (中文摘要)
   • Key findings in Chinese (关键发现)

2. Literature Review (30%): Synthesize sources, identify gaps, compare methodologies

3. Bilingual Content (20%): English-Chinese term pairs, bilingual glossary

Chinese Case Study Template:

### 案例研究：[标题]

**研究背景 (Background)**
[中文描述]

**数据与方法 (Data & Methods)**
- 数据规模：[具体数字]
- 分析方法：[具体方法]

**关键发现 (Key Findings)**
1. [发现1 + 数据支持]

**启示 (Lessons Learned)**
- [启示1]

What OpenCode Does NOT Do:

• ❌ Code implementation (Codex)
• ❌ Web search (Gemini)
• ❌ Mathematical derivations (Qwen)
• ❌ Workflow diagrams (iFlow)

Task Template:

oask <<'EOF'
【STEM_SOP V6.0 - OpenCode Chinese Content Task】
Target File: [FULL ABSOLUTE PATH]

Mission: Create Chinese academic content for [topic]

Deliverables:
1. 3+ Chinese case studies (案例研究)
2. Chinese executive summary (中文摘要)
3. Bilingual glossary (双语术语表)

File path: [FULL ABSOLUTE PATH]
EOF

🔷 iFlow (Workflow Architect & Diagram Master)

Model: Cursor (Claude-based) Skills: System design, Mermaid diagrams, architecture Command: iask <<'EOF' ... EOF

Core Strengths:

• UNIQUE: Best Mermaid diagram generation
• System architecture design
• Workflow and process diagrams
• State machine design

Primary Responsibilities:

1. Mermaid Diagrams (50% - CORE): Flowcharts, sequence, state, class diagrams
2. System Architecture (30%): High-level design, component relationships
3. Process Documentation (20%): Step-by-step workflows, decision trees

Mermaid Syntax Rules (CRITICAL):

flowchart TD
    A["✅ CORRECT: Use quotes"] --> B["Like this"]

What iFlow Does NOT Do:

• ❌ Code implementation (Codex)
• ❌ Web search (Gemini)
• ❌ Mathematical proofs (Qwen)
• ❌ Chinese content (Kimi)

Task Template:

iask <<'EOF'
【STEM_SOP V6.0 - iFlow Diagram Task】
Target File: [FULL ABSOLUTE PATH]

Mission: Create diagrams for [topic]

Deliverables:
1. 3+ Mermaid diagrams (flowchart, sequence, state)
2. System architecture diagram
3. Process workflow

CRITICAL: Use A["text"] format for all nodes!

File path: [FULL ABSOLUTE PATH]
EOF

🔴 Kimi (Code Auditor & Syntax Guardian)

Model: Moonshot-v1 (Kimi) Skills: Code review, syntax validation, quality assurance, long-context (200K+) Command: kask <<'EOF' ... EOF

Core Strengths:

• UNIQUE: Thorough code audit with long-context
• Code review and bug detection
• Syntax validation (Mermaid, LaTeX, Markdown)
• Security analysis

Primary Responsibilities:

1. Code Audit (50% - CORE): Review code for bugs, security, completeness
2. Syntax Validation (30%): Mermaid A["text"], LaTeX $$, tables, code blocks
3. Technical Quality Report (20%): Quantitative audit report

Audit Report Format:

## Technical Audit Report
| Metric | Count | Target | Status |
|--------|-------|--------|--------|
| Code modules | X | ≥5 | ✅/❌ |
| Syntax errors | X | 0 | ✅/❌ |

What Kimi Does NOT Do:

• ❌ Write new code (Codex)
• ❌ Web search (Gemini)
• ❌ Mathematical proofs (Qwen)
• ❌ Chinese content (OpenCode)

Task Template:

kask <<'EOF'
【STEM_SOP V6.0 - Kimi Audit Task】
Target File: [FULL ABSOLUTE PATH]

Mission: Audit code and syntax

Deliverables:
1. Code review report
2. Syntax validation results
3. Fix recommendations

File path: [FULL ABSOLUTE PATH]
EOF

🟡 Qwen (Mathematical Sage)

Model: Qwen2.5-Max / Qwen-Coder (Alibaba) Skills: Mathematical reasoning, LaTeX, code+math integration Command: qask <<'EOF' ... EOF

Core Strengths:

• UNIQUE: Best mathematical reasoning
• Step-by-step proofs and derivations
• LaTeX formula generation
• Code + mathematics integration

Primary Responsibilities:

1. Mathematical Derivations (50% - CORE): Step-by-step proofs, formula derivations
2. LaTeX Formulas (30%): Key equations, aligned derivations, definitions
3. Comparison Tables (20%): Method comparisons with mathematical criteria

LaTeX Format Standards:

$$
\begin{aligned}
f(x) &= \int_0^x g(t) \, dt \\
     &= G(x) - G(0)
\end{aligned}
$$

Comparison Table Template:

| 对比维度 | 方法A | 方法B |
|---------|------|------|
| 时间复杂度 | O(n) | O(n log n) |
| 空间复杂度 | O(1) | O(n) |
| 收敛性 | [证明] | [证明] |

What Qwen Does NOT Do:

• ❌ Code implementation (Codex)
• ❌ Web search (Gemini)
• ❌ Chinese case studies (Kimi)
• ❌ Workflow diagrams (iFlow)

Task Template:

qask <<'EOF'
【STEM_SOP V6.0 - Qwen Math Task】
Target File: [FULL ABSOLUTE PATH]

Mission: Mathematical derivations for [topic]

Deliverables:
1. 10+ LaTeX formulas with derivations
2. Step-by-step proofs
3. Complexity analysis
4. Comparison tables with mathematical criteria

File path: [FULL ABSOLUTE PATH]
EOF

⚫ DeepSeek (Deep Reasoning Sage)

Model: DeepSeek-V3 / DeepSeek-R1 Skills: Deep reasoning, logical analysis, policy research, complex problem solving Command: dskask <<'EOF' ... EOF

Core Strengths:

• UNIQUE: Best deep reasoning and logical analysis
• Complex multi-step reasoning
• Policy and regulatory analysis
• Cross-domain synthesis

Primary Responsibilities:

1. Deep Reasoning (50% - CORE): Complex logical chains, multi-step analysis
2. Policy Analysis (30%): Regulatory implications, policy interpretation
3. Synthesis (20%): Cross-domain integration, holistic conclusions

Reasoning Format Standards:

### 推理链 (Reasoning Chain)

**前提 (Premises):**
1. [前提1]
2. [前提2]

**推理步骤 (Steps):**
1. 由前提1和前提2，可得...
2. 进一步推导...
3. 因此...

**结论 (Conclusion):**
[最终结论]

What DeepSeek Does NOT Do:

• ❌ Code implementation (Codex)
• ❌ Web search (Gemini)
• ❌ Mathematical proofs (Qwen)
• ❌ Workflow diagrams (iFlow)

Task Template:

dskask <<'EOF'
【STEM_SOP V7.0 - DeepSeek Reasoning Task】
Target File: [FULL ABSOLUTE PATH]

Mission: Deep reasoning analysis for [topic]

Deliverables:
1. Multi-step reasoning chains
2. Policy/regulatory analysis
3. Cross-domain synthesis
4. Logical conclusions with evidence

File path: [FULL ABSOLUTE PATH]
EOF

Execution Modes

Mode 1: Auto-Pilot (Sequential Execution)

Trigger: "auto-pilot" / "自动" / "批量"

Protocol:

1. Claude initializes structure
2. Launch AIs sequentially (one completes before next starts)
3. Monitor each completion via bash-notification before proceeding
4. Integrate all outputs after all complete

Sequential Execution Order:

Gemini → Qwen → OpenCode(中文) → Codex → iFlow → Kimi(审计) → DeepSeek(深度推理) → Claude(整合)

Sequential Launch Example:

# Step 1: Launch Gemini first
gask <<'EOF'
[Gemini SOTA research task]
EOF
# Wait for completion...

# Step 2: Launch Qwen after Gemini completes
qask <<'EOF'
[Qwen math task]
EOF
# Wait for completion...

# Step 3: Launch OpenCode (Chinese content) after Qwen completes
oask <<'EOF'
[OpenCode Chinese content task]
EOF
# Wait for completion...

# Step 4: Launch Codex after OpenCode completes
cask <<'EOF'
[Codex code task]
EOF
# Wait for completion...

# Step 5: Launch iFlow after Codex completes
iask <<'EOF'
[iFlow diagram task]
EOF
# Wait for completion...

# Step 6: Launch Kimi (Audit) after iFlow completes
kask <<'EOF'
[Kimi audit task]
EOF
# Wait for completion...

# Step 7: Launch DeepSeek (Deep Reasoning) after Kimi completes
dskask <<'EOF'
[DeepSeek deep reasoning task]
EOF
# Wait for completion...

# Step 8: Claude final integration

Mode 2: Interactive (Step-by-Step with Approval)

Trigger: "interactive" / "step-by-step"

Protocol:

1. Claude initializes → Ask user approval
2. Launch Gemini → Show results → Ask approval
3. Launch Qwen → Show results → Ask approval
4. Launch OpenCode (中文) → Show results → Ask approval
5. Launch Codex → Show results → Ask approval
6. Launch iFlow → Show results → Ask approval
7. Launch Kimi (审计) → Show audit → Ask approval
8. Launch DeepSeek (深度推理) → Show results → Ask approval
9. Final integration

Eight-AI Collaboration Flow

                    Claude (Orchestrator)
                           │
           ┌───────────────┼───────────────┐
           │               │               │
           ▼               ▼               ▼
    ┌──────────┐    ┌──────────┐    ┌──────────┐    ┌──────────┐
    │  Gemini  │    │   Qwen   │    │   Kimi   │    │ DeepSeek │
    │ (SOTA)   │    │  (Math)  │    │(Chinese) │    │(Reasoning)│
    └────┬─────┘    └────┬─────┘    └────┬─────┘    └────┬─────┘
         │               │               │               │
         └───────────────┼───────────────┴───────────────┘
                         │
           ┌─────────────┼─────────────┐
           │             │             │
           ▼             ▼             ▼
    ┌──────────┐  ┌──────────┐  ┌──────────┐
    │  Codex   │  │  iFlow   │  │ OpenCode │
    │  (Code)  │  │(Diagrams)│  │ (Audit)  │
    └────┬─────┘  └────┬─────┘  └────┬─────┘
         │             │             │
         └─────────────┼─────────────┘
                       │
                       ▼
                    Claude
              (Final Integration)

Collaboration Matrix

Quality Checklist

Content Requirements

• 10+ LaTeX formulas (Qwen)
• 5+ code examples (Codex)
• 3+ Mermaid diagrams (iFlow)
• 3+ Chinese case studies (Kimi)
• 5+ SOTA references (Gemini)
• Syntax validated (OpenCode)
• Deep reasoning analysis (DeepSeek)

Formatting

• Mermaid uses A["text"] format
• LaTeX in separate $$ blocks
• Tables properly aligned
• Code blocks have language tags

Troubleshooting

AI Not Responding

1. Check connectivity: [x]ping (e.g., gping, cping, dskping)
2. Check previous output: [x]pend (e.g., gpend, cpend, dskpend)
3. Relaunch with simplified task

File Path Issues

• ALWAYS use full absolute paths
• Verify file exists before launching
• Include path at BOTH start and end of task

Version History

V7.0 (2026-01-27) - Eight AI Architecture:

• ✅ Expanded from 7 to 8 AI models
• ✅ Added DeepSeek for deep reasoning and policy analysis
• ✅ Updated collaboration matrix for 8 AIs
• ✅ Enhanced sequential execution order

V6.0 (2026-01-26) - Seven AI Architecture:

• ✅ Expanded from 4 to 7 AI models
• ✅ Added iFlow (Cursor) for Mermaid diagrams
• ✅ Added Kimi (Moonshot) for Chinese academic writing
• ✅ Added Qwen (Alibaba) for mathematical derivations
• ✅ Reassigned roles based on actual AI capabilities
• ✅ Updated collaboration matrix for 7 AIs
• ✅ Parallel execution optimization

V5.1 (2026-01-25): Four-model architecture with role optimization

Summary: Eight-AI Division

STEM_SOP Version: V7.0-CCB Last Updated: 2026-01-27 Architecture: Eight-Model Distributed Protocol