@wentorai/research-plugins 1.2.2 → 1.3.0

package/skills/research/automation/ai-scientist-guide/SKILL.md

@@ -1,228 +0,0 @@
----
-name: ai-scientist-guide
-description: "End-to-end automated scientific discovery with AI Scientist v2"
-metadata:
-  openclaw:
-    emoji: "🤖"
-    category: "research"
-    subcategory: "automation"
-    keywords: ["ai-scientist", "research automation", "scientific workflow", "AI experiment design"]
-    source: "https://github.com/SakanaAI/AI-Scientist-v2"
----
-
-# AI Scientist Guide
-
-## Overview
-
-The AI Scientist v2 is a fully autonomous scientific research system developed by Sakana AI that can generate hypotheses, run experiments, analyze data, and write complete scientific manuscripts. It represents the cutting edge of AI-driven research automation, having produced the first workshop paper written entirely by AI and accepted through peer review at ICLR 2025.
-
-Unlike its predecessor (v1), the AI Scientist v2 removes reliance on human-authored templates, generalizes across Machine Learning domains, and employs a progressive agentic tree search guided by an experiment manager agent. This guide explains how to set up, configure, and use the system effectively, as well as how to integrate its principles into your own research workflows.
-
-This skill is relevant for researchers interested in accelerating their experimental cycles, exploring automated hypothesis generation, or understanding how agentic AI systems approach scientific discovery. Even if you do not use AI Scientist v2 directly, the concepts behind its design -- structured ideation, tree-based experiment exploration, automated writing -- can inform how you organize your own research.
-
-## System Architecture
-
-The AI Scientist v2 operates through a multi-stage pipeline:
-
-```
-Topic Description (.md)
-    |
-    v
-[Ideation Stage] --> Research Ideas (.json)
-    |
-    v
-[Experiment Stage] --> Best-First Tree Search (BFTS)
-    |                   - Multiple parallel workers
-    |                   - Automatic debugging
-    |                   - Experiment manager agent
-    v
-[Analysis Stage] --> Results + Figures
-    |
-    v
-[Writing Stage] --> Complete Paper (.pdf)
-    |
-    v
-[Review Stage] --> Automated Peer Review
-```
-
-### Key Components
-
-| Component | Role | Model Used |
-|-----------|------|------------|
-| Ideation Agent | Generates research hypotheses | Configurable (GPT-4o, Claude) |
-| Experiment Manager | Guides tree search exploration | Claude 3.5 Sonnet (default) |
-| Analysis Agent | Interprets results, creates figures | Same as experiment |
-| Writing Agent | Drafts full paper with LaTeX | o1-preview (default) |
-| Citation Agent | Finds and integrates references | GPT-4o (default) |
-| Review Agent | Simulates peer review | GPT-4o (default) |
-
-## Installation and Setup
-
-### Prerequisites
-
-- Linux with NVIDIA GPU (CUDA support required)
-- Python 3.11+
-- conda or mamba package manager
-
-### Step-by-Step Installation
-
-```bash
-# 1. Create and activate environment
-conda create -n ai_scientist python=3.11
-conda activate ai_scientist
-
-# 2. Install PyTorch with CUDA
-conda install pytorch torchvision torchaudio pytorch-cuda=12.4 \
-  -c pytorch -c nvidia
-
-# 3. Install PDF and LaTeX tools
-conda install anaconda::poppler conda-forge::chktex
-
-# 4. Clone and install
-git clone https://github.com/SakanaAI/AI-Scientist-v2.git
-cd AI-Scientist-v2
-pip install -r requirements.txt
-
-# 5. Set API keys
-export OPENAI_API_KEY=<key>
-export S2_API_KEY=<key>  # Optional but recommended
-```
-
-## Running the Pipeline
-
-### Stage 1: Ideation
-
-Create a topic description file following this structure:
-
-```markdown
-# Title
-Exploring Efficient Fine-Tuning Methods for Large Language Models
-
-# Keywords
-LoRA, parameter-efficient fine-tuning, LLM adaptation, low-rank
-
-# TL;DR
-Investigate novel parameter-efficient methods for adapting LLMs to
-domain-specific tasks with minimal compute.
-
-# Abstract
-Large language models require substantial resources for full fine-tuning.
-Parameter-efficient methods like LoRA reduce this cost but may sacrifice
-performance. We seek to explore new approaches that balance efficiency
-and effectiveness across diverse downstream tasks.
-```
-
-Run ideation:
-
-```bash
-python ai_scientist/perform_ideation_temp_free.py \
-  --workshop-file "ai_scientist/ideas/my_topic.md" \
-  --model gpt-4o-2024-05-13 \
-  --max-num-generations 20 \
-  --num-reflections 5
-```
-
-This produces a JSON file with structured research ideas including hypotheses, proposed experiments, and related work.
-
-### Stage 2: Experiment and Paper Generation
-
-```bash
-python launch_scientist_bfts.py \
-  --load_ideas "ai_scientist/ideas/my_topic.json" \
-  --load_code \
-  --add_dataset_ref \
-  --model_writeup o1-preview-2024-09-12 \
-  --model_citation gpt-4o-2024-11-20 \
-  --model_review gpt-4o-2024-11-20 \
-  --model_agg_plots o3-mini-2025-01-31 \
-  --num_cite_rounds 20
-```
-
-### Configuration: bfts_config.yaml
-
-Key parameters to tune:
-
-```yaml
-agent:
-  num_workers: 3        # Parallel exploration paths
-  steps: 21             # Maximum nodes to explore
-  num_seeds: 3          # Initial root nodes
-
-search:
-  max_debug_depth: 3    # Max debug attempts per failing node
-  debug_prob: 0.5       # Probability of debugging vs. abandoning
-  num_drafts: 3         # Number of independent search trees
-```
-
-## Cost and Performance Estimates
-
-| Phase | Typical Cost | Duration |
-|-------|-------------|----------|
-| Ideation (20 ideas) | $2-5 | 15-30 min |
-| Experimentation (BFTS) | $15-20 | 2-6 hours |
-| Writing + Citation | $5 | 20-30 min |
-| Review | $1-2 | 5-10 min |
-| **Total per run** | **$23-32** | **3-7 hours** |
-
-## Integrating AI Scientist Principles Into Your Research
-
-Even without running the full system, you can adopt its methodological ideas:
-
-### Structured Ideation
-
-Use LLMs to brainstorm research directions systematically:
-
-```python
-prompt = """
-Given the research area of [TOPIC], generate 5 research ideas.
-For each idea, provide:
-1. Hypothesis (one sentence)
-2. Key experiment to test it
-3. Expected outcome if hypothesis is true
-4. Expected outcome if hypothesis is false
-5. Why this matters (impact)
-"""
-```
-
-### Tree-Based Experiment Design
-
-Instead of running experiments linearly, structure them as a tree:
-
-1. Start with 2-3 seed experiments (broad exploration)
-2. Evaluate results at each node
-3. Expand the most promising branches
-4. Prune branches that show diminishing returns
-5. Debug failures before abandoning (up to a depth limit)
-
-### Automated Literature Checks
-
-Use Semantic Scholar API to check novelty before investing in an idea:
-
-```python
-import requests
-
-def check_novelty(query, max_results=10):
-    url = "https://api.semanticscholar.org/graph/v1/paper/search"
-    params = {"query": query, "limit": max_results,
-              "fields": "title,year,citationCount"}
-    resp = requests.get(url, params=params)
-    papers = resp.json().get('data', [])
-    return papers
-```
-
-## Best Practices
-
-- **Always disclose AI involvement.** If AI Scientist generates any part of your paper, disclose this clearly in the methods section.
-- **Validate all generated results.** Automated systems can produce plausible but incorrect code. Review experiments manually.
-- **Use sandboxed environments.** The system executes LLM-generated code. Run it in Docker containers.
-- **Start with well-defined topics.** Narrow, concrete research questions produce better results than broad ones.
-- **Iterate on the topic description.** The quality of the input topic file strongly influences output quality.
-- **Combine with human judgment.** Use AI Scientist for ideation and draft generation, but apply human expertise for final decisions.
-
-## References
-
-- [AI-Scientist-v2 Repository](https://github.com/SakanaAI/AI-Scientist-v2) -- Source code (2,229+ stars)
-- [AI Scientist v2 Paper](https://pub.sakana.ai/ai-scientist-v2/paper) -- Workshop-Level Automated Scientific Discovery via Agentic Tree Search
-- [AI Scientist Blog Post](https://sakana.ai/ai-scientist-first-publication/) -- Sakana AI announcement
-- [AIDE: ML Engineering Agent](https://github.com/WecoAI/aideml) -- Foundation for the tree search component
-- [Semantic Scholar API](https://api.semanticscholar.org/) -- Literature search API

package/skills/research/automation/coexist-ai-guide/SKILL.md

@@ -1,149 +0,0 @@
----
-name: coexist-ai-guide
-description: "Modular MCP-based research assistant framework"
-metadata:
-  openclaw:
-    emoji: "🧩"
-    category: "research"
-    subcategory: "automation"
-    keywords: ["CoexistAI", "MCP", "research assistant", "modular agents", "academic workflow", "framework"]
-    source: "https://github.com/SPThole/CoexistAI"
----
-
-# CoexistAI Research Assistant Guide
-
-## Overview
-
-CoexistAI is a modular research assistant framework built on MCP (Model Context Protocol) that provides composable academic research capabilities. It offers MCP servers for paper search, citation management, data analysis, and writing assistance that can be mixed and matched with any MCP-compatible LLM client. Designed for researchers who want customizable, extensible AI research tooling.
-
-## Architecture
-
-```
-LLM Client (Claude Code / Open WebUI / etc.)
-      ↓ MCP Protocol
-  CoexistAI MCP Servers
-  ├── paper-search     (arXiv, S2, PubMed)
-  ├── citation-manager (BibTeX, formatting)
-  ├── data-analysis    (Python execution)
-  ├── writing-assist   (drafting, revision)
-  ├── note-taker       (structured notes)
-  └── custom modules   (extend as needed)
-```
-
-## Installation
-
-```bash
-# Install CoexistAI
-npm install -g @coexistai/mcp-servers
-
-# Or individual modules
-npm install -g @coexistai/paper-search
-npm install -g @coexistai/citation-manager
-npm install -g @coexistai/data-analysis
-```
-
-## MCP Configuration
-
-```json
-{
-  "mcpServers": {
-    "coexist-paper-search": {
-      "command": "npx",
-      "args": ["@coexistai/paper-search"],
-      "env": {}
-    },
-    "coexist-citation": {
-      "command": "npx",
-      "args": ["@coexistai/citation-manager"],
-      "env": {
-        "BIB_FILE": "./references.bib"
-      }
-    },
-    "coexist-analysis": {
-      "command": "npx",
-      "args": ["@coexistai/data-analysis"],
-      "env": {
-        "SANDBOX": "true"
-      }
-    }
-  }
-}
-```
-
-## Module: Paper Search
-
-```markdown
-### Available Tools
-- `search_papers(query, source, limit)` — Search academic DBs
-- `get_paper_details(id)` — Fetch full metadata
-- `find_related(id, limit)` — Find related papers
-- `get_citations(id)` — List citing papers
-- `get_references(id)` — List referenced papers
-
-### Example Usage
-"Search for papers on contrastive learning in NLP from 2023"
-→ Returns top results with title, authors, abstract, DOI
-```
-
-## Module: Citation Manager
-
-```markdown
-### Available Tools
-- `add_citation(doi)` — Add to .bib file from DOI
-- `format_citation(id, style)` — Format in APA/MLA/etc
-- `search_bib(query)` — Search existing bibliography
-- `check_duplicates()` — Find duplicate entries
-- `export_bib(format)` — Export in various formats
-
-### Example Usage
-"Add the BERT paper to my bibliography"
-→ Fetches metadata, generates BibTeX, appends to refs.bib
-```
-
-## Module: Data Analysis
-
-```markdown
-### Available Tools
-- `run_python(code)` — Execute Python in sandbox
-- `analyze_csv(path, task)` — Auto-analyze data file
-- `generate_plot(data, type)` — Create visualizations
-- `run_statistics(data, test)` — Statistical tests
-
-### Example Usage
-"Run a t-test comparing groups A and B in experiment.csv"
-→ Loads data, runs scipy.stats.ttest_ind, returns results
-```
-
-## Custom Module Development
-
-```typescript
-// Create a custom MCP module
-import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
-
-const server = new McpServer({
-  name: "my-research-tool",
-  version: "1.0.0",
-});
-
-server.tool(
-  "custom_analysis",
-  { description: "Run custom research analysis" },
-  async (args) => {
-    // Your custom logic
-    return { result: "Analysis complete" };
-  }
-);
-```
-
-## Use Cases
-
-1. **Modular research**: Pick-and-choose research capabilities
-2. **Custom pipelines**: Build tailored research workflows
-3. **Tool integration**: Connect to existing research infrastructure
-4. **Team standardization**: Shared MCP modules across lab
-5. **Extension**: Add domain-specific research tools
-
-## References
-
-- [CoexistAI GitHub](https://github.com/SPThole/CoexistAI)
-- [MCP Specification](https://modelcontextprotocol.io/)

package/skills/research/automation/foam-agent-guide/SKILL.md

@@ -1,203 +0,0 @@
----
-name: foam-agent-guide
-description: "Agent for automated knowledge graph building from research notes"
-metadata:
-  openclaw:
-    emoji: "🫧"
-    category: "research"
-    subcategory: "automation"
-    keywords: ["Foam", "knowledge graph", "Zettelkasten", "research notes", "linked notes", "PKM agent"]
-    source: "https://github.com/foambubble/foam"
----
-
-# Foam Agent Guide
-
-## Overview
-
-Foam is a personal knowledge management and note-sharing system built on VS Code, inspired by Roam Research. This guide covers using AI agents to automate Foam-based research workflows — extracting concepts from papers, building knowledge graphs, discovering connections between notes, and maintaining a Zettelkasten-style research knowledge base. Particularly useful for researchers managing large literature collections.
-
-## Foam Basics
-
-### Workspace Structure
-
-```
-research-foam/
-├── .foam/
-│   └── config.json          # Foam configuration
-├── .vscode/
-│   └── settings.json        # VS Code settings
-├── inbox/                   # Unprocessed notes
-├── literature/              # Paper notes (one per paper)
-├── concepts/                # Concept/topic notes
-├── projects/                # Project-specific notes
-├── daily/                   # Daily research journal
-└── README.md                # Graph entry point
-```
-
-### Note Format
-
-```markdown
-# Attention Mechanism
-
-Tags: #transformer #nlp #deep-learning
-
-## Summary
-The attention mechanism allows models to focus on relevant parts
-of the input sequence when producing output.
-
-## Key Concepts
-- [[self-attention]] computes relationships within a sequence
-- [[cross-attention]] relates two different sequences
-- [[multi-head-attention]] runs parallel attention functions
-
-## References
-- [[vaswani-2017-attention]] — original Transformer paper
-- [[bahdanau-2015-attention]] — attention for NMT
-
-## Related
-- [[transformer-architecture]]
-- [[positional-encoding]]
-```
-
-## Automated Note Generation
-
-```python
-from foam_agent import FoamAgent
-
-agent = FoamAgent(
-    workspace="./research-foam",
-    llm_provider="anthropic",
-)
-
-# Generate literature note from PDF
-note = agent.process_paper(
-    pdf_path="papers/vaswani2017attention.pdf",
-    template="literature",
-)
-# Creates: literature/vaswani-2017-attention.md
-# - Extracts title, authors, year, venue
-# - Summarizes key contributions
-# - Identifies main concepts → [[wikilinks]]
-# - Adds BibTeX reference
-
-# Batch process papers directory
-agent.process_papers_batch(
-    input_dir="papers/",
-    output_dir="literature/",
-    skip_existing=True,
-)
-```
-
-## Knowledge Graph Building
-
-```python
-# Discover connections between notes
-connections = agent.discover_connections(
-    scope="literature/",
-    min_similarity=0.7,
-)
-
-for conn in connections:
-    print(f"{conn.source} ↔ {conn.target}")
-    print(f"  Reason: {conn.reason}")
-    print(f"  Similarity: {conn.score:.2f}")
-
-# Auto-insert wikilinks for discovered connections
-agent.link_notes(connections, mode="suggest")
-# Adds [[suggested-link]] with <!-- foam:auto --> comment
-
-# Build concept map from literature notes
-concept_map = agent.build_concept_map(
-    scope="literature/",
-    min_mentions=3,
-)
-for concept in concept_map.nodes:
-    print(f"{concept.name}: mentioned in {concept.count} papers")
-```
-
-## Daily Research Journal
-
-```python
-# Generate daily note with research context
-agent.create_daily_note(
-    include=[
-        "recent_papers",        # Papers read recently
-        "open_questions",       # Unresolved questions from notes
-        "stale_notes",          # Notes not updated in 30+ days
-        "orphan_notes",         # Notes with no incoming links
-        "suggested_reading",    # Papers related to recent work
-    ],
-)
-
-# Output: daily/2025-03-10.md
-# ## Recently Added
-# - [[vaswani-2017-attention]] (added yesterday)
-#
-# ## Open Questions
-# - How does [[rotary-position-embedding]] compare to [[alibi]]?
-#
-# ## Suggested Reading
-# Based on recent notes on [[flash-attention]], consider:
-# - "FlashAttention-2: Faster Attention" (Dao, 2023)
-```
-
-## Graph Analysis
-
-```python
-# Analyze knowledge graph structure
-stats = agent.analyze_graph()
-
-print(f"Total notes: {stats.total_notes}")
-print(f"Total links: {stats.total_links}")
-print(f"Orphan notes: {stats.orphan_count}")
-print(f"Most connected: {stats.top_nodes[:5]}")
-print(f"Clusters: {stats.cluster_count}")
-
-# Identify research gaps
-gaps = agent.find_gaps()
-for gap in gaps:
-    print(f"Gap: {gap.description}")
-    print(f"  Between clusters: {gap.cluster_a} ↔ {gap.cluster_b}")
-    print(f"  Suggested topics: {gap.suggested_topics}")
-```
-
-## Configuration
-
-```json
-{
-  "foam.agent": {
-    "llm_provider": "anthropic",
-    "templates": {
-      "literature": "templates/literature.md",
-      "concept": "templates/concept.md",
-      "daily": "templates/daily.md"
-    },
-    "auto_link": {
-      "enabled": true,
-      "min_similarity": 0.75,
-      "max_suggestions_per_note": 5
-    },
-    "graph": {
-      "exclude_patterns": ["daily/*", "templates/*"],
-      "tag_colors": {
-        "transformer": "#3B82F6",
-        "nlp": "#10B981"
-      }
-    }
-  }
-}
-```
-
-## Use Cases
-
-1. **Literature management**: Auto-generate linked notes from papers
-2. **Concept mapping**: Build and visualize research concept graphs
-3. **Research journaling**: AI-assisted daily research summaries
-4. **Gap discovery**: Find under-explored connections between topics
-5. **Collaborative wikis**: Shared team research knowledge bases
-
-## References
-
-- [Foam GitHub](https://github.com/foambubble/foam)
-- [Foam Documentation](https://foambubble.github.io/foam/)
-- [Zettelkasten Method](https://zettelkasten.de/introduction/)