scientify 1.2.1 → 1.3.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "scientify",
-  "version": "1.2.1",
+  "version": "1.3.0",
   "description": "Scientify - AI-powered research workflow automation for OpenClaw. Includes idea generation, literature review, research pipeline skills, and arxiv tool.",
   "type": "module",
   "main": "dist/index.js",

package/skills/_shared/workspace-spec.md

@@ -0,0 +1,129 @@
+# Workspace Directory Specification
+
+All Scientify skills share a unified project-based workspace structure.
+
+## Base Path
+
+```
+~/.openclaw/workspace/projects/
+├── .active                      # Current project ID (plain text)
+├── {project-id}/                # Each research topic has its own project
+│   └── ...
+└── {another-project}/
+```
+
+## Project Structure
+
+```
+~/.openclaw/workspace/projects/{project-id}/
+├── project.json                 # Project metadata
+├── task.json                    # Research task definition
+│
+├── survey/                      # /literature-survey outputs
+│   ├── search_terms.json        # Generated search keywords
+│   ├── raw_results.json         # All search results
+│   ├── filtered_papers.json     # Papers with relevance scores
+│   ├── clusters.json            # Clustered by research direction
+│   └── report.md                # Final survey report
+│
+├── papers/                      # Downloaded paper sources
+│   ├── {direction-1}/           # Organized by cluster
+│   │   ├── paper_list.md
+│   │   └── {arxiv_id}/          # .tex source files
+│   ├── {direction-2}/
+│   └── uncategorized/
+│
+├── repos/                       # Cloned reference repositories
+│   ├── {repo-name-1}/
+│   └── {repo-name-2}/
+│
+├── ideas/                       # /idea-generation outputs
+│   ├── gaps.md                  # Identified research gaps
+│   ├── idea_1.md ... idea_5.md  # Generated ideas
+│   ├── selected_idea.md         # Enhanced best idea
+│   ├── implementation_report.md # Code mapping
+│   └── summary.md               # Final summary
+│
+├── review/                      # /write-review-paper outputs
+│   ├── reading_plan.md          # Prioritized reading list
+│   ├── notes/                   # Per-paper reading notes
+│   │   └── {paper_id}.md
+│   ├── comparison.md            # Method comparison table
+│   ├── timeline.md              # Research timeline
+│   ├── taxonomy.md              # Classification system
+│   ├── draft.md                 # Survey paper draft
+│   └── bibliography.bib         # References
+│
+├── plan_res.md                  # /research-pipeline: implementation plan
+├── project/                     # /research-pipeline: code implementation
+│   ├── model/
+│   ├── data/
+│   ├── training/
+│   ├── testing/
+│   ├── run.py
+│   └── requirements.txt
+├── iterations/                  # Review iterations
+│   ├── judge_v1.md
+│   └── judge_v2.md
+└── experiment_res.md            # Final results
+```
+
+## Conventions
+
+### File Existence = Step Completion
+
+Check output file before executing any step. If exists, skip.
+
+Enables:
+- **Crash recovery**: resume from last completed step
+- **Incremental progress**: rerunning skips completed work
+- **Transparency**: inspect progress by listing directory
+
+### Project Metadata
+
+**project.json:**
+```json
+{
+  "id": "battery-rul-prediction",
+  "name": "Battery RUL Prediction",
+  "created": "2024-01-15T10:00:00Z",
+  "topics": ["battery", "remaining useful life", "prediction"]
+}
+```
+
+**task.json:**
+```json
+{
+  "domain": "battery health",
+  "focus": "RUL prediction using transformer",
+  "date_limit": "2024-01-01",
+  "created": "2024-01-15"
+}
+```
+
+### Immutability
+
+Once written, do NOT modify outputs unless user explicitly asks.
+Exception: `project/` is mutable during implement-review-iterate loop.
+
+### Active Project
+
+```bash
+# Read active project
+cat ~/.openclaw/workspace/projects/.active
+
+# Set active project
+echo "battery-rul-prediction" > ~/.openclaw/workspace/projects/.active
+
+# Set $WORKSPACE variable
+WORKSPACE=~/.openclaw/workspace/projects/$(cat ~/.openclaw/workspace/projects/.active)
+```
+
+## Skill Outputs Summary
+
+| Skill | Primary Outputs |
+|-------|-----------------|
+| `/literature-survey` | `survey/`, `papers/` |
+| `/idea-generation` | `ideas/` |
+| `/write-review-paper` | `review/` |
+| `/research-pipeline` | `project/`, `iterations/`, `experiment_res.md` |

package/skills/idea-generation/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: idea-generation
-description: "Generate innovative research ideas from a topic. SEARCHES arXiv/GitHub automatically, downloads papers, analyzes literature, outputs 5 novel ideas with citations. Use for: 找研究方向, 生成创新点, find research gaps, propose new methods. NOT for: summarizing papers (use /write-review-paper), literature survey (use /literature-survey)."
+description: "Generate 5 innovative research ideas from collected papers. Analyzes literature, identifies gaps, proposes novel methods with citations. Use for: 找研究方向, 生成创新点, find research gaps. Requires papers in workspace (run /literature-survey first if needed)."
 metadata:
   {
     "openclaw":
@@ -13,389 +13,168 @@ metadata:
 
 # Idea Generation
 
-End-to-end workflow for generating innovative research ideas from a research topic. This skill implements a full research idea generation pipeline:
+Generate innovative research ideas grounded in literature analysis. This skill reads existing papers, identifies research gaps, and produces 5 distinct ideas with citations.
 
-1. Search papers and code repositories
-2. Select and download references
-3. Analyze literature and codebases
-4. Generate multiple ideas
-5. Select and enhance the best idea
-6. Map to code implementations
+**Core principle:** Ideas MUST be grounded in actual papers, not generated from model knowledge.
 
----
-
-## ⚠️ CRITICAL: EXECUTION MODE
-
-**AUTONOMOUS EXECUTION**: Execute ALL steps without asking for user confirmation at each step.
-- Do NOT ask "要我继续吗?" or "Should I proceed?"
-- You MAY spawn subagents for parallel tasks (e.g., downloading multiple papers)
-- Only ask user when there's a genuine ambiguity (e.g., which focus area to choose)
-- Checkpoints are for YOUR internal verification, not for asking user
-
-**Run the entire workflow from Step 1 to Step 8 automatically.**
+**Workspace:** See `../_shared/workspace-spec.md` for directory structure. Outputs go to `$WORKSPACE/ideas/`.
 
 ---
 
-## ⚠️ CRITICAL: MANDATORY TOOL USAGE
-
-**DO NOT generate ideas from your own knowledge.** All ideas MUST be grounded in actual literature research.
-
-### Blocking Requirements
-
-1. **MUST call `arxiv` tool** to search papers - NO EXCEPTIONS
-2. **MUST call `github_search` tool** to find repositories - NO EXCEPTIONS
-3. **MUST write `search_results.md`** BEFORE proceeding to idea generation
-4. **MUST reference specific papers** (with arXiv IDs) in generated ideas
-5. **MUST clone actual repos** before code survey
-
-### Anti-Pattern: DO NOT DO THIS
+## Step 1: Check Workspace Resources
 
-❌ User asks about "time series forecasting" → Agent immediately lists methods from memory
-❌ Agent generates ideas without calling any search tools
-❌ Agent skips to idea generation without `search_results.md` existing
+First, check what resources already exist:
 
-### Correct Pattern: DO THIS
-
-✅ User asks about "time series forecasting" → Agent calls `arxiv` tool with query
-✅ Agent calls `github_search` tool to find implementations
-✅ Agent writes search results to file
-✅ Agent reads downloaded papers before generating ideas
-✅ Ideas reference specific papers by arXiv ID
-
----
-
-## Workspace Convention (Project-based)
+```bash
+# Check active project
+cat ~/.openclaw/workspace/projects/.active 2>/dev/null
 
-**IMPORTANT**: Each research topic uses its own project directory. Agent auto-selects or creates projects.
+# Check papers
+ls ~/.openclaw/workspace/projects/*/papers/ 2>/dev/null | head -20
 
-```
-~/.openclaw/workspace/
-└── projects/
-    ├── .active                   # Current project ID (plain text file)
-    ├── nlp-summarization/        # Project A
-    │   ├── project.json          # Project metadata
-    │   ├── task.json             # Research task definition
-    │   ├── search_results.md     # Search results
-    │   ├── prepare_res.md        # Selected repos summary
-    │   ├── papers/               # Downloaded papers
-    │   ├── repos/                # Cloned repositories
-    │   └── ideas/                # Generated ideas
-    ├── image-segmentation/       # Project B
-    │   └── ...
-    └── ...
+# Check survey results
+cat ~/.openclaw/workspace/projects/*/survey/clusters.json 2>/dev/null | head -5
 ```
 
-**All paths are project-relative**: `~/.openclaw/workspace/projects/{project_id}/`
+### Assess Available Resources
 
-**File existence = step completion.** Skip steps whose output already exists.
+| Resource | Location | Status |
+|----------|----------|--------|
+| Papers | `$WORKSPACE/papers/` | Count: ? |
+| Survey clusters | `$WORKSPACE/survey/clusters.json` | Exists: Y/N |
+| Repos | `$WORKSPACE/repos/` | Count: ? |
 
 ---
 
-## Step 0: Auto Project Management (REQUIRED)
+## Step 2: Ask User About Search Strategy
 
-**Autonomous - DO NOT ask user for confirmation.**
+Based on workspace state, ask user:
 
-1. Extract topic from user query → convert to kebab-case ID
-2. Check `~/.openclaw/workspace/projects/` for existing match
-3. Use existing or create new: `mkdir -p $PROJECT_ID/{papers,repos,ideas}`
-4. Update `.active` file and set `$WORKSPACE` path
+**If papers exist (≥5):**
+> 📚 Found {N} papers in workspace from previous survey.
+>
+> Options:
+> 1. **Use existing papers** - Generate ideas from current collection
+> 2. **Search more** - Run `/literature-survey` to expand collection
+> 3. **Quick search** - Add 5-10 more papers on specific topic
 
-> 📁 Using project: `{project_id}` (new/existing)
+**If no papers:**
+> 📭 No papers found in workspace.
+>
+> To generate grounded ideas, I need literature. Options:
+> 1. **Run /literature-survey** - Comprehensive search (100+ papers, recommended)
+> 2. **Quick search** - Fetch 10-15 papers on your topic now
+> 3. **You provide papers** - Point me to existing PDFs/tex files
 
 ---
 
-## Step 1: Parse Task
+## Step 3: Acquire Resources (if needed)
 
-Check `$WORKSPACE/task.json`. If missing, extract from user query:
+### Option A: Delegate to /literature-survey (Recommended)
 
-- **domain**: Research domain (e.g., "graph neural networks", "recommendation")
-- **focus** (optional): Specific problem or technique
-- **date_limit** (optional): Only consider papers before this date
-
-```bash
-cat $WORKSPACE/task.json 2>/dev/null || echo "No task.json"
+If user wants comprehensive search:
 ```
+Please run: /literature-survey {topic}
 
-Create task.json:
-```json
-{
-  "domain": "graph neural networks",
-  "focus": "scalable transformers for node classification",
-  "date_limit": "2024-01-01",
-  "created": "2024-XX-XX"
-}
-```
-
-**Output:** `$WORKSPACE/task.json`
-
----
-
-## Step 2: Search Papers and Code (MANDATORY)
-
-**⚠️ BLOCKING: You MUST complete this step before ANY idea generation.**
-
-### 2.1 ArXiv Search (REQUIRED)
-
-**You MUST call the `arxiv` tool.** Example:
+This will:
+- Search 100+ papers systematically
+- Filter by relevance (score ≥4)
+- Cluster into research directions
+- Save to $WORKSPACE/papers/
 
-```
-Tool: arxiv
-Arguments:
-  query: "text summarization transformer model"
-  max_results: 10
-  sort_by: "relevance"
-```
-
-If `arxiv` tool is not available, use `WebSearch` with `site:arxiv.org`:
-```
-Tool: WebSearch
-Arguments:
-  query: "site:arxiv.org text summarization transformer model"
+After survey completes, run /idea-generation again.
 ```
 
-### 2.2 GitHub Search (REQUIRED)
+### Option B: Quick Search (5-10 papers)
 
-**You MUST call the `github_search` tool or search GitHub.** Example:
+For fast iteration, do minimal search:
 
+1. **ArXiv search:**
 ```
-Tool: github_search
+Tool: arxiv_search
 Arguments:
-  query: "text summarization pytorch huggingface"
-  sort: "stars"
-  max_results: 20
-```
-
-If `github_search` tool is not available, use `WebSearch`:
-```
-Tool: WebSearch
-Arguments:
-  query: "site:github.com text summarization pytorch stars:>100"
-```
-
-### 2.3 CHECKPOINT: Verify Search Completed
-
-Before proceeding, confirm:
-- [ ] Called arxiv/WebSearch for papers
-- [ ] Called github_search/WebSearch for repositories
-- [ ] Have at least 5 paper results
-- [ ] Have at least 5 repository results
-
-**If search returns 0 results, try different queries. DO NOT proceed without results.**
-
-### 2.4 Compile Results
-
-Write to `$WORKSPACE/search_results.md`:
-
-```markdown
-# Search Results
-
-## Task
-- Domain: {domain}
-- Focus: {focus}
-- Date: {date}
-
-## ArXiv Papers Found
-
-| # | Title | ArXiv ID | Year | Relevance |
-|---|-------|----------|------|-----------|
-| 1 | [Title](pdf_url) | 2401.xxxxx | 2024 | [Why relevant] |
-| 2 | ... | ... | ... | ... |
-
-## GitHub Repositories Found
-
-| # | Repository | Stars | Language | Relevance |
-|---|------------|-------|----------|-----------|
-| 1 | [owner/repo](url) | 1.2k | Python | [Why relevant] |
-| 2 | ... | ... | ... | ... |
+  query: "{user_topic}"
+  max_results: 10
 ```
 
-**Output:** `$WORKSPACE/search_results.md`
-
----
-
-## Step 3: Prepare - Select Repositories
-
-Read search results and select **at least 5** most valuable repositories.
-
-Selection criteria:
-- Direct implementation of relevant papers
-- High code quality (stars, documentation)
-- Active maintenance
-- Covers key techniques in the domain
-
-### 3.1 Clone Selected Repos
-
+2. **Clone 3-5 reference repos:**
 ```bash
 mkdir -p $WORKSPACE/repos
-cd $WORKSPACE/repos
-
-# For each selected repo:
-git clone --depth 1 https://github.com/owner/repo1.git
-git clone --depth 1 https://github.com/owner/repo2.git
-# ... at least 5 repos
-```
-
-### 3.2 Document Selection
-
-Write to `$WORKSPACE/prepare_res.md`:
-
-```markdown
-# Selected Reference Codebases
-
-## Selection Rationale
-[Why these repos were chosen]
-
-## Repositories
-
-### 1. repo1
-- **URL**: https://github.com/owner/repo1
-- **Paper**: [Associated paper if any]
-- **Key Components**:
-  - `model/` - Model architecture
-  - `train.py` - Training loop
-- **Usage**: [How this will help implement our idea]
-
-### 2. repo2
-...
-
-## Reference Papers
-Based on these repos, the key papers to read are:
-1. [Paper Title 1] - ArXiv: 2401.xxxxx
-2. [Paper Title 2] - ArXiv: 2401.xxxxx
-...
+git clone --depth 1 {repo_url} $WORKSPACE/repos/{name}
 ```
 
-**Output:** `$WORKSPACE/prepare_res.md` + `$WORKSPACE/repos/`
-
----
-
-## Step 4: Download Papers
-
-For each paper referenced in prepare_res.md, download the source.
-
-**IMPORTANT: Download .tex source, NOT PDF.** .tex files are much easier for AI to read and extract information from.
-
-### 4.1 Download .tex Source (RECOMMENDED - Use arxiv tool)
-
-Use the `arxiv` tool with `download: true` to automatically download and extract .tex sources:
-
-```
-Tool: arxiv
-Arguments:
-  query: "abstractive summarization long document"
-  max_results: 10
-  download: true
-  output_dir: "$WORKSPACE/papers"
-```
-
-The tool will:
-1. Search for papers matching your query
-2. Download .tex source from `https://arxiv.org/src/{arxiv_id}`
-3. Extract tar.gz archives automatically
-4. Fall back to PDF if .tex is unavailable
-5. Return a `downloads` array showing what was downloaded
-
-**Output format:**
-```json
-{
-  "papers": [...],
-  "downloads": [
-    {"arxiv_id": "2404.04429", "format": "tex", "files": ["main.tex", "methods.tex"]},
-    {"arxiv_id": "2308.03664", "format": "pdf", "files": ["2308.03664.pdf"], "error": "tex unavailable"}
-  ],
-  "output_dir": "$WORKSPACE/papers"
-}
-```
-
-### 4.2 Manual Download (Fallback)
-
-If the arxiv tool is unavailable, use bash:
+3. **Download paper sources:**
 ```bash
 mkdir -p $WORKSPACE/papers/{arxiv_id}
-cd $WORKSPACE/papers/{arxiv_id}
-curl -L "https://arxiv.org/src/{arxiv_id}" -o source.tar.gz
-tar -xzf source.tar.gz 2>/dev/null || mv source.tar.gz main.tex
+curl -L "https://arxiv.org/src/{arxiv_id}" | tar -xz -C $WORKSPACE/papers/{arxiv_id}
 ```
 
-### 4.3 Document Downloads
-
-Write to `$WORKSPACE/papers/download_log.md`:
-
-```markdown
-# Downloaded Papers
-
-| ArXiv ID | Title | Format | Status |
-|----------|-------|--------|--------|
-| 2404.04429 | Physics-Informed ML for Battery... | .tex | ✓ |
-| 2308.03664 | Two-stage Early Prediction... | .tex | ✓ |
-| 2401.99999 | Some Other Paper | .pdf | ✓ (tex unavailable) |
-```
-
-**Output:** `$WORKSPACE/papers/`
-
 ---
 
-## Step 5: Generate Ideas (5 Ideas)
+## Step 4: Analyze Literature
 
-**⚠️ BLOCKING: DO NOT start this step unless Steps 2-4 are complete.**
+**Prerequisites:** At least 5 papers in `$WORKSPACE/papers/`
 
-### Pre-requisite Checkpoint
+### 4.1 Read Papers
 
-Before generating ANY ideas, verify these files exist:
-- [ ] `$WORKSPACE/search_results.md` - search results from Step 2
-- [ ] `$WORKSPACE/prepare_res.md` - selected repos from Step 3
-- [ ] At least 3 papers downloaded in `$WORKSPACE/papers/`
+For each paper, extract:
+- Core contribution (1 sentence)
+- Key method/formula
+- Limitations mentioned
+- Future work suggestions
 
-**If any file is missing, GO BACK and complete the previous steps.**
+**Long papers (>50KB):** See `references/reading-long-papers.md`
 
-This is the core intellectual step. Generate **exactly 5 distinct innovative ideas**.
-
-**IMPORTANT: Ideas must be grounded in the literature you just read. Each idea MUST:**
-- Reference at least 2 specific papers by arXiv ID
-- Identify specific limitations from those papers
-- Propose improvements based on gaps found in the literature
-
-### 5.1 Analyze Literature First (REQUIRED)
-
-For each paper in `papers/`:
-1. Read thoroughly (especially: abstract, method, experiments, limitations)
-2. Extract: core contribution, math formulas, limitations, future work
-3. Note connections to other papers
-
-**Long Papers (>50KB):** See `references/reading-long-papers.md` for chunked reading strategy.
-
-For each repo in `repos/`:
-1. Understand structure: `gen_code_tree_structure` equivalent
-2. Identify key implementations
-3. Note reusable components
-
-### 5.2 Identify Research Gaps
+### 4.2 Identify Research Gaps
 
 Look for:
 - Common limitations across papers
-- Unexplored combinations of techniques
+- Unexplored technique combinations
 - Scalability issues
 - Assumptions that could be relaxed
 
-### 5.3 Generate 5 Ideas
+Document gaps in `$WORKSPACE/ideas/gaps.md`:
+```markdown
+# Research Gaps Identified
+
+## Gap 1: [Description]
+- Mentioned in: [paper1], [paper2]
+- Why important: ...
+
+## Gap 2: [Description]
+...
+```
+
+---
+
+## Step 5: Generate 5 Ideas
 
 Create `$WORKSPACE/ideas/idea_1.md` through `idea_5.md` using template in `references/idea-template.md`.
 
-**Key requirements:**
-- Each idea must cite ≥2 papers by arXiv ID
-- Use different strategies (see template): combination, simplification, generalization, constraint relaxation, architecture innovation
+**Requirements:**
+- Each idea cites ≥2 papers by arXiv ID
+- Use different strategies:
 
-**❌ REJECTED if:** No arXiv IDs cited, or ideas not grounded in literature
+| Idea | Strategy |
+|------|----------|
+| 1 | Combination - merge 2+ techniques |
+| 2 | Simplification - reduce complexity |
+| 3 | Generalization - extend to new domain |
+| 4 | Constraint relaxation - remove assumption |
+| 5 | Architecture innovation - new design |
 
-**Output:** `$WORKSPACE/ideas/idea_1.md` through `idea_5.md`
+**❌ REJECTED if:** No arXiv IDs cited, or ideas not grounded in literature
 
 ---
 
 ## Step 6: Select and Enhance Best Idea
 
-### 6.1 Evaluate All Ideas
+### 6.1 Score All Ideas
 
-Score each idea on Novelty/Feasibility/Impact (1-5). Select highest total.
+| Idea | Novelty | Feasibility | Impact | Total |
+|------|---------|-------------|--------|-------|
+| 1 | /5 | /5 | /5 | /15 |
+| ... | | | | |
 
 ### 6.2 Enhance Selected Idea
 
@@ -404,62 +183,25 @@ Create `$WORKSPACE/ideas/selected_idea.md` with:
 - Architecture choices
 - Hyperparameters
 - Implementation roadmap
-- Failure modes & mitigations
-
-**Output:** `$WORKSPACE/ideas/selected_idea.md`
 
 ---
 
-## Step 7: Code Survey - Map Idea to Implementations
+## Step 7: Code Survey
 
-Map each **atomic concept** in the selected idea to code in reference repos.
+Map idea concepts to reference implementations.
 
-See `references/code-mapping.md` for detailed template.
-
-**Quick steps:**
-1. Extract atomic concepts from `selected_idea.md`
-2. Search repos: `grep -r "class.*Attention" $WORKSPACE/repos/`
-3. Document mapping to `$WORKSPACE/ideas/implementation_report.md`
+See `references/code-mapping.md` for template.
 
 **Output:** `$WORKSPACE/ideas/implementation_report.md`
 
 ---
 
-## Step 8: Final Summary
+## Step 8: Summary
 
-Create `$WORKSPACE/ideas/summary.md` with:
-- Task overview (domain, focus)
-- Resources gathered (papers, repos count)
+Create `$WORKSPACE/ideas/summary.md`:
 - All 5 ideas with scores
 - Selected idea details
-- Next steps: `/research-pipeline` or manual implementation
-
-**Output:** `$WORKSPACE/ideas/summary.md`
-
----
-
-## Quality Checklist
-
-Before completing, verify:
-
-- [ ] At least 5 repos cloned in `repos/`
-- [ ] At least 3 papers downloaded in `papers/`
-- [ ] All 5 ideas are substantially different
-- [ ] Selected idea has complete math formulations
-- [ ] Implementation report covers ALL atomic concepts
-- [ ] Each concept has actual code reference (not placeholder)
-- [ ] Evaluation plan has specific datasets and metrics
-
----
-
-## Integration with Other Skills
-
-**After idea-generation:**
-- `/research-pipeline` → Implement the selected idea
-
-**To gather more resources:**
-- `/literature-survey` → Comprehensive paper collection
-- `/write-review-paper` → Synthesize into review
+- Next steps: `/research-pipeline` to implement
 
 ---
 
@@ -467,19 +209,15 @@ Before completing, verify:
 
 | User Says | Action |
 |-----------|--------|
-| "Generate research ideas for NLP" | Full workflow (Steps 1-8) |
-| "Search papers on X" | Steps 1-2 only |
-| "I have papers, generate ideas" | Skip to Step 5 |
-| "Enhance this idea: ..." | Skip to Step 6-7 |
-| "Map this idea to code" | Step 7 only |
+| "Generate ideas for X" | Check workspace → ask strategy → generate |
+| "I have papers, generate ideas" | Skip to Step 4 |
+| "Enhance idea N" | Jump to Step 6 |
+| "Map to code" | Jump to Step 7 |
 
 ---
 
-## Batch Processing Rule
-
-If more than 10 papers/repos to analyze:
-1. First pass: Quick scan all (abstract/README only)
-2. Select top 5-7 for deep analysis
-3. Generate ideas from deep analysis
+## Integration
 
-Do NOT process all resources with full detail - context will overflow.
+- **Before:** `/literature-survey` to collect papers
+- **After:** `/research-pipeline` to implement selected idea
+- **Alternative:** `/write-review-paper` to write survey instead

package/skills/literature-survey/SKILL.md

@@ -14,6 +14,8 @@ metadata:
 
 Comprehensive literature discovery workflow for a research domain. This skill searches broadly, filters by relevance, clusters by direction, and iterates to ensure complete coverage.
 
+**Workspace:** See `../_shared/workspace-spec.md` for directory structure. Outputs go to `$WORKSPACE/survey/` and `$WORKSPACE/papers/`.
+
 ## Architecture: Isolated Sub-agent
 
 This survey runs in an **isolated sub-session** to avoid context pollution. The main session only receives the final report.

package/skills/research-pipeline/SKILL.md

@@ -13,44 +13,22 @@ metadata:
 
 # Research Pipeline
 
-Automate an end-to-end ML research workflow: idea -> literature search -> survey -> plan -> implement -> review -> iterate.
+Automate an end-to-end ML research workflow: idea → literature → survey → plan → implement → review → iterate.
 
-All intermediate results live in a project-based workspace directory. **File existence = step completion.** If a step's output file already exists, skip that step and move on. This enables crash recovery and incremental progress.
+**Workspace:** See `../_shared/workspace-spec.md` for directory structure. Outputs go to `$WORKSPACE/project/`, `$WORKSPACE/iterations/`.
 
----
+**File existence = step completion.** Skip steps whose output already exists.
 
-## Workspace Convention (Project-based)
+---
 
-**IMPORTANT**: This skill uses the same project-based workspace as `idea-generation`. Check or set the active project first.
+## Step 0: Check Active Project
 
-### Check Active Project
 ```bash
 cat ~/.openclaw/workspace/projects/.active 2>/dev/null
 ```
 
-If a project is active, set `$WORKSPACE = ~/.openclaw/workspace/projects/{project_id}/`.
-
-If no active project exists, create one based on the research idea (see Step 1).
-
-### Directory Structure
-```
-$WORKSPACE/
-├── project.json          # Project metadata
-├── task.json             # Research task/idea definition
-├── search_results.md     # Search results (Step 2)
-├── prepare_res.md        # Selected repos (Step 3)
-├── papers/               # Downloaded papers (Step 4)
-├── repos/                # Cloned repositories (Step 3)
-├── notes/                # Paper notes (Step 5)
-├── survey_res.md         # Literature survey (Step 5)
-├── plan_res.md           # Implementation plan (Step 6)
-├── project/              # Code implementation (Step 7)
-├── ml_res.md             # Implementation report (Step 7)
-├── iterations/           # Review iterations (Step 8-9)
-│   ├── judge_v1.md
-│   └── ...
-└── experiment_res.md     # Final results (Step 10)
-```
+If active, set `$WORKSPACE = ~/.openclaw/workspace/projects/{project_id}/`.
+If none, create based on research idea in Step 1.
 
 ---
 

package/skills/research-pipeline/references/workspace-spec.md

@@ -1,81 +1,5 @@
-# Workspace Directory Specification
+# Workspace Specification
 
-All research pipeline artifacts live in a `workspace/` directory. The location is either specified by the user or defaults to the current working directory plus `workspace/`.
+**This file has moved to the shared location.**
 
-## Directory Layout
-
-```
-workspace/
-  task.json                 # Input: research task definition
-  search_results.md         # Step 2: arxiv + github search results
-  prepare_res.md            # Step 3: selected repos and rationale
-  survey_res.md             # Step 5: synthesized literature survey
-  plan_res.md               # Step 6: four-part implementation plan
-  ml_res.md                 # Step 7: implementation report
-  experiment_res.md         # Step 10: full training results
-
-  repos/                    # Step 3: cloned reference repositories
-    repo-name-1/
-    repo-name-2/
-
-  papers/                   # Step 4: downloaded paper sources
-    2401.12345.tex
-    2401.67890.tex
-
-  notes/                    # Step 5: per-paper survey notes
-    paper_001.md
-    paper_002.md
-
-  iterations/               # Steps 8-9: review history
-    judge_v1.md
-    judge_v2.md
-
-  project/                  # Step 7: implementation code
-    model/
-    data/
-    training/
-    testing/
-    utils/
-    run.py
-    requirements.txt
-```
-
-## Conventions
-
-### File Existence = Step Completion
-
-The research pipeline uses file existence as the checkpoint mechanism. Before executing any step, check whether its output file already exists. If it does, skip the step.
-
-This enables:
-- **Crash recovery**: resume from the last completed step.
-- **Incremental progress**: re-running the pipeline skips completed work.
-- **Transparency**: a human can inspect progress by listing the directory.
-
-### Naming Rules
-
-- Markdown files (`.md`) for human-readable outputs.
-- JSON files (`.json`) for structured data (task definition).
-- Paper notes use sequential numbering: `paper_001.md`, `paper_002.md`.
-- Review iterations use version numbering: `judge_v1.md`, `judge_v2.md`.
-
-### Immutability
-
-Once a step's output is written, do NOT modify it unless the user explicitly asks. If a step needs to be re-done, delete the output file first, then re-execute.
-
-Exception: `workspace/project/` is mutable during the implement-review-iterate loop (Steps 7-9).
-
-### task.json Schema
-
-```json
-{
-  "idea": "A 1-3 sentence description of the research idea",
-  "references": ["2401.12345", "paper title string"],
-  "domain": "recommendation systems",
-  "date_limit": "2024-01-01"
-}
-```
-
-- `idea` (required): The core research idea to implement.
-- `references` (optional): ArXiv IDs or paper titles as starting points.
-- `domain` (optional): Research domain for focused searching.
-- `date_limit` (optional): Only consider papers published after this date.
+See: `../../_shared/workspace-spec.md` for the unified workspace specification used by all Scientify skills.

package/skills/write-review-paper/SKILL.md

@@ -14,6 +14,8 @@ metadata:
 
 Guide for writing a structured literature review or survey paper from papers you've already collected. This skill helps with reading strategy, note organization, and academic writing.
 
+**Workspace:** See `../_shared/workspace-spec.md` for directory structure. Outputs go to `$WORKSPACE/review/`.
+
 ## Prerequisites
 
 Before starting, ensure you have: