diff --git a/container/skills/README.md b/container/skills/README.md
index 2b58a20..b6b475d 100644
--- a/container/skills/README.md
+++ b/container/skills/README.md
@@ -13,6 +13,15 @@ Anything the WhatsApp / WeCom / Discord / local-web agent should use belongs **h
 | `sds-gel-review/` | SDS gel image review |
 | `query-*` | Database / API usage (Ensembl, UniProt, KEGG, …) |
 | `blast-search/`, `pubmed-search/`, `sequence-analysis/` | Literature & sequence workflows |
+| `bio-manuscript-*` | Community-contributed manuscript planning pipeline for idea screening, figure planning, manuscript drafting, refinement, and implementation blueprints |
+| `bio-manuscript-common/` | Shared templates and helper scripts used by the manuscript pipeline skill family |
+
+## Community Skills
+
+Some runtime skills may be integrated from BioClaw community contributors when they prove useful in real workflows. The manuscript pipeline skill family currently staged here is integrated as a community-contributed workflow.
+
+Contributor reference:
+- Yuhong Dong, Westlake University PhD candidate, BioClaw community contributor
 
 ## Developer-only skills
 
diff --git a/container/skills/bio-analysis-system/SKILL.md b/container/skills/bio-analysis-system/SKILL.md
new file mode 100644
index 0000000..39954ba
--- /dev/null
+++ b/container/skills/bio-analysis-system/SKILL.md
@@ -0,0 +1,161 @@
+# bio-analysis-system
+
+**Step 5: Analysis system design (分析方法体系构建)**
+
+Build the analysis layer for the manuscript by identifying which analyses, tools, and biological validations should support each figure and each task.
+
+## Purpose
+
+1. Extract analysis patterns from related work
+2. Borrow useful analyses from adjacent domains when needed
+3. Map analyses to BioClaw-compatible tools or fallback software
+4. Explain why each analysis is included and what biological claim it supports
+5. Connect analyses to figure panels
+
+## Input Format
+
+```text
+topic: [research topic]
+paper_count: [number of related papers]
+task_system: [task system]
+metric_system: [metric system]
+dataset_catalog: [dataset catalog]
+```
+
+## Workflow
+
+### Step 5.1: Extract analyses from existing work
+
+If enough related papers exist, inspect their figures and extract:
+
+- panel type
+- analysis method
+- software / package
+- important parameters
+- the scientific or biological conclusion the panel supports
+
+### Step 5.2: Borrow from adjacent fields
+
+If the field is still thin, adapt common analyses from nearby areas such as:
+
+- clustering
+- marker visualization
+- latent embedding visualization
+- pathway enrichment
+- cell-cell communication
+- spatial statistics
+- GRN analysis
+
+### Step 5.3: Categorize analyses
+
+Use three broad groups:
+
+- **Quantitative analyses**
+  - clustering
+  - metric computation
+  - statistical tests
+  - baseline comparisons
+- **Qualitative analyses**
+  - spatial visualization
+  - feature / violin plots
+  - UMAP / t-SNE
+  - before / after alignment comparisons
+  - heatmaps
+- **Biological analyses**
+  - cell annotation
+  - marker genes
+  - pathway enrichment
+  - GRN
+  - ligand-receptor communication
+  - spatial statistics
+  - trajectory analysis
+
+### Step 5.4: Map to BioClaw or fallback tools
+
+Whenever possible, map analysis needs to BioClaw-compatible skills or established tools.
+
+Examples:
+
+- clustering -> Scanpy / Leiden
+- annotation -> CellTypist / SingleR
+- marker plots -> Scanpy
+- enrichment -> gseapy
+- spatial statistics -> squidpy
+- GRN -> pySCENIC
+- communication -> CellChat-like workflow
+
+### Step 5.5: Standardize analysis descriptions
+
+For each analysis, define:
+
+- category
+- purpose
+- biological claim supported
+- preferred tool
+- fallback tool
+- key function
+- recommended parameters
+- inputs / outputs
+- mapped task
+- mapped figure / panel
+
+## Output Format
+
+```markdown
+# Analysis System
+
+## Analysis Sources
+- Extracted from related papers:
+- Borrowed from adjacent domains:
+
+## Quantitative Analyses
+
+### Clustering
+- Category:
+- Purpose:
+- Biological claim supported:
+- Preferred tool:
+- Fallback tool:
+- Key function:
+- Recommended parameters:
+- Inputs / outputs:
+- Relevant tasks:
+- Figure mapping:
+
+### Metric computation
+- Category:
+- Purpose:
+- Preferred tools:
+- Relevant tasks:
+- Figure mapping:
+
+## Qualitative Analyses
+- spatial plot
+- marker / feature plot
+- latent embedding plot
+- heatmap
+- before / after alignment visualization
+
+## Biological Analyses
+- annotation
+- marker recovery
+- pathway enrichment
+- GRN
+- communication
+- trajectory
+
+## Next Step
+- Use the analysis system to design figures in Step 6
+```
+
+## Usage
+
+```bash
+/bio-analysis-system "spatial multi-omics integration | paper_count: 5 | task_system: [...] | metric_system: [...] | dataset_catalog: [...]"
+```
+
+## Notes
+
+1. Prefer analyses that directly support paper claims.
+2. Make the biological readouts visible early; they should not appear only at the very end.
+3. Map each major analysis to a concrete figure panel.
diff --git a/container/skills/bio-dataset-search/SKILL.md b/container/skills/bio-dataset-search/SKILL.md
new file mode 100644
index 0000000..1b9ec2f
--- /dev/null
+++ b/container/skills/bio-dataset-search/SKILL.md
@@ -0,0 +1,135 @@
+# bio-dataset-search
+
+**Step 3: Dataset search and task matching (数据集搜索与匹配)**
+
+Find suitable datasets for each task and map datasets to the task system defined earlier in the manuscript pipeline.
+
+## Purpose
+
+1. Extract datasets from related papers when possible
+2. Search public repositories directly when needed
+3. Normalize dataset metadata into a common structure
+4. Match datasets to tasks in a defendable way
+
+## Input Format
+
+```text
+topic: [research topic]
+task_system: [task system from Step 2]
+paper_count: [number of related papers]
+existing_papers: [optional list of related papers]
+```
+
+## Workflow
+
+### Step 3.1: Extract datasets from existing work
+
+If `paper_count >= 5`, start from the strongest existing papers.
+
+Read Methods / Data Availability sections and extract:
+
+- dataset name
+- data source
+- platform
+- modality
+- sample scale
+- download path
+- annotation availability
+
+### Step 3.2: Search datasets directly
+
+If there is not enough prior work, search repositories such as:
+
+- GEO
+- ArrayExpress
+- project-specific public portals
+
+Use keyword sets built from:
+
+- topic
+- modality
+- tissue / disease
+- benchmark intent
+
+### Step 3.3: Normalize dataset metadata
+
+For each dataset, record:
+
+- source
+- platform
+- species
+- tissue / disease
+- sample size
+- feature count
+- modalities
+- annotation quality
+- histology / region metadata
+- format
+- preprocessing needs
+- recommended task fit
+
+### Step 3.4: Match datasets to tasks
+
+A good match should satisfy:
+
+1. Every major task has at least one viable dataset
+2. Dataset structure matches the task's technical assumptions
+3. Download remains feasible
+4. Metadata quality is sufficient for evaluation
+5. Prefer at least one backup dataset per important task
+
+## Output Format
+
+```markdown
+# Dataset Catalog
+
+## Data Sources
+- Extracted from related papers:
+- Direct repository search:
+- Borrowed from adjacent domains:
+
+## Dataset Entries
+
+### Dataset 1: [name]
+- Source:
+- Platform:
+- Species:
+- Tissue / disease:
+- Modalities:
+- Sample scale:
+- Annotation quality:
+- Download URL:
+- Format:
+- Recommended tasks:
+- Why it fits:
+
+## Dataset-Task Mapping
+| Task | Recommended dataset | Why it fits | Notes |
+|------|---------------------|-------------|-------|
+| ... | ... | ... | ... |
+
+## Acquisition Notes
+- GEO download hints
+- Public portal download hints
+
+## Preprocessing Recommendations
+| Dataset | Preprocessing needs | Suggested skill / tool |
+|---------|---------------------|------------------------|
+| ... | ... | ... |
+
+## Next Step
+- Build the metric system in Step 4
+```
+
+## Usage
+
+```bash
+/bio-dataset-search "spatial multi-omics integration | paper_count: 5 | task_system: [task system from Step 2]"
+```
+
+## Notes
+
+1. Prefer datasets already used in related work when possible.
+2. Verify links before committing them to the benchmark plan.
+3. Capture QC and annotation metadata whenever available.
+4. Match datasets to tasks based on actual experimental needs, not just popularity.
diff --git a/container/skills/bio-figure-design/SKILL.md b/container/skills/bio-figure-design/SKILL.md
new file mode 100644
index 0000000..496552c
--- /dev/null
+++ b/container/skills/bio-figure-design/SKILL.md
@@ -0,0 +1,125 @@
+# bio-figure-design
+
+**Step 6: Figure design (Figure 详细设计)**
+
+Design the manuscript figures panel by panel, including the figure logic, panel content, and caption intent.
+
+## Purpose
+
+1. Design Figure 1 as the method / framework overview
+2. Design Figures 2-N as task-driven application figures
+3. Plan supplementary figures
+4. Draft figure captions
+5. Keep figure logic synchronized with manuscript claims
+
+## Input Format
+
+```text
+topic: [research topic]
+task_system: [task system]
+dataset_catalog: [dataset catalog]
+metric_system: [metric system]
+analysis_system: [analysis system]
+target_journal: [optional, default nat-communications]
+```
+
+## Workflow
+
+### Step 6.1: Design Figure 1
+
+Figure 1 should explain the overall method and paper framing:
+
+- Panel a: method / model overview
+- Panel b: data or modality overview
+- Panel c: task overview
+- Panel d: metric overview
+- Panel e: analysis overview
+
+The goal is to make the full manuscript logic visible in one figure.
+
+### Step 6.2: Design Figures 2-N
+
+Each application figure should be task-first:
+
+- one major task per figure
+- panel a: data flow / experimental setup
+- panels b-d: quantitative evaluation
+- panel e or later: qualitative / biological validation
+
+This keeps the paper organized around claims rather than around plots.
+
+### Step 6.3: Supplementary figures
+
+Use supplementary figures for:
+
+- ablations
+- robustness checks
+- extra markers
+- extended datasets
+- alternative parameter settings
+
+### Step 6.4: Caption planning
+
+Every figure should have a caption plan that explains:
+
+- what each panel shows
+- what claim it supports
+- what dataset it uses
+- what metric or biological conclusion it demonstrates
+
+## Output Format
+
+```markdown
+# Figure Designs
+
+## Figure 1: Framework Overview
+- Panel a:
+- Panel b:
+- Panel c:
+- Panel d:
+- Panel e:
+- Caption intent:
+
+## Figure 2: [Task 1]
+- Panel a:
+- Panel b:
+- Panel c:
+- Panel d:
+- Panel e:
+- Caption intent:
+
+## Figure 3: [Task 2]
+...
+
+## Supplementary Figures
+- Supplementary Figure 1:
+- Supplementary Figure 2:
+
+## Design Notes
+- visual consistency
+- panel ordering logic
+- expected take-home message per figure
+
+## Next Step
+- Use the figure plan to draft manuscript text in Step 7
+```
+
+## Figure Design Principles
+
+1. One main claim per figure
+2. Quantitative evidence should appear before broad interpretation
+3. Biological validation should be visible, not hidden
+4. Reviewer-facing clarity matters more than decorative complexity
+5. Figure order should match the manuscript story
+
+## Usage
+
+```bash
+/bio-figure-design "topic: spatial multi-omics integration | task_system: [...] | dataset_catalog: [...] | metric_system: [...] | analysis_system: [...] | target_journal: nat-communications"
+```
+
+## Notes
+
+1. Keep Figure 1 conceptual and clean.
+2. For application figures, tie every panel to a concrete task and metric.
+3. Do not overload a figure if a supplementary figure can carry the extra material.
diff --git a/container/skills/bio-human-feedback/SKILL.md b/container/skills/bio-human-feedback/SKILL.md
new file mode 100644
index 0000000..a53aba3
--- /dev/null
+++ b/container/skills/bio-human-feedback/SKILL.md
@@ -0,0 +1,80 @@
+# bio-human-feedback
+
+**Phase 2.6: Human review checkpoint (人类反馈验证)**
+
+Present the refined proposal to a human reviewer, collect feedback, and decide whether to continue or loop back for revision.
+
+## Purpose
+
+1. Present the current proposal in a concise reviewable form
+2. Wait for explicit human feedback
+3. Record approval or requested changes
+4. Route the workflow back to the correct phase if needed
+
+## Inputs
+
+- `FINAL_PROPOSAL.md`
+- current round number
+
+## Workflow
+
+1. Present the proposal summary
+2. Wait for approval or feedback
+3. Classify the feedback severity
+4. Record the response and decide the next phase
+
+## Human Review Summary Format
+
+```markdown
+# Proposal Review - Round X
+
+## Innovation Summary
+- ...
+
+## Figure Overview
+- Figure 1:
+- Figure 2:
+
+## Experimental Plan
+| Task | Dataset | Metric | Baseline |
+|------|---------|--------|----------|
+| ... | ... | ... | ... |
+
+## Key Decisions Requiring Human Approval
+1. ...
+2. ...
+
+## Reviewer Summary
+| Reviewer | Score / status | Main advice |
+|----------|----------------|-------------|
+| Editor | ... | ... |
+| Computational | ... | ... |
+| Biological | ... | ... |
+
+## Human Decision
+- Approve and continue
+- Request revisions
+```
+
+## Feedback Severity
+
+- **Critical**: return to early planning
+- **Major**: return to design / manuscript refinement
+- **Minor**: revise locally and continue
+
+## Files To Record
+
+- `refine-logs/human-feedback/feedback-round-X.md`
+- `refine-logs/HUMAN_APPROVAL.md`
+
+## Usage
+
+```bash
+/bio-human-feedback --round 2 --proposal refine-logs/FINAL_PROPOSAL.md
+```
+
+## Notes
+
+1. Do not continue automatically without explicit human approval.
+2. Record all human feedback in the workspace.
+3. Keep the requested decisions concise and actionable.
diff --git a/container/skills/bio-innovation-check/SKILL.md b/container/skills/bio-innovation-check/SKILL.md
new file mode 100644
index 0000000..09a9e37
--- /dev/null
+++ b/container/skills/bio-innovation-check/SKILL.md
@@ -0,0 +1,141 @@
+# bio-innovation-check
+
+**Step 1: Innovation assessment (创新性检测)**
+
+Estimate whether a research idea is sufficiently novel for a strong methods-style paper by expanding the topic and searching the literature.
+
+## Purpose
+
+1. Generate multiple topic variants and synonyms
+2. Search PubMed, bioRxiv, and arXiv q-bio
+3. Count and de-duplicate related papers
+4. Assign a novelty level
+5. Suggest how to sharpen or reposition the idea if needed
+
+## Input Format
+
+```text
+topic: [research topic]
+```
+
+## Workflow
+
+### Step 1.1: Topic expansion
+
+Use several types of expansions:
+
+1. Core term substitution
+2. Phrase re-ordering
+3. Parent / child concept expansion
+4. Adjacent-domain vocabulary borrowing
+5. Method keyword enrichment
+
+Example:
+
+- "spatial multi-omics integration"
+- "integration of spatial transcriptomics and proteomics"
+- "spatial multi-modal data fusion"
+
+Target output: 15-20 topic variants by default.
+
+### Step 1.2: Literature search
+
+Search these sources:
+
+- PubMed
+- bioRxiv
+- arXiv q-bio
+
+Suggested pattern:
+
+```python
+for variant in topic_variants:
+    results = search(variant, platforms=["PubMed", "bioRxiv", "arXiv"])
+    all_papers.extend(results)
+
+unique_papers = deduplicate(all_papers, threshold=0.8)
+```
+
+### Step 1.3: Novelty scoring
+
+Use a simple first-pass threshold:
+
+```python
+if paper_count <= 2:
+    level = "strong novelty / methods-journal candidate"
+elif paper_count <= 5:
+    level = "promising but needs sharpening"
+else:
+    level = "needs repositioning"
+```
+
+This is only a heuristic. Final judgment should still use human reasoning.
+
+### Step 1.4: Repositioning suggestions
+
+If the project is not yet strong enough, suggest improvements from one or more of these angles:
+
+1. Method angle
+2. Task angle
+3. Data / validation angle
+4. Analysis angle
+
+## Output Format
+
+```markdown
+# Innovation Assessment Report
+
+## Search Strategy
+- Number of variants:
+- Search sources:
+- Search date:
+
+## Topic Variants
+| No. | Variant |
+|-----|---------|
+| 1 | ... |
+
+## Search Results Summary
+| Variant | PubMed | bioRxiv | arXiv | Total |
+|---------|--------|---------|-------|-------|
+| ... | ... | ... | ... | ... |
+
+## De-duplicated Counts
+- Total related studies:
+- Published papers:
+- Preprints:
+
+## Novelty Decision
+- Level:
+- Reason:
+
+## Representative Related Work
+1. [title]
+   - Source:
+   - Year:
+   - Main method:
+   - Overlap with the proposed idea:
+
+## Repositioning Suggestions
+1. Method:
+2. Task:
+3. Data / validation:
+
+## Next Step
+- If novelty is strong: continue to Step 2
+- If the idea needs sharpening: refine and continue
+- If it needs repositioning: redesign before proceeding
+```
+
+## Usage
+
+```bash
+/bio-innovation-check "spatial multi-omics integration"
+```
+
+## Notes
+
+1. Use timeouts because search latency varies by source.
+2. De-duplication matters; otherwise novelty will be overestimated or underestimated.
+3. Overlap scoring still needs human judgment.
+4. Journal-specific novelty expectations can differ by field.
diff --git a/container/skills/bio-manuscript-common/README.md b/container/skills/bio-manuscript-common/README.md
new file mode 100644
index 0000000..c278d1c
--- /dev/null
+++ b/container/skills/bio-manuscript-common/README.md
@@ -0,0 +1,13 @@
+## Bio Manuscript Common
+
+Shared assets for the BioClaw integration of `bio-manuscript-forge`.
+
+Contents:
+- `templates/`: journal and manuscript planning templates
+- `scripts/`: helper scripts copied from the upstream repository
+
+Integration notes:
+- These assets were copied from `external/bio-manuscript-forge/bio-manuscript-forge/`.
+- Runtime skills live under `/home/node/.claude/skills/` inside the BioClaw container.
+- Keep skill-to-skill references within `container/skills/` and avoid `~/.openclaw/...` assumptions.
+- Treat this directory as shared support data for the manuscript pipeline skill family.
diff --git a/container/skills/bio-manuscript-common/scripts/generate_figure_image.py b/container/skills/bio-manuscript-common/scripts/generate_figure_image.py
new file mode 100644
index 0000000..3072d11
--- /dev/null
+++ b/container/skills/bio-manuscript-common/scripts/generate_figure_image.py
@@ -0,0 +1,247 @@
+#!/usr/bin/env python3
+"""
+Figure 图文生成器
+
+调用 Gemini API 生成 Figure 1 Panel b-c-d-e 的图文。
+"""
+
+import os
+from typing import Optional, Dict, List
+
+
+class FigureImageGenerator:
+    """Figure 图文生成器"""
+    
+    def __init__(self, api_key: Optional[str] = None):
+        """
+        初始化
+        
+        Args:
+            api_key: Gemini API key（也可从环境变量获取）
+        """
+        self.api_key = api_key or os.environ.get("GEMINI_API_KEY")
+        if not self.api_key:
+            print("Warning: No Gemini API key provided. Image generation will be simulated.")
+    
+    def generate_panel_image(self,
+                            panel_type: str,
+                            content: str,
+                            description: str,
+                            output_path: str) -> Dict:
+        """
+        生成 Panel 图文
+        
+        Args:
+            panel_type: Panel 类型（data/task/metric/analysis）
+            content: 内容描述
+            description: 简短描述
+            output_path: 输出路径
+            
+        Returns:
+            生成结果
+        """
+        prompt = self._build_prompt(panel_type, content, description)
+        
+        # 调用 Gemini API（简化版）
+        result = self._call_gemini(prompt, output_path)
+        
+        return result
+    
+    def _build_prompt(self, panel_type: str, content: str, description: str) -> str:
+        """构建 prompt"""
+        
+        panel_descriptions = {
+            "data": "数据类型介绍",
+            "task": "任务层级介绍",
+            "metric": "评价指标介绍",
+            "analysis": "分析方法介绍"
+        }
+        
+        prompt = f"""
+生成一个简洁的科学论文插图，用于 Nature Methods 论文 Figure 1。
+
+Panel 类型：{panel_descriptions.get(panel_type, panel_type)}
+
+内容：{content}
+
+简短描述：{description}
+
+要求：
+1. 风格：Nature Methods 论文插图，简洁清晰，配色专业
+2. 布局：包含标题标签 + 示意图 + 简短描述文字
+3. 颜色：专业科学论文配色，避免过于鲜艳
+4. 尺寸：适合 Figure 1 Panel（约 60mm × 30mm 每个项目）
+5. 格式：矢量图或高分辨率位图
+
+请生成一个框框样式的设计，框框内包含：
+- 内容名称（如 "Proteomics"）
+- 简洁的示意图
+- 一句话描述
+"""
+        return prompt
+    
+    def _call_gemini(self, prompt: str, output_path: str) -> Dict:
+        """调用 Gemini API"""
+        
+        if not self.api_key:
+            # 模拟生成
+            return {
+                "success": False,
+                "message": "No API key provided. This is a simulation.",
+                "prompt": prompt,
+                "output_path": output_path,
+                "note": "Please provide GEMINI_API_KEY to enable actual image generation."
+            }
+        
+        try:
+            # 实际调用 Gemini API
+            # import google.generativeai as genai
+            # genai.configure(api_key=self.api_key)
+            # model = genai.GenerativeModel('gemini-pro-vision')
+            # response = model.generate_content(prompt)
+            # ... 保存图片
+            
+            return {
+                "success": True,
+                "prompt": prompt,
+                "output_path": output_path,
+                "message": "Image generated successfully (placeholder)"
+            }
+        except Exception as e:
+            return {
+                "success": False,
+                "error": str(e),
+                "prompt": prompt
+            }
+    
+    def generate_figure1_panel_b(self, data_types: List[Dict]) -> Dict:
+        """
+        生成 Figure 1 Panel b（数据介绍）
+        
+        Args:
+            data_types: 数据类型列表
+                [{"name": "Proteomics", "description": "蛋白质表达测量"}, ...]
+                
+        Returns:
+            生成结果
+        """
+        results = []
+        
+        for data_type in data_types:
+            result = self.generate_panel_image(
+                panel_type="data",
+                content=data_type["name"],
+                description=data_type["description"],
+                output_path=f"figure1_panel_b_{data_type['name'].lower()}.png"
+            )
+            results.append(result)
+        
+        return {
+            "panel": "Figure 1 Panel b",
+            "results": results
+        }
+    
+    def generate_figure1_panel_c(self, tasks: List[Dict]) -> Dict:
+        """
+        生成 Figure 1 Panel c（任务介绍）
+        
+        Args:
+            tasks: 任务列表
+                [{"name": "Vertical Integration", "description": "同细胞多模态"}, ...]
+                
+        Returns:
+            生成结果
+        """
+        results = []
+        
+        for task in tasks:
+            result = self.generate_panel_image(
+                panel_type="task",
+                content=task["name"],
+                description=task["description"],
+                output_path=f"figure1_panel_c_{task['name'].lower().replace(' ', '_')}.png"
+            )
+            results.append(result)
+        
+        return {
+            "panel": "Figure 1 Panel c",
+            "results": results
+        }
+    
+    def generate_figure1_panel_d(self, metrics: List[Dict]) -> Dict:
+        """
+        生成 Figure 1 Panel d（指标介绍）
+        
+        Args:
+            metrics: 指标列表
+                [{"name": "ARI", "description": "聚类一致性评估"}, ...]
+                
+        Returns:
+            生成结果
+        """
+        results = []
+        
+        for metric in metrics:
+            result = self.generate_panel_image(
+                panel_type="metric",
+                content=metric["name"],
+                description=metric["description"],
+                output_path=f"figure1_panel_d_{metric['name'].lower()}.png"
+            )
+            results.append(result)
+        
+        return {
+            "panel": "Figure 1 Panel d",
+            "results": results
+        }
+    
+    def generate_figure1_panel_e(self, analyses: List[Dict]) -> Dict:
+        """
+        生成 Figure 1 Panel e（分析介绍）
+        
+        Args:
+            analyses: 分析列表
+                [{"name": "Clustering", "description": "Leiden聚类"}, ...]
+                
+        Returns:
+            生成结果
+        """
+        results = []
+        
+        for analysis in analyses:
+            result = self.generate_panel_image(
+                panel_type="analysis",
+                content=analysis["name"],
+                description=analysis["description"],
+                output_path=f"figure1_panel_e_{analysis['name'].lower()}.png"
+            )
+            results.append(result)
+        
+        return {
+            "panel": "Figure 1 Panel e",
+            "results": results
+        }
+
+
+def main():
+    """主函数"""
+    generator = FigureImageGenerator()
+    
+    # 示例：生成 Panel b（数据介绍）
+    data_types = [
+        {"name": "Proteomics", "description": "蛋白质表达测量，提供细胞功能信息"},
+        {"name": "Transcriptomics", "description": "全转录组表达，揭示基因调控"},
+        {"name": "Epigenomics", "description": "表观遗传修饰，反映染色质状态"},
+    ]
+    
+    result = generator.generate_figure1_panel_b(data_types)
+    
+    print(f"Panel: {result['panel']}")
+    print(f"Generated {len(result['results'])} items")
+    
+    for r in result['results']:
+        print(f"  - {r.get('output_path', 'N/A')}: {r.get('message', r.get('error', 'Unknown'))}")
+
+
+if __name__ == "__main__":
+    main()
\ No newline at end of file
diff --git a/container/skills/bio-manuscript-common/scripts/integrate_omicsclaw.py b/container/skills/bio-manuscript-common/scripts/integrate_omicsclaw.py
new file mode 100644
index 0000000..afcf11e
--- /dev/null
+++ b/container/skills/bio-manuscript-common/scripts/integrate_omicsclaw.py
@@ -0,0 +1,194 @@
+#!/usr/bin/env python3
+"""
+OmicsClaw 联动脚本
+
+封装 OmicsClaw CLI 调用，用于实际数据分析。
+"""
+
+import subprocess
+import os
+from typing import Optional, Dict, List
+
+
+class OmicsClawRunner:
+    """OmicsClaw 运行器"""
+    
+    def __init__(self, omicsclaw_path: str = "omicsclaw"):
+        """
+        初始化
+        
+        Args:
+            omicsclaw_path: omicsclaw 命令路径
+        """
+        self.omicsclaw_path = omicsclaw_path
+    
+    def run_skill(self, 
+                  skill_name: str,
+                  input_file: str,
+                  output_dir: Optional[str] = None,
+                  params: Optional[Dict] = None) -> Dict:
+        """
+        运行 OmicsClaw skill
+        
+        Args:
+            skill_name: skill 名称
+            input_file: 输入文件路径
+            output_dir: 输出目录
+            params: 参数字典
+            
+        Returns:
+            运行结果
+        """
+        cmd = [self.omicsclaw_path, "run", skill_name, "--input", input_file]
+        
+        if output_dir:
+            cmd.extend(["--output", output_dir])
+        
+        if params:
+            for key, value in params.items():
+                cmd.extend([f"--{key}", str(value)])
+        
+        # 运行命令
+        result = subprocess.run(cmd, capture_output=True, text=True)
+        
+        return {
+            "skill": skill_name,
+            "command": " ".join(cmd),
+            "returncode": result.returncode,
+            "stdout": result.stdout,
+            "stderr": result.stderr,
+            "success": result.returncode == 0
+        }
+    
+    def preprocess(self, input_file: str, output_dir: str) -> Dict:
+        """
+        数据预处理
+        
+        Args:
+            input_file: 输入文件
+            output_dir: 输出目录
+            
+        Returns:
+            运行结果
+        """
+        return self.run_skill(
+            skill_name="spatial-preprocess",
+            input_file=input_file,
+            output_dir=output_dir
+        )
+    
+    def cluster(self, 
+                input_file: str, 
+                method: str = "leiden",
+                resolution: float = 0.5) -> Dict:
+        """
+        职类分析
+        
+        Args:
+            input_file: 输入文件
+            method: 职类方法
+            resolution: 分辨率参数
+            
+        Returns:
+            运行结果
+        """
+        return self.run_skill(
+            skill_name="spatial-domains",
+            input_file=input_file,
+            params={"method": method, "resolution": resolution}
+        )
+    
+    def annotate(self, input_file: str, method: str = "markers") -> Dict:
+        """
+        细胞类型注释
+        
+        Args:
+            input_file: 输入文件
+            method: 注释方法
+            
+        Returns:
+            运行结果
+        """
+        return self.run_skill(
+            skill_name="spatial-annotate",
+            input_file=input_file,
+            params={"method": method}
+        )
+    
+    def find_markers(self, input_file: str, output_dir: str) -> Dict:
+        """
+        Marker 基因识别
+        
+        Args:
+            input_file: 输入文件
+            output_dir: 输出目录
+            
+        Returns:
+            运行结果
+        """
+        return self.run_skill(
+            skill_name="sc-markers",
+            input_file=input_file,
+            output_dir=output_dir
+        )
+    
+    def spatial_statistics(self, input_file: str) -> Dict:
+        """
+        空间统计分析
+        
+        Args:
+            input_file: 输入文件
+            
+        Returns:
+            运行结果
+        """
+        return self.run_skill(
+            skill_name="spatial-statistics",
+            input_file=input_file
+        )
+    
+    def enrichment(self, 
+                   gene_list: List[str], 
+                   database: str = "KEGG_2021_Human") -> Dict:
+        """
+        通路富集分析
+        
+        Args:
+            gene_list: 基因列表
+            database: 富集数据库
+            
+        Returns:
+            运行结果
+        """
+        # 创建临时文件
+        temp_file = "/tmp/gene_list.txt"
+        with open(temp_file, "w") as f:
+            f.write("\n".join(gene_list))
+        
+        return self.run_skill(
+            skill_name="spatial-enrichment",
+            input_file=temp_file,
+            params={"database": database}
+        )
+
+
+def main():
+    """主函数"""
+    runner = OmicsClawRunner()
+    
+    # 示例：预处理
+    print("示例：运行 spatial-preprocess")
+    result = runner.preprocess(
+        input_file="data/example.h5ad",
+        output_dir="results/"
+    )
+    
+    print(f"命令: {result['command']}")
+    print(f"成功: {result['success']}")
+    
+    if not result['success']:
+        print(f"错误: {result['stderr']}")
+
+
+if __name__ == "__main__":
+    main()
\ No newline at end of file
diff --git a/container/skills/bio-manuscript-common/scripts/topic_synonym_transform.py b/container/skills/bio-manuscript-common/scripts/topic_synonym_transform.py
new file mode 100644
index 0000000..3744699
--- /dev/null
+++ b/container/skills/bio-manuscript-common/scripts/topic_synonym_transform.py
@@ -0,0 +1,171 @@
+#!/usr/bin/env python3
+"""
+Topic 同义变换生成器
+
+生成研究主题的多个语义等价表达，用于文献搜索。
+"""
+
+import re
+from typing import List, Dict
+
+
+class TopicSynonymTransformer:
+    """Topic 同义变换器"""
+    
+    # 领域术语映射
+    DOMAIN_TERMS = {
+        "空间组学": ["spatial omics", "spatial transcriptomics", "spatial biology"],
+        "多组学": ["multi-omics", "multi-modal", "multimodal"],
+        "整合": ["integration", "fusion", "alignment", "mapping"],
+        "单细胞": ["single-cell", "scRNA-seq", "single cell"],
+        "转录组": ["transcriptomics", "RNA-seq", "gene expression"],
+        "蛋白质组": ["proteomics", "protein expression"],
+        "表观遗传": ["epigenomics", "epigenetic", "ATAC-seq", "ChIP-seq"],
+        "细胞类型": ["cell type", "cell identity", "cell annotation"],
+        "轨迹": ["trajectory", "pseudotime", "development"],
+        "通讯": ["communication", "interaction", "signaling"],
+    }
+    
+    def __init__(self):
+        self.variants = []
+    
+    def transform(self, topic: str) -> List[str]:
+        """
+        生成 topic 的同义变换
+        
+        Args:
+            topic: 原始研究主题
+            
+        Returns:
+            变体列表
+        """
+        self.variants = []
+        
+        # 1. 原始 topic
+        self.variants.append(topic)
+        
+        # 2. 中文翻译
+        en_topic = self._translate_to_english(topic)
+        if en_topic != topic:
+            self.variants.append(en_topic)
+        
+        # 3. 术语替换
+        term_variants = self._replace_terms(en_topic)
+        self.variants.extend(term_variants)
+        
+        # 4. 词序调整
+        order_variants = self._reorder_words(en_topic)
+        self.variants.extend(order_variants)
+        
+        # 5. 上下位扩展
+        hierarchy_variants = self._expand_hierarchy(en_topic)
+        self.variants.extend(hierarchy_variants)
+        
+        # 去重
+        self.variants = list(set(self.variants))
+        
+        return self.variants
+    
+    def _translate_to_english(self, topic: str) -> str:
+        """中文翻译为英文"""
+        # 简单的术语替换
+        result = topic
+        for cn, en_list in self.DOMAIN_TERMS.items():
+            if cn in topic:
+                result = result.replace(cn, en_list[0])
+        return result
+    
+    def _replace_terms(self, topic: str) -> List[str]:
+        """术语替换"""
+        variants = []
+        
+        for cn, en_list in self.DOMAIN_TERMS.items():
+            for en in en_list:
+                # 检查是否有相关术语
+                for other_cn, other_en_list in self.DOMAIN_TERMS.items():
+                    for other_en in other_en_list:
+                        if other_en in topic.lower():
+                            new_topic = topic.lower().replace(other_en, en)
+                            if new_topic != topic.lower():
+                                variants.append(new_topic)
+        
+        return variants
+    
+    def _reorder_words(self, topic: str) -> List[str]:
+        """词序调整"""
+        variants = []
+        words = topic.lower().split()
+        
+        if len(words) >= 3:
+            # 交换前两个词
+            variant1 = ' '.join([words[1], words[0]] + words[2:])
+            variants.append(variant1)
+            
+            # "integration of X and Y" → "X and Y integration"
+            if "integration" in words:
+                idx = words.index("integration")
+                if idx == 0:
+                    variant2 = ' '.join(words[1:] + ["integration"])
+                    variants.append(variant2)
+        
+        return variants
+    
+    def _expand_hierarchy(self, topic: str) -> List[str]:
+        """上下位扩展"""
+        variants = []
+        
+        # 上位词
+        if "spatial" in topic.lower():
+            # 上位：multi-modal
+            variants.append(topic.lower().replace("spatial", "multi-modal"))
+            variants.append(topic.lower().replace("spatial", "single-cell"))
+        
+        if "multi-omics" in topic.lower():
+            # 下位：具体模态
+            variants.append(topic.lower().replace("multi-omics", "transcriptomics proteomics"))
+            variants.append(topic.lower().replace("multi-omics", "RNA protein"))
+        
+        return variants
+    
+    def generate_search_queries(self, topic: str, max_variants: int = 15) -> List[Dict]:
+        """
+        生成搜索查询
+        
+        Returns:
+            查询列表，每个包含 topic 和搜索平台
+        """
+        variants = self.transform(topic)[:max_variants]
+        
+        queries = []
+        for i, variant in enumerate(variants):
+            queries.append({
+                "id": i + 1,
+                "query": variant,
+                "platforms": ["PubMed", "bioRxiv", "arXiv q-bio"]
+            })
+        
+        return queries
+
+
+def main():
+    """主函数"""
+    transformer = TopicSynonymTransformer()
+    
+    # 示例
+    topic = "空间多组学整合"
+    variants = transformer.transform(topic)
+    
+    print(f"原始 Topic: {topic}")
+    print(f"\n生成的 {len(variants)} 个变体:")
+    for i, v in enumerate(variants, 1):
+        print(f"  {i}. {v}")
+    
+    print(f"\n搜索查询:")
+    queries = transformer.generate_search_queries(topic)
+    for q in queries:
+        print(f"  Query {q['id']}: {q['query']}")
+        print(f"    Platforms: {', '.join(q['platforms'])}")
+
+
+if __name__ == "__main__":
+    main()
\ No newline at end of file
diff --git a/container/skills/bio-manuscript-common/templates/nature-methods/figure_template.md b/container/skills/bio-manuscript-common/templates/nature-methods/figure_template.md
new file mode 100644
index 0000000..2a90e31
--- /dev/null
+++ b/container/skills/bio-manuscript-common/templates/nature-methods/figure_template.md
@@ -0,0 +1,33 @@
+# Figure Template
+
+## Figure Planning Block
+
+### Figure title
+
+[Figure title]
+
+### Scientific claim
+
+[What this figure proves]
+
+### Panel layout
+
+- Panel a:
+- Panel b:
+- Panel c:
+- Panel d:
+- Panel e:
+
+### Dataset / benchmark
+
+- dataset:
+- baseline:
+- metric:
+
+### Biological interpretation
+
+[What biological conclusion or validation this figure supports]
+
+### Caption intent
+
+[2-4 sentences describing what the full figure caption should communicate]
diff --git a/container/skills/bio-manuscript-common/templates/nature-methods/introduction_template.md b/container/skills/bio-manuscript-common/templates/nature-methods/introduction_template.md
new file mode 100644
index 0000000..090e40f
--- /dev/null
+++ b/container/skills/bio-manuscript-common/templates/nature-methods/introduction_template.md
@@ -0,0 +1,35 @@
+# Introduction Template
+
+## Five-Paragraph Structure
+
+### Paragraph 1: Field background
+
+Introduce the field, explain why the problem matters, and highlight the broad motivation.
+
+### Paragraph 2: Related work
+
+Summarize the strongest prior methods and their main ideas.
+
+### Paragraph 3: Current limitations
+
+State the main gaps in existing methods, benchmarks, analyses, or biological interpretation.
+
+### Paragraph 4: Our method
+
+Introduce the proposed method and clearly separate:
+
+- conceptual novelty
+- algorithmic novelty
+- task / analysis novelty
+
+### Paragraph 5: Significance
+
+Explain why the method matters technically, biologically, and methodologically.
+
+## Writing Checklist
+
+- concise field motivation
+- reviewer-friendly related work summary
+- explicit gap statement
+- concrete method positioning
+- realistic significance claim
diff --git a/container/skills/bio-manuscript-common/templates/nature-methods/manuscript_template.md b/container/skills/bio-manuscript-common/templates/nature-methods/manuscript_template.md
new file mode 100644
index 0000000..7895bbf
--- /dev/null
+++ b/container/skills/bio-manuscript-common/templates/nature-methods/manuscript_template.md
@@ -0,0 +1,45 @@
+# Manuscript Template
+
+## Title
+
+[Working title]
+
+## Abstract
+
+- background:
+- gap:
+- method:
+- results:
+- significance:
+
+## Introduction
+
+- field background
+- related work
+- limitations
+- our method
+- significance
+
+## Results
+
+- Figure 1 overview
+- Figure 2 task result
+- Figure 3 task result
+- Figure 4 task result
+- Figure 5 task result
+
+## Discussion
+
+- strengths
+- biological implications
+- limitations
+- future directions
+
+## Methods
+
+- preprocessing
+- model
+- training
+- metrics
+- analyses
+- baselines
diff --git a/container/skills/bio-manuscript-common/templates/nature-methods/methods_template.md b/container/skills/bio-manuscript-common/templates/nature-methods/methods_template.md
new file mode 100644
index 0000000..d8ce1b0
--- /dev/null
+++ b/container/skills/bio-manuscript-common/templates/nature-methods/methods_template.md
@@ -0,0 +1,46 @@
+# Methods Template
+
+## Data
+
+- dataset sources
+- accession numbers
+- preprocessing assumptions
+
+## Preprocessing
+
+- QC
+- normalization
+- feature selection
+- dimensionality reduction
+
+## Model
+
+- architecture overview
+- key modules
+- latent space definition
+
+## Training
+
+- losses
+- optimization
+- scheduling
+- calibration / regularization
+
+## Evaluation
+
+- metrics
+- baselines
+- statistical testing
+
+## Biological Analyses
+
+- marker analysis
+- enrichment
+- spatial statistics
+- communication / GRN / trajectory if relevant
+
+## Reproducibility
+
+- software versions
+- config files
+- output artifacts
diff --git a/container/skills/bio-manuscript-pipeline/SKILL.md b/container/skills/bio-manuscript-pipeline/SKILL.md
new file mode 100644
index 0000000..74b03c6
--- /dev/null
+++ b/container/skills/bio-manuscript-pipeline/SKILL.md
@@ -0,0 +1,622 @@
+# bio-manuscript-pipeline
+
+**End-to-end pipeline from structured research input to a full manuscript plan (一条龙 Pipeline)**
+
+BioClaw integration notes:
+- This skill is staged under `container/skills/` as part of a multi-skill manuscript pipeline.
+- Shared templates and helper scripts are available under the sibling directory `bio-manuscript-common/`.
+- When this pipeline needs supporting capabilities, prefer the copied BioClaw sibling skills in `container/skills/` over any `~/.openclaw/...` layout assumptions.
+- This skill family is being integrated as a BioClaw community-contributed workflow.
+- Contributor reference for attribution and documentation: Yuhong Dong, Westlake University PhD candidate, BioClaw community contributor / BioClaw 共创团队 contributor.
+- In BioClaw, treat the sibling manuscript skills as stepwise companion skills. You should explicitly follow their guidance phase by phase rather than assuming an automatic runtime dispatcher.
+- If a later phase depends on outputs from an earlier phase, write those outputs into the group workspace first and then continue with the next sibling skill using those artifacts as context.
+- At the end of each substantial run, also write a concise human-readable execution summary. Prefer `FINAL_EXEC_SUMMARY.md`; for integration-focused runs, also write `INTEGRATION_TEST_REPORT.md`.
+
+---
+
+## Welcome
+
+Welcome to Bio-Manuscript-Forge. This workflow helps turn a rough research idea into a manuscript-ready planning package.
+
+### Input Template
+
+Provide your project in the following structure:
+
+```text
+topic: [research topic]
+base_work:
+  - paper: [related paper link]
+  - code: [related code repository]
+
+innovation: [one-sentence innovation summary]
+- algorithmic novelty (算法创新性): [core method novelty]
+- tasks (任务): [task1, task2, task3, ...]
+- data (数据): [dataset source or type]
+- benchmark: [evaluation benchmark]
+- metrics (计算指标): [metric1, metric2, ...]
+- biological analyses (生物学分析手段): [how biological significance will be shown]
+
+demo_data: [demo dataset link]
+target_journal: [optional, default nat-communications]
+num_refine_rounds: [optional, default 2]
+```
+
+### Example Input
+
+```
+topic: spatial multi-omics integration
+base_work:
+  - paper: https://www.nature.com/articles/s41592-021-01336-8
+  - code: https://github.com/broadinstitute/Tangram
+
+innovation: jointly align spatial transcriptomics and proteomics while preserving tissue-domain boundaries
+- algorithmic novelty (算法创新性): boundary-aware cross-modal alignment with explicit domain-consistency regularization
+- tasks (任务): cell annotation, spatial domain detection, cross-modal integration, biological interpretation
+- data (数据): public spatial transcriptomics and spatial proteomics cohorts with matched single-cell references
+- benchmark: compare against mapping, domain, and integration baselines on public tumor datasets
+- metrics (计算指标): ARI, NMI, Macro-F1, boundary preservation score, biological consistency
+- biological analyses (生物学分析手段):
+  - marker recovery across modalities
+  - pathway enrichment consistency
+  - neighborhood preservation
+  - tissue-boundary case studies
+
+demo_data: https://zenodo.org/record/0000000
+target_journal: nat-communications
+num_refine_rounds: 2
+```
+
+### Expected Outputs
+
+| File | Content |
+|------|---------|
+| **PPT** | Lab meeting / progress presentation |
+| **FINAL_PROPOSAL** | Full research proposal |
+| **Figure 2-7 (v3)** | Detailed task-wise figure designs |
+| **Manuscript text (v2)** | Introduction, Results, Discussion, Methods |
+
+---
+
+Provide the project description and the pipeline can begin.
+
+---
+
+## Purpose
+
+Run the full manuscript pipeline, generate a journal-style plan, and iteratively refine it through reviewer-style feedback.
+
+## Input Schema
+
+```text
+topic: [research topic]
+base_work: [paper links + code links]
+innovation: [high-level innovation summary]
+- algorithmic novelty (算法创新性): [core algorithmic novelty]
+- tasks (任务): [downstream tasks, comma-separated]
+- data (数据): [dataset source / type]
+- benchmark: [benchmark dataset or evaluation setup]
+- metrics (计算指标): [safety + task metrics such as ASR, ARI, etc.]
+- biological analyses (生物学分析手段): [marker genes, pathways, neighborhood analysis, etc.]
+
+demo_data: [demo dataset link]
+target_journal: [optional, default nat-communications]
+num_refine_rounds: [optional, default 2]
+```
+
+### Example Input (public-safe sample)
+
+```
+topic: spatial multi-omics integration
+base_work:
+  - paper: https://www.nature.com/articles/s41592-021-01336-8
+  - code: https://github.com/broadinstitute/Tangram
+
+innovation: jointly align spatial transcriptomics and proteomics while preserving tissue-domain boundaries
+- algorithmic novelty (算法创新性): boundary-aware cross-modal alignment with explicit domain-consistency regularization
+- tasks (任务): cell annotation, spatial domain detection, cross-modal integration, biological interpretation
+- data (数据): public spatial transcriptomics and spatial proteomics cohorts with matched single-cell references
+- benchmark: compare against mapping, domain, and integration baselines on public tumor datasets
+- metrics (计算指标): ARI, NMI, Macro-F1, boundary preservation score, biological consistency
+- biological analyses (生物学分析手段):
+  - marker recovery across modalities
+  - pathway enrichment consistency
+  - neighborhood preservation
+  - tissue-boundary case studies
+
+demo_data: https://zenodo.org/record/0000000
+target_journal: nat-communications
+num_refine_rounds: 2
+```
+
+### Field Guide
+
+| Field | Required | Description |
+|------|----------|-------------|
+| topic | yes | concise research topic |
+| base_work | yes | paper + code links |
+| innovation | yes | high-level idea plus structured subfields |
+| demo_data | yes | demo dataset link |
+| target_journal | no | default `nat-communications` |
+| num_refine_rounds | no | default `2` |
+
+### Innovation Subfields
+
+| Subfield | Description | Example |
+|----------|-------------|---------|
+| algorithmic novelty (算法创新性) | core method novelty | attention entropy, loss redesign, architecture change |
+| tasks (任务) | downstream tasks covered | cell annotation, perturbation, GRN inference |
+| data (数据) | dataset source / type | public cohorts, user data, target tissue |
+| benchmark | evaluation setup | existing benchmark or new benchmark |
+| metrics (计算指标) | safety + task metrics | ASR, Accuracy, F1, ARI, Pearson |
+| biological analyses (生物学分析手段) | how biology will be demonstrated | marker gene, pathway, regulatory links |
+
+## Execution Flow
+
+### Phase 1: System building (Steps 1-5)
+
+**Input parsing**: first extract the key signals from user input:
+- `topic` → 用于创新性搜索
+- `base_work` → 提取已有工作数据集、指标、方法
+- `innovation.algorithmic novelty` / `innovation.算法创新性` -> novelty assessment
+- `innovation.tasks` / `innovation.任务` -> task system design
+- `innovation.data` / `innovation.数据` -> dataset search direction
+- `innovation.metrics` / `innovation.计算指标` -> metric system design
+- `innovation.biological analyses` / `innovation.生物学分析手段` -> analysis system design
+
+```
+Step 1: 创新性检测
+├─ 解析输入：topic, base_work, innovation.算法创新性
+├─ 调用 searxng/web_search 搜索
+├─ Topic 同义变换生成 10-20 个变体
+├─ 搜索 PubMed + bioRxiv + arXiv q-bio
+├─ 统计相似文章数量
+├─ 结合 innovation.算法创新性 判断创新性级别
+└─ 输出：01_INNOVATION_ASSESSMENT.md
+
+Step 2: 任务体系构建
+├─ 解析输入：innovation.任务
+├─ 若用户提供任务列表 → 直接使用
+├─ 若未提供 → 搜索领域主要任务分类
+├─ 识别任务层级（Level 1-4）
+├─ 确保难度递进
+└─ 输出：02_TASK_SYSTEM.md
+
+Step 3: 数据集搜索
+├─ 解析输入：innovation.数据, innovation.benchmark, demo_data
+├─ 若用户提供数据描述 → 搜索匹配数据集
+├─ 从 base_work 论文提取数据集
+├─ 数据集与任务匹配
+└─ 输出：03_DATASET_CATALOG.md
+
+Step 4: 指标体系构建
+├─ 解析输入：innovation.计算指标
+├─ 若用户提供指标 → 直接使用并补充
+├─ 若未提供 → 从已有工作提取指标
+├─ 分类：安全指标 + 任务指标
+└─ 输出：04_METRIC_SYSTEM.md
+
+Step 5: 分析方法体系
+├─ 解析输入：innovation.生物学分析手段
+├─ 若用户提供分析手段 → 直接使用并补充
+├─ 若未提供 → 从已有工作提取分析方法
+├─ 标注 OmicsClaw/Bioclaw skill
+├─ 说明为什么用、证明什么、体现什么生物学意义
+└─ 输出：05_ANALYSIS_SYSTEM.md
+```
+
+### Phase 2: 设计与文案（Steps 6-7）
+
+**⚠️ 核心原则**：
+1. **任务为先**：Figure 2-N 每个对应一个任务，数据/指标/分析随任务而定
+2. **分析增强**：每个 Figure 必须包含安全 + 生物学分析
+3. **文案同步**：Figure 改完立即更新 Results
+
+```
+Step 6: Figure 设计
+│
+├─ Figure 1：算法创新性（方法框架）
+│   ├─ Panel a：方法 Overview
+│   ├─ Panel b：创新点示意
+│   ├─ Panel c：模型覆盖
+│   ├─ Panel d：任务覆盖
+│   └─ Panel e：指标体系
+│
+├─ Figure 2-N：每个 Figure = 一个任务 ⭐ 任务为先原则
+│   │
+│   ├─ Panel a: 任务 Overview（数据流）
+│   │
+│   ├─ Panel b-d: 定量测评
+│   │   ├─ 多模型对比
+│   │   ├─ ASR 降低
+│   │   └─ 任务指标保持
+│   │
+│   ├─ Panel e: Technical analysis ⭐ must include
+│   │   ├─ representation pattern shifts
+│   │   ├─ error / uncertainty analysis
+│   │   └─ failure-mode or boundary-case inspection
+│   │
+│   ├─ Panel f: 生物学分析 ⭐ 必须包含
+│   │   ├─ Marker gene recovery
+│   │   ├─ Pathway preservation
+│   │   └─ 具体生物学意义
+│   │
+│   ├─ Panel g: In-depth case studies ⭐ 1-2 cases
+│   │   ├─ concrete biological question
+│   │   ├─ baseline vs proposed method comparison
+│   │   └─ interpretation of recovered biological structure
+│   │
+│   └─ 数据/指标/分析依据任务选取
+│
+├─ Figure N+1: Summary + 生物学意义总结
+│
+└─ 输出：06_FIGURE_DESIGNS/
+
+Step 6.5: 文案同步检查 ⭐ 必须
+├─ Figure 有这个 Panel → Results 有对应段落？
+├─ Figure 有这个案例 → Results 有详细展开？
+└─ 检查通过才能进入下一步
+
+Step 7: 论文文案生成
+│
+├─ Introduction（5段）
+│   ├─ 第一段：领域介绍
+│   ├─ 第二段：相关工作调研
+│   ├─ 第三段：现有方法不足
+│   ├─ 第四段：本文方法介绍
+│   └─ 第五段：意义与应用
+│
+├─ Results（与 Figure 对应）⭐ 结构对齐
+│   ├─ 2.1 Overview（对应 Figure 1）
+│   ├─ 2.2 Task 1 / Main claim（对应 Figure 2）
+│   │   ├─ quantitative evaluation
+│   │   ├─ technical analysis
+│   │   ├─ biological analysis
+│   │   └─ case study
+│   ├─ 2.3 Task 2 / Main claim（对应 Figure 3）
+│   ├─ ...每个任务一个 section
+│   └─ 2.N Summary（对应最后一个 Figure）
+│
+├─ Discussion
+│   ├─ 方法优势总结
+│   ├─ 安全-生物学结合意义 ⭐
+│   ├─ 与现有方法对比
+│   ├─ 方法局限性
+│   └─ 未来方向
+│
+├─ Methods
+│   ├─ 数据预处理
+│   ├─ 模型架构
+│   ├─ 任务特定方法 ⭐ 按任务组织
+│   ├─ 生物学分析方法 ⭐
+│   ├─ 统计分析
+│   └─ 代码与数据可用性
+│
+└─ 输出：07_MANUSCRIPT_TEXT/
+```
+
+**Figure 设计检查清单**：
+
+```
+- [ ] Figure 1 是方法框架？
+- [ ] Figure 2-N 每个对应一个任务？
+- [ ] 每个 Figure 包含多模型对比？
+- [ ] 每个 Figure 有定量测评 Panel？
+- [ ] 每个 Figure 有安全分析 Panel？ ⭐
+- [ ] 每个 Figure 有生物学分析 Panel？ ⭐
+- [ ] 每个 Figure 有 1-2 个深入案例？ ⭐
+- [ ] 分析手段多样化？
+- [ ] Results 结构与 Figure 对应？ ⭐
+```
+
+### Phase 2.5: Refine Loop ⭐
+
+```
+Step 7.5: 三审稿人迭代优化
+│
+├─ Round 0: 保存初始方案
+│   └─ 输出：refine-logs/round-0-initial-proposal.md
+│
+├─ Round 1 Review:
+│   ├─ Editor Review（创新性评估，Nature子刊标准）
+│   │   ├─ 概念创新 / 方法创新 / 应用创新
+│   │   └─ 评分：创新性 / 可行性 / 推荐度
+│   │
+│   ├─ 计算审稿人 Review（算法/方法评审）
+│   │   ├─ 算法设计合理性 / 方法创新性
+│   │   ├─ 实验设计严谨性（Baseline/指标/Ablation）
+│   │   └─ 评分：方法创新 / 技术严谨 / 代码可行
+│   │
+│   ├─ 生物分析审稿人 Review（生物学意义评审）
+│   │   ├─ 生物学意义 / 分析设计合理性
+│   │   ├─ 数据集选择合理性
+│   │   └─ 评分：生物意义 / 分析设计 / 数据选择
+│   │
+│   └─ 输出：refine-logs/round-1/
+│
+├─ Round 1 Refinement:
+│   ├─ 汇总三审稿人意见
+│   ├─ 问题分类（Critical/Major/Minor）
+│   ├─ 逐条响应和修改
+│   ├─ 更新 Proposal
+│   └─ 输出：refine-logs/round-1/refinement.md
+│
+├─ Round 2 Review:（同 Round 1）
+│   └─ 输出：refine-logs/round-2/
+│
+├─ Round 2 Refinement:
+│   └─ 输出：refine-logs/round-2/refinement.md
+│
+└─ 最终输出：
+    ├─ refine-logs/REVIEW_SUMMARY.md（每轮汇总）
+    ├─ refine-logs/FINAL_PROPOSAL.md（最终方案）
+    ├─ refine-logs/score-history.md（评分历史）
+    └─ refine-logs/REFINEMENT_REPORT.md（完整报告）
+```
+
+### Phase 2.6: 人类反馈验证 ⭐ NEW
+
+```
+Step 7.6: 人类反馈循环
+│
+├─ 呈现 Proposal
+│   ├─ 展示 FINAL_PROPOSAL.md 核心内容
+│   ├─ 包含：创新点、Figure 设计、实验方案、关键修改
+│   └─ 格式：结构化摘要 + 关键决策点
+│
+├─ 等待人类反馈
+│   ├─ 选项 A: 同意 → 继续 Phase 3
+│   └─ 选项 B: 有意见 → 收集反馈内容
+│
+├─ 反馈处理
+│   ├─ 如果同意 → 记录并进入 Phase 3
+│   └─ 如果不同意 → 
+│       ├─ 记录反馈意见到 refine-logs/human-feedback/
+│       ├─ 根据反馈类型决定返回点：
+│       │   ├─ Phase 1 级问题：创新性/任务体系需重构
+│       │   ├─ Phase 2 级问题：Figure/文案需调整
+│       │   └─ Phase 2.5 级问题：细节优化
+│       ├─ 执行迭代修改
+│       ├─ 重新运行 Phase 2.5 Refine Loop
+│       └─ 再次呈现给人类验证
+│
+└─ 输出：
+    ├─ refine-logs/human-feedback/feedback-round-X.md
+    └─ refine-logs/HUMAN_APPROVAL.md（最终批准记录）
+```
+
+**人类反馈处理流程：**
+
+```
+人类反馈 → 问题分类 → 返回点决策
+│
+├─ Critical 问题（创新性方向错误）
+│   └─ 返回 Phase 1 → 重新评估创新点
+│
+├─ Major 问题（设计/方案需要大改）
+│   └─ 返回 Phase 2 → 调整 Figure/文案
+│
+├─ Minor 问题（细节优化）
+│   └─ 返回 Phase 2.5 → Refine Loop
+│
+└─ 批准
+    └─ 进入 Phase 3
+```
+
+**反馈收集格式：**
+
+```markdown
+## 人类反馈 Round X
+
+**反馈时间**: YYYY-MM-DD HH:MM
+**反馈内容**: [用户意见]
+**问题级别**: Critical / Major / Minor
+**返回阶段**: Phase 1 / Phase 2 / Phase 2.5
+**修改建议**: [AI 分析后的修改方案]
+
+---
+
+## 修改执行记录
+
+- [ ] 修改项 1
+- [ ] 修改项 2
+...
+```
+
+### Phase 3: 验证与汇报（Steps 8-11）
+
+```
+Step 8: 代码修改方案
+├─ 克隆原有代码仓库
+├─ 分析代码结构
+├─ 映射创新点到修改位置
+├─ 设计新增文件 + 修改文件
+└─ 输出：08_CODE_MODIFICATION_PLAN.md
+
+Step 9: Demo 快速验证
+├─ 应用代码修改
+├─ 下载 Demo 数据
+├─ Subsample + 少 epoch 快速运行
+├─ 可行性判断
+├─ 如果不可行 → 修改建议
+└─ 输出：09_DEMO_VALIDATION.md
+
+Step 10: 详细分析执行（可选）
+├─ 调用 OmicsClaw/Bioclaw
+├─ 运行完整分析
+├─ 生成实际数据
+└─ 输出：10_ANALYSIS_RESULTS.md
+
+Step 11: 生成组会汇报 PPT（⭐ 新增）
+├─ 从 FINAL_PROPOSAL.md 提取核心内容
+├─ 从 DEMO_VALIDATION.md 提取 Demo 结果
+├─ 生成 12-15 页组会汇报 PPT
+├─ 格式：Markdown (Marp) / HTML (reveal.js) / PPTX
+└─ 输出：11_PPT_PRESENTATION.md
+
+Step 12: 执行总结与汇报摘要（⭐ BioClaw 集成建议）
+├─ 汇总本次实际跑过的阶段
+├─ 汇总关键输出文件与路径
+├─ 标注哪些步骤真正跑通、哪些仅为草案/脚手架
+├─ 标注当前 blocker
+├─ 给出下一步建议（最多 3 条）
+├─ 记录适合集成汇报的结论
+└─ 输出：FINAL_EXEC_SUMMARY.md
+```
+
+## 输出目录结构
+
+```
+manuscript-plan/
+├── 01_INNOVATION_ASSESSMENT.md
+├── 02_TASK_SYSTEM.md
+├── 03_DATASET_CATALOG.md
+├── 04_METRIC_SYSTEM.md
+├── 05_ANALYSIS_SYSTEM.md
+│
+├── 06_FIGURE_DESIGNS/
+│   ├── FIGURE_1_DESIGN.md
+│   ├── FIGURE_2_DESIGN.md
+│   ├── FIGURE_3_DESIGN.md
+│   ├── FIGURE_4_DESIGN.md
+│   ├── FIGURE_5_DESIGN.md
+│   └── SUPPLEMENTARY_DESIGN.md
+│
+├── 07_MANUSCRIPT_TEXT/
+│   ├── INTRODUCTION.md
+│   ├── RESULTS.md
+│   ├── DISCUSSION.md
+│   └── METHODS.md
+│
+├── refine-logs/                    # ⭐ 新增
+│   ├── round-0-initial-proposal.md
+│   │
+│   ├── round-1/
+│   │   ├── editor-review.md
+│   │   ├── computational-review.md
+│   │   ├── biological-review.md
+│   │   ├── review-summary.md
+│   │   └── refinement.md
+│   │
+│   ├── round-2/
+│   │   ├── editor-review.md
+│   │   ├── computational-review.md
+│   │   ├── biological-review.md
+│   │   ├── review-summary.md
+│   │   └── refinement.md
+│   │
+│   ├── human-feedback/              # ⭐ NEW: 人类反馈记录
+│   │   ├── feedback-round-1.md
+│   │   ├── feedback-round-2.md
+│   │   └── ...
+│   │
+│   ├── REVIEW_SUMMARY.md
+│   ├── FINAL_PROPOSAL.md
+│   ├── HUMAN_APPROVAL.md            # ⭐ NEW: 人类批准记录
+│   ├── score-history.md
+│   └── REFINEMENT_REPORT.md
+│
+├── 08_CODE_MODIFICATION_PLAN.md
+├── 09_DEMO_VALIDATION.md
+├── 10_ANALYSIS_RESULTS.md
+│
+├── 11_PPT_PRESENTATION.md           # ⭐ 新增：组会汇报 PPT
+├── FINAL_EXEC_SUMMARY.md            # ⭐ 新增：面向人类汇报的执行摘要
+├── INTEGRATION_TEST_REPORT.md       # ⭐ 可选：集成/验证测试报告
+│
+└── FINAL_MANUSCRIPT_PLAN.md
+```
+
+## 执行摘要模板
+
+每次较完整运行结束后，补一个汇报友好的摘要文件，至少覆盖以下内容：
+
+```md
+# FINAL_EXEC_SUMMARY
+
+## Run Scope
+- Topic:
+- Date:
+- Workspace:
+- Pipeline entry:
+
+## Stages Executed
+- Step / Phase:
+- Step / Phase:
+
+## Key Files Generated
+- path/to/file
+- path/to/file
+
+## Verified Outputs
+- What actually ran successfully
+- What was only drafted / scaffolded
+
+## Current Blockers
+- blocker 1
+- blocker 2
+
+## Recommended Next Steps
+1. ...
+2. ...
+3. ...
+
+## Attribution
+- Workflow family: Bio-Manuscript-Forge
+- BioClaw integration: community-contributed workflow
+- Contributor reference: Yuhong Dong, Westlake University PhD candidate, BioClaw community contributor
+```
+
+## 三审稿人评审标准
+
+### Editor（编辑）
+- **职责**：初审，判断是否达到 Nature 子刊水平
+- **评审维度**：创新性、可行性、期刊匹配度
+- **评分**：创新性/10、可行性/10、推荐意见
+
+### 计算审稿人
+- **职责**：从计算/算法角度评审
+- **评审维度**：算法设计、方法创新、实验严谨性、代码可行性
+- **评分**：方法创新/10、技术严谨/10、代码可行/10
+
+### 生物分析审稿人
+- **职责**：从生物学/分析角度评审
+- **评审维度**：生物学意义、分析设计、数据选择
+- **评分**：生物意义/10、分析设计/10、数据选择/10
+
+## 使用方式
+
+```bash
+/bio-manuscript-pipeline "topic: spatial multi-omics integration | base_work: https://github.com/example/project | innovation: boundary-aware cross-modal alignment | demo_data: https://example.com/data.h5ad | target_journal: nat-communications | num_refine_rounds: 2"
+```
+
+## 子 Skill 调用
+
+本 Pipeline 会依次调用以下子 Skill：
+- `bio-innovation-check`（Step 1）
+- `bio-task-system`（Step 2）
+- `bio-dataset-search`（Step 3）
+- `bio-metric-system`（Step 4）
+- `bio-analysis-system`（Step 5）
+- `bio-figure-design`（Step 6）
+- `bio-manuscript-text`（Step 7）
+- `bio-manuscript-refine`（Step 7.5）⭐
+- `bio-human-feedback`（Step 7.6）⭐ NEW - 人类反馈验证
+- `bio-code-modification`（Step 8）
+- `bio-demo-validate`（Step 9）
+- `bio-ppt-generate`（Step 11）⭐
+
+## 注意事项
+
+1. **Phase 1 完成后**：检查创新性评估结果
+2. **Phase 2 完成后**：检查 Figure 设计和文案
+3. **Phase 2.5（Refine Loop）**：每轮评分需达到 7+ 才能进入下一阶段
+4. **Phase 2.6（人类反馈验证）**：⭐ 关键检查点
+   - 呈现 FINAL_PROPOSAL.md 给人类审阅
+   - 必须等待人类明确反馈
+   - 同意 → 继续 Phase 3
+   - 不同意 → 根据问题级别返回对应阶段迭代
+   - 所有反馈记录到 refine-logs/human-feedback/
+5. **Phase 3**：Demo 验证如果不可行，回到 Step 8 重新设计
+6. **迭代收敛**：通常 2 轮 Refine 后评分趋于稳定
+7. **最终检查**：使用 FINAL_PROPOSAL.md 作为执行依据
+8. **人类批准**：必须有人类批准记录（HUMAN_APPROVAL.md）才能进入 Phase 3
diff --git a/container/skills/bio-manuscript-refine/SKILL.md b/container/skills/bio-manuscript-refine/SKILL.md
new file mode 100644
index 0000000..f42c670
--- /dev/null
+++ b/container/skills/bio-manuscript-refine/SKILL.md
@@ -0,0 +1,117 @@
+# bio-manuscript-refine
+
+**Refine loop: three-reviewer iterative refinement (三审稿人迭代优化)**
+
+Run a reviewer-style refinement loop over the manuscript plan using three perspectives: editor, computational reviewer, and biological reviewer.
+
+## Purpose
+
+1. Review the current manuscript plan
+2. Produce structured review comments
+3. Revise the proposal round by round
+4. Track score history and revision history
+
+## Input Format
+
+```text
+manuscript_plan: [full manuscript plan generated in previous steps]
+target_journal: [target journal, default nat-communications]
+num_rounds: [number of refine rounds, default 2]
+```
+
+## Workflow
+
+### Round 0
+
+- save the initial proposal snapshot
+
+### Each review round
+
+Produce three reviews:
+
+1. **Editor**
+   - novelty
+   - feasibility
+   - journal fit
+2. **Computational reviewer**
+   - method design
+   - technical rigor
+   - benchmark quality
+   - implementation feasibility
+3. **Biological reviewer**
+   - biological significance
+   - analysis design
+   - dataset suitability
+
+Then generate:
+
+- a review summary
+- a revision response
+- a refined proposal
+
+## Output Format
+
+```markdown
+# Refine Report
+
+## Round 0
+- initial proposal snapshot
+
+## Round 1 Reviews
+### Editor
+- scores:
+- key concerns:
+
+### Computational Reviewer
+- scores:
+- key concerns:
+
+### Biological Reviewer
+- scores:
+- key concerns:
+
+## Round 1 Revision
+- addressed concerns:
+- remaining risks:
+
+## Score History
+| Round | Editor | Computational | Biological | Overall |
+|-------|--------|---------------|------------|---------|
+| 0 | ... | ... | ... | ... |
+
+## Final Proposal Status
+- ready for next phase / needs more revision
+```
+
+## Reviewer Criteria
+
+### Editor
+
+- novelty
+- feasibility
+- journal fit
+
+### Computational reviewer
+
+- algorithmic soundness
+- method novelty
+- benchmark rigor
+- code feasibility
+
+### Biological reviewer
+
+- biological significance
+- analysis relevance
+- dataset realism
+
+## Usage
+
+```bash
+/bio-manuscript-refine "manuscript_plan: [path to proposal] | target_journal: nat-communications | num_rounds: 2"
+```
+
+## Notes
+
+1. Revision should update the proposal itself, not only append comments.
+2. Keep a full history of each round.
+3. Scores are guidance, not absolute truth; comments matter more than raw numbers.
diff --git a/container/skills/bio-manuscript-text/SKILL.md b/container/skills/bio-manuscript-text/SKILL.md
new file mode 100644
index 0000000..f03b913
--- /dev/null
+++ b/container/skills/bio-manuscript-text/SKILL.md
@@ -0,0 +1,127 @@
+# bio-manuscript-text
+
+**Step 7: Manuscript drafting (论文文案生成)**
+
+Draft the main manuscript text from the figure plan, metric system, and analysis system.
+
+## Purpose
+
+1. Write the Introduction
+2. Draft the Results around the figure logic
+3. Draft the Discussion
+4. Draft the Methods skeleton
+
+## Input Format
+
+```text
+topic: [research topic]
+figure_designs: [figure design document]
+innovation: [innovation summary]
+base_work: [related prior work]
+metric_system: [metric system]
+analysis_system: [analysis system]
+paper_count: [number of related papers]
+related_papers: [optional list]
+target_journal: [target journal]
+```
+
+## Workflow
+
+### Step 7.1: Write the Introduction
+
+Use a five-paragraph structure:
+
+1. field background
+2. related work
+3. limitations of current methods
+4. introduce the proposed method
+5. significance and broader impact
+
+### Step 7.2: Write the Results
+
+Organize the Results around figure order:
+
+- Figure 1: framework / method overview
+- Figure 2-N: one main task or claim per figure
+
+For each figure section, explain:
+
+- setup
+- quantitative findings
+- qualitative or biological findings
+- take-home message
+
+### Step 7.3: Write the Discussion
+
+Cover:
+
+- method strengths
+- comparison to prior methods
+- biological implications
+- limitations
+- future directions
+
+### Step 7.4: Write the Methods skeleton
+
+Include at minimum:
+
+- preprocessing
+- model architecture
+- training strategy
+- metrics
+- biological analyses
+- baselines
+
+## Output Format
+
+```markdown
+# Manuscript Text
+
+## INTRODUCTION
+### Paragraph 1: Field background
+### Paragraph 2: Related work
+### Paragraph 3: Current limitations
+### Paragraph 4: Our method
+### Paragraph 5: Significance
+
+## RESULTS
+### 2.1 Framework overview
+### 2.2 Task 1 / Figure 2
+### 2.3 Task 2 / Figure 3
+### 2.4 Task 3 / Figure 4
+### 2.5 Task 4 / Figure 5
+
+## DISCUSSION
+- strengths
+- comparisons
+- biological implications
+- limitations
+- future work
+
+## METHODS
+- preprocessing
+- model
+- training
+- metrics
+- biological analyses
+- baselines
+```
+
+## Writing Principles
+
+1. Make the manuscript track the figure logic.
+2. Keep claims tied to evidence.
+3. Separate technical and biological claims clearly.
+4. Prefer reviewer-friendly clarity over stylistic complexity.
+
+## Usage
+
+```bash
+/bio-manuscript-text "topic: spatial multi-omics integration | figure_designs: [...] | innovation: [...] | base_work: [...] | paper_count: 5 | target_journal: nat-communications"
+```
+
+## Notes
+
+1. Do not draft text before the figure logic is stable.
+2. Every Results subsection should map to a figure.
+3. Keep Methods detailed enough that later implementation planning remains consistent.
diff --git a/container/skills/bio-metric-system/SKILL.md b/container/skills/bio-metric-system/SKILL.md
new file mode 100644
index 0000000..5af5e0f
--- /dev/null
+++ b/container/skills/bio-metric-system/SKILL.md
@@ -0,0 +1,149 @@
+# bio-metric-system
+
+**Step 4: Metric system design (评价指标体系构建)**
+
+Build a defensible set of quantitative and qualitative metrics by extracting them from related work or adapting them from adjacent fields.
+
+## Purpose
+
+1. Extract evaluation metrics from existing literature
+2. Borrow metrics from adjacent domains when needed
+3. Organize metrics into quantitative and qualitative groups
+4. Explain what each metric measures and how it should be computed
+
+## Input Format
+
+```text
+topic: [research topic]
+paper_count: [number of related papers]
+task_system: [task system from Step 2]
+```
+
+## Workflow
+
+### Step 4.1: Extract metrics from existing work
+
+If `paper_count >= 5`, review the Results / Benchmark sections of the strongest related papers and extract:
+
+- metric name
+- what it evaluates
+- formula or computation method
+- expected range
+- how often it appears in the field
+
+### Step 4.2: Borrow metrics from adjacent domains
+
+If the literature is still thin, adapt metrics from a nearby field.
+
+Examples:
+
+- clustering agreement -> ARI / NMI
+- modality agreement -> Pearson / cosine similarity
+- reconstruction / registration -> MSE / MAE
+- biological relevance -> marker recovery / enrichment scores
+
+### Step 4.3: Organize the metric system
+
+Split metrics into:
+
+- **Quantitative metrics**
+  - integration quality
+  - modality consistency
+  - registration / alignment quality
+  - biological agreement
+- **Qualitative metrics**
+  - spatial plots
+  - feature plots
+  - latent visualizations
+  - heatmaps
+  - pathway / enrichment figures
+
+### Step 4.4: Standardize each metric
+
+For each metric, define:
+
+- English name
+- optional Chinese reference in parentheses
+- category
+- what it measures
+- formula (if needed)
+- range / interpretation
+- software implementation
+- task relevance
+- mapped figure / panel
+
+## Output Format
+
+```markdown
+# Metric System
+
+## Metric Sources
+- Extracted from related papers:
+- Borrowed from adjacent domains:
+
+## Quantitative Metrics
+
+### ARI (Adjusted Rand Index)
+- Category:
+- What it measures:
+- Formula:
+- Range:
+- Interpretation:
+- Implementation:
+- Relevant tasks:
+- Figure mapping:
+
+### NMI (Normalized Mutual Information)
+- Category:
+- What it measures:
+- Formula:
+- Range:
+- Interpretation:
+- Implementation:
+- Relevant tasks:
+- Figure mapping:
+
+### Pearson correlation
+- Category:
+- What it measures:
+- Formula:
+- Range:
+- Interpretation:
+- Implementation:
+- Relevant tasks:
+- Figure mapping:
+
+## Qualitative Metrics / Visual Readouts
+- spatial domain map
+- feature plot
+- violin plot
+- UMAP / latent visualization
+- heatmap
+- pathway enrichment figure
+
+## Next Step
+- Use the metric system to build the analysis system in Step 5
+```
+
+## Recommended Core Metrics
+
+For most manuscript-planning runs, include at least:
+
+- ARI
+- NMI
+- Macro-F1 or annotation accuracy
+- Pearson / cosine similarity when cross-modal agreement matters
+- MSE / MAE when reconstruction or alignment quality matters
+- at least one biological validation readout
+
+## Usage
+
+```bash
+/bio-metric-system "spatial multi-omics integration | paper_count: 5 | task_system: [task system from Step 2]"
+```
+
+## Notes
+
+1. Do not overload the paper with too many metrics; prefer a compact but defendable set.
+2. Match each metric to a specific task claim.
+3. Include at least one metric that reflects biological value, not just technical fit.
diff --git a/container/skills/bio-ppt-generate/SKILL.md b/container/skills/bio-ppt-generate/SKILL.md
new file mode 100644
index 0000000..25e9baa
--- /dev/null
+++ b/container/skills/bio-ppt-generate/SKILL.md
@@ -0,0 +1,84 @@
+# bio-ppt-generate
+
+**Presentation generation (组会汇报 PPT)**
+
+Generate a concise presentation package from the final proposal and demo / validation outputs.
+
+## Purpose
+
+1. Extract the core message from `FINAL_PROPOSAL.md`
+2. Incorporate demo or validation results
+3. Produce a 10-15 slide presentation outline
+4. Output Markdown-first slides that can be converted later
+
+## Input Format
+
+```text
+final_proposal: [path to FINAL_PROPOSAL.md]
+demo_result: [path to DEMO_VALIDATION.md or equivalent]
+ppt_title: [presentation title]
+author: [author]
+date: [date]
+```
+
+## Suggested Slide Order
+
+1. Title
+2. Background
+3. Research question
+4. Innovation
+5. Method overview
+6. Task design
+7. Data and metrics
+8. Demo / validation results
+9. Expected figure set
+10. Biological significance
+11. Next steps
+12. Summary
+13. Q&A
+
+## Output Format
+
+Preferred output:
+
+- Markdown slide deck
+- optionally HTML / reveal.js structure
+- optionally a PowerPoint-style outline
+
+```markdown
+---
+marp: true
+theme: gaia
+paginate: true
+---
+
+# [Title]
+
+**[Author]**
+
+[Date]
+
+---
+
+# Background
+
+- ...
+
+---
+
+# Research Question
+
+- ...
+```
+
+## Usage
+
+```bash
+/bio-ppt-generate "final_proposal: FINAL_PROPOSAL.md | demo_result: DEMO_VALIDATION.md | ppt_title: Project Update | author: Name | date: 2026-04-04"
+```
+
+## Notes
+
+1. Keep the deck concise and review-friendly.
+2. One slide should carry one main idea.
+3. Demo evidence should be included if available.
diff --git a/container/skills/bio-task-system/SKILL.md b/container/skills/bio-task-system/SKILL.md
new file mode 100644
index 0000000..97033c5
--- /dev/null
+++ b/container/skills/bio-task-system/SKILL.md
@@ -0,0 +1,118 @@
+# bio-task-system
+
+**Step 2: Task system design (任务体系构建)**
+
+Identify the main task categories in the field and organize them into a staged difficulty ladder.
+
+## Purpose
+
+1. Find the dominant task taxonomy in the target field
+2. Define Level 1-4 task tiers
+3. Ensure the task ladder increases in difficulty
+4. Prepare the task system for downstream dataset and metric design
+
+## Input Format
+
+```text
+topic: [research topic]
+paper_count: [number of related papers from Step 1]
+```
+
+## Workflow
+
+### Step 2.1: Task taxonomy search
+
+If there is substantial prior work, extract tasks from existing papers.
+
+If there is not enough prior work, borrow the taxonomy from a parent domain and adapt it.
+
+Typical adaptation logic:
+
+- single-cell multi-omics -> spatial multi-omics
+- modality alignment -> spatial-cell alignment
+- batch integration -> cross-sample integration
+
+### Step 2.2: Task tier design
+
+Define four levels:
+
+- **Level 1**: basic validation task
+- **Level 2**: intermediate application task
+- **Level 3**: challenge task
+- **Level 4**: flagship innovation task
+
+Increase across three dimensions:
+
+1. Data complexity
+2. Technical difficulty
+3. Biological value
+
+### Step 2.3: Standardize task descriptions
+
+For each task, write:
+
+- definition
+- difficulty level
+- data requirements
+- technical focus
+- biological value
+- representative methods
+- mapped figure
+
+## Output Format
+
+```markdown
+# Task System Design
+
+## Task Sources
+- Extracted from related papers:
+- Borrowed from parent domain:
+
+## Tier Overview
+| Level | Task type | Difficulty | Figure |
+|-------|-----------|------------|--------|
+| 1 | ... | low | Figure 2 |
+| 2 | ... | medium | Figure 3 |
+| 3 | ... | high | Figure 4 |
+| 4 | ... | highest | Figure 5 |
+
+## Detailed Task Descriptions
+
+### Task 1: [task name]
+- Definition:
+- Difficulty:
+- Data requirements:
+- Technical focus:
+- Biological value:
+- Representative methods:
+- Mapped figure:
+
+### Task 2: ...
+
+## Progression Rationale
+1. Data complexity rises across tasks
+2. Technical difficulty rises across tasks
+3. Biological value rises across tasks
+
+## Next Step
+- Use the task system to search for datasets in Step 3
+```
+
+## Example Ladder
+
+- Level 1: vertical integration
+- Level 2: horizontal / cross-slice integration
+- Level 3: mosaic integration with missing modalities
+- Level 4: diagonal integration across platform / resolution / cohort
+
+## Usage
+
+```bash
+/bio-task-system "spatial multi-omics integration | paper_count: 5"
+```
+
+## Notes
+
+1. Keep the ladder interpretable to reviewers.
+2. Avoid adding too many tasks; four well-designed tiers are usually enough.
+3. Make sure each task can later be tied to datasets, metrics, and figures.