ClawOmics

Professional AI-Driven Bioinformatics Orchestration for OpenClaw

🧬 Your Intelligent Lab Partner

ClawOmics transforms your OpenClaw instance into a bioinformatics agent framework. By combining a master orchestrator with a library of specialized scientific skills, it turns raw biological data into a confirmable and executable analysis workflow.

⚡ Simplest Usage

If you only care about the easiest way to run ClawOmics, use this:

cd /Users/zhangyifan/clawomics
npm install
npm link
clawomics start

After clawomics start, keep that process running and do the rest in your MCP-enabled chat client.

If your MCP client already knows how to spawn clawomics-mcp-server, you usually do not need clawomics start at all. In production, the host should auto-launch the server on demand.

The intended end-user flow is:

start ClawOmics once
open OpenClaw / Codex / another MCP-capable client
say where the data are
say "确认执行" when you want it to proceed

Why ClawOmics?

🧠 Automatic Planning: ClawOmics profiles your dataset (FASTQ, H5AD, BAM, VCF) and generates a structured first-pass analysis plan.
🧩 Mixed Dataset Triage: Mixed input folders are partitioned into analysis units so raw reads, VCFs, and processed tables can be handled separately.
🧪 Assay Routing: Raw sequencing inputs now produce assay candidates such as bulk-rnaseq or dna-seq, with follow-up questions when confidence is low.
🧭 Agent Framework: Outputs include explicit agent state, confirmation gates, and persistent run artifacts that OpenClaw can use across turns.
🛠️ Batteries Included: Pre-integrated with 200+ skills including Scanpy, DeepTools, Biopython, and database connectors for Ensembl, ClinVar, and AlphaFold.
📦 Seamless Environment Control: Automated Conda and Mamba management to ensure reproducible, version-stable scientific workflows.
📖 AI-Driven Narrative: Technical results are translated into biological insights, providing context-aware summaries of complex multi-omics data.

🆕 Release Notes

v1.2

🧵 Context Isolation: chat hosts can now pass context_key so Feishu, Telegram, and other concurrent threads do not share the same remembered bridge state
🧹 Context Cleanup: new clear-context CLI and clawomics_clear_context MCP tool to drop stale lightweight bridge state safely
🔧 CLI Interface: New clawomics.mjs CLI for one-command operations
🧰 Global Command Entry: npm link now exposes a reusable clawomics command
🪄 One-Command Startup: clawomics start checks MCP readiness and starts the chat bridge
🔌 MCP Server: New local MCP server for chat-first integration with OpenClaw and other MCP-capable clients
🗂️ Dataset Profiling: bio-expert now emits structured dataset profiles for OpenClaw
🧭 Auto Planning: New plan command builds first-pass workflows from detected evidence
🪓 Dataset Partitioning: New partition command separates mixed directories into analysis units
💾 JSON Artifacts: profile, partition, and plan can now be written to disk for downstream automation
▶️ Confirmed Run Bootstrap: run creates a tracked workspace with a manifest and step scripts after the user confirms execution
🧪 Demo Data Generator: generate_demo_data.mjs creates test datasets instantly
🧠 Working Orchestrator: bio-expert/scripts/orchestrator.mjs profiles datasets and drafts workflow plans
📊 Resource Summary: Auto-generated skill statistics table in RESOURCES.md
📖 Cookbook: New docs/COOKBOOK.md with prompt templates

🏗️ Architecture

ClawOmics now operates as a chat-first workflow layer with three surfaces:

chat clients call the MCP bridge
the bridge calls the bio-expert orchestrator
the orchestrator emits durable artifacts and run workspaces

This is the intended boundary:

bio-expert skill: domain policy and workflow semantics
orchestrator: real profiling, planning, session, and run logic
MCP bridge: thin transport layer for chat clients

If you only need a local OpenClaw skill, you can think mostly in terms of bio-expert. If you need Feishu, Telegram, or other chat channels to auto-route and resume state, the MCP bridge becomes necessary.

graph TD
    User["User"] --> Client["OpenClaw / Codex / Gemini Host"]
    Client --> MCP["clawomics-mcp-server"]
    MCP --> Turn["clawomics_agent_turn"]
    Turn --> Orchestrator["bio-expert orchestrator"]

    Orchestrator --> Profile["dataset_profile.json"]
    Orchestrator --> Partitions["dataset_partitions.json"]
    Orchestrator --> Plan["analysis_plan.json"]
    Orchestrator --> Session["agent_session.json"]
    Orchestrator --> Bridge[".clawomics/openclaw_context.json or .clawomics/contexts/<context-key>.json"]

    Orchestrator --> Skills["skills/ registry"]
    Skills --> S1["scanpy / scvi-tools"]
    Skills --> S2["deeptools / pysam"]
    Skills --> S3["database connectors"]

    Orchestrator --> Run["clawomics_runs/<run-id>/"]
    Run --> Manifest["run_manifest.json"]
    Run --> Commands["commands/*.sh"]

    classDef client fill:#f7f1e3,stroke:#8c6d1f,color:#222;
    classDef core fill:#e6f4ea,stroke:#2f855a,color:#222;
    classDef artifact fill:#e8f0fe,stroke:#356ac3,color:#222;
    classDef run fill:#fce8e6,stroke:#c53929,color:#222;

    class Client,MCP,Turn client;
    class Orchestrator,Skills,S1,S2,S3 core;
    class Profile,Partitions,Plan,Session,Bridge artifact;
    class Run,Manifest,Commands run;

Runtime Flow

sequenceDiagram
    participant U as User
    participant C as Chat Client
    participant M as MCP Server
    participant O as bio-expert

    U->>C: "/data/project1 里有数据，帮我分析"
    C->>M: clawomics_agent_turn(message)
    M->>O: handleAgentMessage()
    O-->>M: profile + plan + confirmation prompt
    M-->>C: assistantReply
    C-->>U: 展示计划并请求确认
    U->>C: "确认执行"
    C->>M: clawomics_agent_turn(message)
    M->>O: resume latest bridge state
    O-->>M: run workspace + manifest
    M-->>C: assistantReply + run paths
    C-->>U: 告知已创建运行目录

🚀 Quick Start

1. Installation

Clone ClawOmics into your OpenClaw workspace skills directory:

cd ~/.openclaw/workspace/skills
git clone https://github.com/yf8578/clawomics.git

2. Minimal Daily Workflow

For normal use, you usually only need:

cd /Users/zhangyifan/clawomics
npm install
npm link
clawomics start

Then switch to your chat client and talk to it directly.

Advanced CLI and Debug Setup

If you want the lower-level commands for debugging or local testing:

cd clawomics
chmod +x scripts/*.mjs scripts/*.sh

# Initialize environment
node scripts/clawomics.mjs setup

# Simplest daily entrypoint: start the chat bridge once
node scripts/clawomics.mjs start

# Generate demo data for testing
node scripts/clawomics.mjs demo

# Natural-language entrypoint
node scripts/clawomics.mjs agent "demo_data 里有数据，帮我分析一下"

# OpenClaw-friendly compact payload
node scripts/clawomics.mjs agent "demo_data 里有数据，帮我分析一下" --compact

# Confirm and create a run workspace
node scripts/clawomics.mjs agent "确认执行"

# MCP helper commands
node scripts/clawomics.mjs mcp-doctor
node scripts/clawomics.mjs mcp-config
node scripts/clawomics.mjs mcp

# Build profile + partitions + plan in one step
node scripts/clawomics.mjs analyze demo_data --write

# Build a structured dataset profile
node scripts/clawomics.mjs profile demo_data --write

# Generate an automatic analysis plan
node scripts/clawomics.mjs plan demo_data --write

# Split mixed inputs into analysis units
node scripts/clawomics.mjs partition demo_data --write

# After the user confirms, bootstrap a runnable workspace
node scripts/clawomics.mjs run demo_data --approve

4. Initialize Resources

Update the skill inventory to register all 200+ skills:

node scripts/inventory_skills.mjs

This generates docs/RESOURCES.md with a summary table of all available tools.

5. Usage Example

Refer to our 📖 Cookbook for detailed prompt examples and scenarios.

User: "./data 里有一批测序数据，帮我看看该怎么分析。"

ClawOmics: "I detected a mixed directory containing FASTQ, VCF, and tabular outputs. I split this into raw-sequencing and variant-analysis units. For the FASTQ unit, assay routing is still low-confidence, so I recommend confirming whether the reads are DNA-seq or RNA-seq before alignment."

5.1 Generated Artifacts

When you add --write, ClawOmics writes machine-readable artifacts next to the input dataset:

analysis_bundle.json
dataset_profile.json
dataset_partitions.json
analysis_plan.json
agent_session.json

After run --approve, ClawOmics also creates a run workspace:

clawomics_runs/<run-id>/run_manifest.json
clawomics_runs/<run-id>/commands/*.sh

6. OpenClaw Usage Model

ClawOmics is designed to stay simple inside OpenClaw:

OpenClaw provides the model layer for planning and explanation.
ClawOmics provides the dataset profiler and workflow scaffolding.
No separate LLM configuration is required for the first-pass planning flow in this repository.
The recommended production integration is MCP, so users only interact through the chat box.

For manual local testing, the intended operator flow is:

clawomics start

After that, the rest should happen inside the chat client rather than through more ClawOmics commands.

For a real OpenClaw / Feishu / Telegram deployment, prefer letting the MCP host auto-spawn the server instead of keeping a separate terminal alive.

6.1 Intended OpenClaw Flow

User tells OpenClaw where the data live.
OpenClaw calls agent "<user-message>" or agent "<user-message>" --compact.
ClawOmics returns profile, partitions, and a first-pass plan.
User confirms execution.
OpenClaw calls agent "确认执行" and ClawOmics resumes from the persisted bridge state automatically.
ClawOmics creates a tracked run workspace and command templates.

agent_session.json is the durable per-dataset state. The lightweight conversation bridge is stored in .clawomics/openclaw_context.json by default, or .clawomics/contexts/<context-key>.json when the host passes a stable chat-specific context_key.

6.2 Framework Docs

docs/PRODUCT_FRAMEWORK.md: product positioning, scope, boundaries, and principles
docs/AGENT_PROTOCOL.md: state machine, OpenClaw flow, and artifact contract
docs/OPENCLAW_SYSTEM_PROMPT.md: ready-to-use system prompt template for the OpenClaw conversation layer
docs/OPENCLAW_MCP_SETUP.md: MCP-based integration guide for the cleanest chat-first experience
docs/CHAT_CHANNEL_ROUTING.md: automatic routing rules for Feishu, Telegram, and other mixed-purpose chat channels

📂 Project Navigation

skills/bio-expert: The core orchestration logic.
skills/: Library of 200+ integrated scientific skills.
docs/RESOURCES.md: Full inventory of available tools and categories.
docs/PRODUCT_FRAMEWORK.md: Product definition for the ClawOmics agent.
docs/AGENT_PROTOCOL.md: Runtime contract between OpenClaw and ClawOmics.
docs/OPENCLAW_INTEGRATION.md: Practical routing guide for OpenClaw conversation logic.
docs/OPENCLAW_SYSTEM_PROMPT.md: Prompt template for the OpenClaw integration layer.
docs/OPENCLAW_MCP_SETUP.md: MCP server setup for direct OpenClaw tool integration.
docs/CHAT_CHANNEL_ROUTING.md: Auto-routing strategy for Feishu, Telegram, and similar chat channels.
docs/INTEGRATION_PLAN.md: Future capability expansion roadmap.

👥 Contributors

_yf8578

_puppy-0000

🙏 Credits & Attributions

ClawOmics stands on the shoulders of giants. We gratefully acknowledge:

Claude Scientific Skills by K-Dense-AI (170+ core research skills).
BioClaw by Runchuan-BU (Specialized bio-logic and inspirations).
The OpenClaw Community for the underlying agent gateway infrastructure.

📄 License

Distributed under the MIT License. See LICENSE for details.

Built with 🧬 by yf8578

Name		Name	Last commit message	Last commit date
Latest commit History 19 Commits
assets		assets
docs		docs
mcp		mcp
scripts		scripts
skills		skills
.gitignore		.gitignore
README.md		README.md
package-lock.json		package-lock.json
package.json		package.json
permission_test.txt		permission_test.txt

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

ClawOmics

🧬 Your Intelligent Lab Partner

⚡ Simplest Usage

Why ClawOmics?

🆕 Release Notes

🏗️ Architecture

Runtime Flow

🚀 Quick Start

1. Installation

2. Minimal Daily Workflow

4. Initialize Resources

5. Usage Example

5.1 Generated Artifacts

6. OpenClaw Usage Model

6.1 Intended OpenClaw Flow

6.2 Framework Docs

📂 Project Navigation

👥 Contributors

🙏 Credits & Attributions

📄 License

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

ClawOmics

🧬 Your Intelligent Lab Partner

⚡ Simplest Usage

Why ClawOmics?

🆕 Release Notes

🏗️ Architecture

Runtime Flow

🚀 Quick Start

1. Installation

2. Minimal Daily Workflow

4. Initialize Resources

5. Usage Example

5.1 Generated Artifacts

6. OpenClaw Usage Model

6.1 Intended OpenClaw Flow

6.2 Framework Docs

📂 Project Navigation

👥 Contributors

🙏 Credits & Attributions

📄 License

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages