TooluniverseSAFE

Kosmos: An AI Scientist for Autonomous Discovery - An implementation and adaptation to be driven by Claude Code or API - Based on the Kosmos AI Paper - https://arxiv.org/abs/2511.02824

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: tooluniverse
description: Use this skill when working with scientific research tools and workflows across bioinformatics, cheminformatics, genomics, structural biology, proteomics, and drug discovery. This skill provides access to 600+ scientific tools including machine learning models, datasets, APIs, and analysis packages. Use when searching for scientific tools, executing computational biology workflows, composing multi-step research pipelines, accessing databases like OpenTargets/PubChem/UniProt/PDB/ChEMBL, performing tool discovery for research tasks, or integrating scientific computational resources into LLM workflows.
---

# ToolUniverse

## Overview

ToolUniverse is a unified ecosystem that enables AI agents to function as research scientists by providing standardized access to 600+ scientific resources. Use this skill to discover, execute, and compose scientific tools across multiple research domains including bioinformatics, cheminformatics, genomics, structural biology, proteomics, and drug discovery.

**Key Capabilities:**
- Access 600+ scientific tools, models, datasets, and APIs
- Discover tools using natural language, semantic search, or keywords
- Execute tools through standardized AI-Tool Interaction Protocol
- Compose multi-step workflows for complex research problems
- Integration with Claude Desktop/Code via Model Context Protocol (MCP)

## When to Use This Skill

Use this skill when:
- Searching for scientific tools by function or domain (e.g., "find protein structure prediction tools")
- Executing computational biology workflows (e.g., disease target identification, drug discovery, genomics analysis)
- Accessing scientific databases (OpenTargets, PubChem, UniProt, PDB, ChEMBL, KEGG, etc.)
- Composing multi-step research pipelines (e.g., target discovery → structure prediction → virtual screening)
- Working with bioinformatics, cheminformatics, or structural biology tasks
- Analyzing gene expression, protein sequences, molecular structures, or clinical data
- Performing literature searches, pathway enrichment, or variant annotation
- Building automated scientific research workflows

## Quick Start

### Basic Setup
```python
from tooluniverse import ToolUniverse

# Initialize and load tools
tu = ToolUniverse()
tu.load_tools()  # Loads 600+ scientific tools

# Discover tools
tools = tu.run({
    "name": "Tool_Finder_Keyword",
    "arguments": {
        "description": "disease target associations",
        "limit": 10
    }
})

# Execute a tool
result = tu.run({
    "name": "OpenTargets_get_associated_targets_by_disease_efoId",
    "arguments": {"efoId": "EFO_0000537"}  # Hypertension
})
```

### Model Context Protocol (MCP)
For Claude Desktop/Code integration:
```bash
tooluniverse-smcp
```

## Core Workflows

### 1. Tool Discovery

Find relevant tools for your research task:

**Three discovery methods:**
- `Tool_Finder` - Embedding-based semantic search (requires GPU)
- `Tool_Finder_LLM` - LLM-based semantic search (no GPU required)
- `Tool_Finder_Keyword` - Fast keyword search

**Example:**
```python
# Search by natural language description
tools = tu.run({
    "name": "Tool_Finder_LLM",
    "arguments": {
        "description": "Find tools for RNA sequencing differential expression analysis",
        "limit": 10
    }
})

# Review available tools
for tool in tools:
    print(f"{tool['name']}: {tool['description']}")
```

**See `references/tool-discovery.md` for:**
- Detailed discovery methods and search strategies
- Domain-specific keyword suggestions
- Best practices for finding tools

### 2. Tool Execution

Execute individual tools through the standardized interface:

**Example:**
```python
# Execute disease-target lookup
targets = tu.run({
    "name": "OpenTargets_get_associated_targets_by_disease_efoId",
    "arguments": {"efoId": "EFO_0000616"}  # Breast cancer
})

# Get protein structure
structure = tu.run({
    "name": "AlphaFold_get_structure",
    "arguments": {"uniprot_id": "P12345"}
})

# Calculate molecular properties
properties = tu.run({
    "name": "RDKit_calculate_descriptors",
    "arguments": {"smiles": "CCO"}  # Ethanol
})
```

**See `references/tool-execution.md` for:**
- Real-world execution examples across domains
- Tool parameter handling and validation
- Result processing and error handling
- Best practices for production use

### 3. Tool Composition and Workflows

Compose multiple tools for complex research workflows:

**Drug Discovery Example:**
```python
# 1. Find disease targets
targets = tu.run({
    "name": "OpenTargets_get_associated_targets_by_disease_efoId",
    "arguments": {"efoId": "EFO_0000616"}
})

# 2. Get protein structures
structures = []
for target in targets[:5]:
    structure = tu.run({
        "name": "AlphaFold_get_structure",
        "arguments": {"uniprot_id": target['uniprot_id']}
    })
    structures.append(structure)

# 3. Screen compounds
hits = []
for structure in structures:
    compounds = tu.run({
        "name": "ZINC_virtual_screening",
        "arguments": {
            "structure": structure,
            "library": "lead-like",
            "top_n": 100
        }
    })
    hits.extend(compounds)

# 4. Evaluate drug-likeness
drug_candidates = []
for compound in hits:
    props = tu.run({
        "name": "RDKit_calculate_drug_properties",
        "arguments": {"smiles": compound['smiles']}
    })
    if props['lipinski_pass']:
        drug_candidates.append(compound)
```

**See `references/tool-composition.md` for:**
- Complete workflow examples (drug discovery, genomics, clinical)
- Sequential and parallel tool composition patterns
- Output processing hooks
- Workflow best practices

## Scientific Domains

ToolUniverse supports 600+ tools across major scientific domains:

**Bioinformatics:**
- Sequence analysis, alignment, BLAST
- Gene expression (RNA-seq, DESeq2)
- Pathway enrichment (KEGG, Reactome, GO)
- Variant annotation (VEP, ClinVar)

**Cheminformatics:**
- Molecular descriptors and fingerpr

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (0)

No findings outside the package's declared scope.

Gates applied: no_behavioural_pass.