The first bioinformatics-native AI agent skill library.
Built on OpenClaw (180k+ GitHub stars). Local-first. Privacy-focused. Reproducible.

AI Agents for Health: ClawBio Hackathon
Thu 23 April 2026, 12:00-19:00 · University of Westminster, Cavendish Campus, London
Register free on Luma

41 executable skills + 8,000 Galaxy tools. Local-first. No cloud. No guessing.

Snap a photo of a medication in Telegram. ClawBio identifies the drug from the packaging, queries your pharmacogenomic profile from your own genome, and returns a personalised dosage card — on your machine, in seconds:

Warfarin | CYP2C9 *1/*2 Intermediate · VKORC1 High Sensitivity AVOID — DO NOT USE · Standard dose causes over-anticoagulation in this genotype.

Or take any genetic variant (identified by its rsID — a unique label like rs9923231) and search nine genomic databases at once to find every known disease association, tissue-specific effect, and population frequency. Or estimate your genetic predisposition to conditions like type 2 diabetes by combining thousands of small-effect variants into a single polygenic risk score. Or explore the UK Biobank — a half-million-person research dataset — by asking in plain English what fields measure blood pressure, grip strength, or depression, and get back the exact field IDs, descriptions, and linked publications you need.

Every result ships with a reproducibility bundle: commands.sh, environment.yml, and SHA-256 checksums. A reviewer can reproduce your Figure 3 in 30 seconds without emailing you.

ClawBio's demo data is built on a real, fully open human genome: the Corpasome. The 23andMe SNP chip (~600K variants) has been available since launch. Now, the project also ships subsets from a 30x Illumina whole-genome sequence (GRCh37), covering ~4M SNPs, ~600K indels, and structural variants (DEL, DUP, INV, BND, INS, CNVs). All data comes from a single individual (Manuel Corpas), licensed CC0, and published on Zenodo (doi:10.5281/zenodo.19297389). This dataset is provided for research and educational purposes only.

See docs/reference-genome.md for use cases, subsets, and citation details.

You read a paper. You want to reproduce Figure 3. So you:

1. Go to GitHub. Clone the repo.
2. Wrong Python version. Fix dependencies.
3. Need the reference data — where is it?
4. Download 2GB from Zenodo. Link is dead.
5. Email the first author. Wait 3 weeks.
6. Paths are hardcoded to /home/jsmith/data/.
7. Two days later: still broken. You give up.

Now imagine the same paper published a skill:

python ancestry_pca.py --demo --output fig3
# Figure 3 reproduced. Identical. SHA-256 verified. 30 seconds.

That's ClawBio. Every figure in your paper should be one command away from reproduction.

Current agentic bioinformatics systems address either the reasoning layer (constraining LLM outputs with knowledge graphs or fine-tuning) or the connectivity layer (wrapping tools as MCP servers). Neither addresses the specification layer: the encoding of a domain expert's analytical decisions into a machine-readable contract that constrains agent behaviour. Without this layer, the agent must reconstruct expert knowledge from its training distribution, a stochastic, unversioned process.

A skill is a self-contained directory comprising a declarative specification (SKILL.md), validated Python code, demo data, and a reproducibility bundle (commands.sh, environment.yml, SHA-256 checksums). The specification is a contract, not a prompt: it encodes the domain expert's analytical decisions so the LLM orchestrates but does not improvise.

Ad-hoc LLM code generation  = stochastic, unversioned, unverifiable
ClawBio skill                = specification-constrained, versioned, reproducible

• Specification-first: Domain expertise resides in SKILL.md, not in model weights. Specifications are versioned, human-readable, peer-reviewable, and trivially updatable.
• Agent-agnostic: Skills execute identically whether invoked by Claude, ChatGPT, or a locally hosted model via Ollama. Reproducibility is decoupled from any specific AI vendor.
• Local-first: Your genomic data never leaves your laptop. No cloud uploads, no data exfiltration.
• Reproducible: Every analysis exports commands.sh, environment.yml, and SHA-256 checksums. Anyone can reproduce it without the agent.
• MIT licensed: Open-source, free, community-driven.

Ask any LLM to "profile the pharmacogenes in my VCF file." It will produce plausible Python. But the code may use outdated CPIC guidelines, hallucinate star allele functional classifications, or confuse reduced-function and no-function alleles. CYP2D6*4 is a no-function allele; misclassifying it as reduced-function determines whether a patient receives a dose adjustment or is told to avoid a drug entirely. Approximately 7% of individuals of European ancestry are CYP2D6 poor metabolisers for whom codeine provides zero analgesic effect, and roughly 0.5% carry DPYD variants where standard fluorouracil dosing can be lethal.

An LLM should not be improvising these from training data. ClawBio encodes the correct bioinformatics decisions in versioned, peer-reviewable specifications so the agent executes them correctly every time.

Every ClawBio analysis ships with a reproducibility bundle — not as an afterthought, but as part of the output:

report/
├── report.md              # Full analysis with figures and tables
├── figures/               # Publication-quality PNGs
├── tables/                # CSV data tables
├── commands.sh            # Exact commands to reproduce
├── environment.yml        # Conda environment snapshot
└── checksums.sha256       # SHA-256 of every input and output file

Why this matters: a reviewer can re-run your analysis in 30 seconds. A collaborator can reproduce your Figure 3 without emailing you. Future-you can regenerate results two years later from the same bundle.

Wrap your bioinformatics pipeline as a skill and submit a PR. One community-contributed skill (NutriGx Advisor) is already in production; eight more have specifications authored and are awaiting implementation.

cp templates/SKILL-TEMPLATE.md skills/<your-skill-name>/SKILL.md
# Edit SKILL.md, add Python implementation, demo data, and tests

See CONTRIBUTING.md for the full submission process. Join the contributors community on Telegram: t.me/ClawBioContributors.

Genomebook is a synthetic-genetics sandbox built into ClawBio. It turns fictional or historical character profiles ("souls") into diploid genomes, scores compatibility, and breeds offspring across generations, complete with Mendelian inheritance, de novo mutations, and clinical evaluation.

SOUL.md  -->  Soul2DNA  -->  .genome.json  -->  GenomeMatch  -->  Recombinator  -->  Gen-N offspring
 (trait       (compiler)     (diploid loci)     (M x F rank)     (meiosis +        (.genome.json
  scores)                                                         mutation)          with clinical
                                                                                     history)

• 20 souls in GENOMEBOOK/DATA/SOULS/ (Einstein, Curie, Turing, Hypatia, Da Vinci, ...)
• 20 generation-0 genomes in GENOMEBOOK/DATA/GENOMES/
• 26 traits across 55 loci (trait_registry.json)
• Disease registry with penetrance, fitness costs, and onset probabilities

# Compile all souls to genomes
python skills/soul2dna/soul2dna.py --demo

# Score all M x F compatibility pairings
python skills/genome-match/genome_match.py --demo

# Breed Einstein x Anning (3 offspring)
python skills/recombinator/recombinator.py --demo

• Additive, dominant, recessive inheritance at each locus
• Heterozygosity advantage (heterosis) in compatibility scoring
• Carrier risk flagging for autosomal recessive conditions
• De novo mutations with hotspot categories (cognitive, immune, metabolic)
• Clinical evaluation with penetrance, onset probability, and fitness costs
• Health scores derived from cumulative condition burden

Generates a pharmacogenomic report from consumer genetic data (23andMe, AncestryDNA):

• Parses raw genetic data (auto-detects format, including gzip)
• Extracts 31 pharmacogenomic SNPs across 12 genes (CYP2C19, CYP2D6, CYP2C9, VKORC1, SLCO1B1, DPYD, TPMT, UGT1A1, CYP3A5, CYP2B6, NUDT15, CYP1A2)
• Calls star alleles and determines metabolizer phenotypes
• Looks up CPIC drug recommendations for 51 medications
• Zero dependencies. Runs in < 1 second.

python pharmgx_reporter.py --input demo_patient.txt --output report

Demo result: CYP2D6 *4/*4 (Poor Metabolizer) → 10 drugs AVOID (codeine, tramadol, 7 TCAs, tamoxifen), 20 caution, 21 standard.

~7% of people are CYP2D6 Poor Metabolizers — codeine gives them zero pain relief. ~0.5% carry DPYD variants where standard 5-FU dose can be lethal. This skill catches both.

Snap a photo of any medication in Telegram. ClawBio identifies the drug from the packaging and returns a personalised dosage card against your own genotype.

• Claude vision extracts drug name and visible dose from the photo
• Cross-references your 23andMe genotype against 31 PGx SNPs
• Four-tier classification: STANDARD DOSING / USE WITH CAUTION / AVOID / INSUFFICIENT DATA
• Correct VKORC1 complement-strand handling (23andMe reports minus strand for rs9923231)
• Works for warfarin, clopidogrel, codeine, simvastatin, tamoxifen, sertraline, and 20+ others

python pharmgx_reporter.py --drug warfarin --dose "5mg" --input my_23andme.txt --output report

No command needed in Telegram — send any medication photo and RoboTerri triggers the skill automatically.

Federated variant query across nine genomic databases in a single command:

python gwas_lookup.py --rsid rs3798220 --output report
python gwas_lookup.py --demo --output /tmp/gwas_lookup_demo

Semantic search across the UK Biobank schema. Ask in plain English what UK Biobank measures about any phenotype — get field IDs, descriptions, data types, participant counts, and linked publications back instantly.

python ukb_navigator.py --query "grip strength"   --output report
python ukb_navigator.py --field 21001              --output report   # BMI
python ukb_navigator.py --demo                     --output /tmp/ukb_demo

Built on a ChromaDB embedding of the full UKB Data Showcase (22,000+ fields).

Runs principal component analysis on your cohort against the SGDP reference panel (345 samples, 164 global populations):

• Contig normalisation (chr1 vs 1)
• IBD removal (related individuals filtered)
• Common biallelic SNPs only
• Confidence ellipses per population
• Publication-quality 4-panel figure generated instantly

python ancestry_pca.py --demo --output ancestry_report

Demo result: 736 Peruvian samples across 28 indigenous populations. Amazonian groups (Matzes, Awajun, Candoshi) sit in genetic space that no SGDP population occupies — genuinely underrepresented, not just in GWAS, but in the reference panels themselves.

Computes a Semantic Isolation Index for diseases using 13.1M PubMed abstracts and PubMedBERT embeddings (768-dim):

• SII (Semantic Isolation Index): higher = more isolated in literature
• KTP (Knowledge Transfer Potential): higher = more cross-disease spillover
• RCC (Research Clustering Coefficient): diversity of research approaches
• Temporal Drift: how research focus evolves over time
• Publication-quality 4-panel figure

python semantic_sim.py --demo --output sem_report

Key finding: Neglected tropical diseases are +38% more semantically isolated (P < 0.0001, Cohen's d = 0.84). 14 of the 25 most isolated diseases are Global South priority conditions. Knowledge silos kill innovation — a malaria immunology breakthrough could help leishmaniasis, but the literatures don't talk to each other.

Corpas et al. (2026). HEIM: Health Equity Index for Measuring structural bias in biomedical research. Under review.

git clone https://github.com/ClawBio/ClawBio.git && cd ClawBio
pip install -r requirements.txt
python clawbio.py run pharmgx --demo

PharmGx demo runs in <2 seconds. Only needs Python 3.10+.

Note: ClawBio is currently installed by cloning the repository. There is no pip install clawbio package yet (planned for a future release).

from clawbio import run_skill, list_skills

# List available skills
skills = list_skills()

# Run pharmacogenomics with demo data
result = run_skill("pharmgx", demo=True)

# Run with your own input
result = run_skill("gwas-lookup", rsid="rs3798220", output="/tmp/report")

Inside Claude Code:

/plugin marketplace add ClawBio/ClawBio
/plugin install clawbio

All skills are then available as agent-routable commands. Alternatively, clone the repo and open it as your working directory in Claude Code; the CLAUDE.md at the repo root teaches Claude how to route requests to skills automatically.

python clawbio.py list                           # See available skills
python clawbio.py run pharmgx --demo             # Pharmacogenomics (1s)
python clawbio.py run equity --demo              # Equity scoring (55s)
python clawbio.py run nutrigx --demo             # Nutrigenomics (60s)
python clawbio.py run metagenomics --demo        # Metagenomics (3s)
python clawbio.py run scrna --demo               # scRNA clustering + marker detection (PBMC3k-first demo)
python clawbio.py run scrna --demo --doublet-method scrublet
                                                 # Optional doublet detection before clustering
python clawbio.py run compare --demo             # Manuel Corpas vs George Church (10s)
python clawbio.py run gwas-lookup --demo         # rs3798220 across 9 databases (5s)
python clawbio.py run prs --demo                 # Polygenic risk scores (10s)
python clawbio.py run ukb-navigator --demo       # UK Biobank schema search (5s)
python clawbio.py run profile --demo             # Unified genomic profile (30s)
python clawbio.py run galaxy --demo              # Galaxy Bridge FastQC demo (offline)
python clawbio.py run rnaseq --demo              # RNA-seq DE demo (bulk/pseudo-bulk)
python clawbio.py run methylation --demo        # Epigenetic methylation clocks via PyAging
python clawbio.py run protocols-io --demo       # Protocols.io protocol search
python clawbio.py run variant-annotation --demo # VCF variant annotation (VEP + ClinVar + gnomAD)
python clawbio.py run proteomics --demo         # Proteomics differential expression
python clawbio.py run clinical-trial --demo     # Clinical trial finder

python clawbio.py run pharmgx --input my_23andme.txt --output results/
python clawbio.py run rnaseq --input counts.csv,metadata.csv --output results_rnaseq/
python clawbio.py run methylation --geo-id GSE139307 --output results_methylation/

pip install pytest
python -m pytest

Core dependencies (requirements.txt): biopython, pandas, numpy, scikit-learn, matplotlib, openai, pydeseq2. Most skills run with just these.

Some skills have additional requirements:

No Docker or Singularity required for core functionality. Skills that need external bioinformatics tools document their setup in their own SKILL.md.

RoboTerri — ClawBio's Telegram agent, inspired by Prof. Teresa K. Attwood

ClawBio skills are available through RoboTerri, a public Telegram bot running against a real human genome (Manuel Corpas, CC0 public domain). Named after Prof. Teresa K. Attwood — a pioneer of bioinformatics education, founding Chair of GOBLET, and winner of the 2021 ISCB Outstanding Contributions Award.

Try RoboTerri now — no install needed →

Ask it anything:

• "Give me my pharmacogenomic summary" — analyses 12 genes, 51 drugs
• "What diseases am I at risk for?" — polygenic risk scores for 6 conditions
• Send a photo of any medication — checks CYP2D6/CYP2C19 metaboliser status
• /demo pharmgx /demo prs /demo nutrigx /demo compare /demo profile

You:        [send 23andMe file]
RoboTerri:  Running PharmGx Reporter...
            CYP2D6 *4/*4 — Poor Metabolizer → 10 drugs AVOID
            [report.md attached]
            [3 figures attached]

You:        [send photo of warfarin packet]
RoboTerri:  Warfarin detected. Running Drug Photo skill...
            CYP2C9 *1/*2 · VKORC1 High Sensitivity
            AVOID — DO NOT USE at standard dose.

You:        run gwas-lookup rs3798220
RoboTerri:  Querying 9 databases...
            rs3798220 (LPA) — coronary artery disease, Lp(a) levels.
            eQTL in liver (GTEx). gnomAD MAF 0.07.

RoboTerri auto-detects file type (23andMe .txt, AncestryDNA .csv, VCF, FASTQ) and routes to the right skill via the Bio Orchestrator. Photos of medications trigger the Drug Photo skill automatically — no command needed.

Install your own RoboTerri: Set up your own Telegram bot running ClawBio skills in ~20 minutes.

ClawBio indexes 8,000+ bioinformatics tools from usegalaxy.org via the Galaxy Bridge skill. Search by natural language, inspect tool schemas, and execute remotely — all from the CLI.

# Search Galaxy tools by keyword
python skills/galaxy-bridge/galaxy_bridge.py --search "metagenomics"

# Browse all 86 tool categories
python skills/galaxy-bridge/galaxy_bridge.py --list-categories

# Run a tool on Galaxy (requires GALAXY_API_KEY)
python skills/galaxy-bridge/galaxy_bridge.py --run fastqc --input reads.fq.gz --output results/

# Demo mode (offline, no API key)
python skills/galaxy-bridge/galaxy_bridge.py --demo

Cross-platform chaining: Galaxy VEP annotates variants → ClawBio PharmGx generates dosage report. Galaxy Kraken2 classifies reads → ClawBio metagenomics profiler. Neither can do this alone.

Built on BioBlend (Galaxy Python SDK). Developed in collaboration with the Galaxy ML SIG.

Telegram (RoboTerri)     CLI (clawbio.py)     Python (import clawbio)
         │                      │                       │
         └──────────┬───────────┘───────────────────────┘
                    │
             ┌──────▼──────┐
             │  Bio         │  ← routes by file type + keywords
             │  Orchestrator│
             └──────┬──────┘
                    │
  ┌─────────────────▼──────────────────────────────────────┐
  │                                                         │
  PharmGx    Equity     NutriGx    Metagenomics   Ancestry
  Reporter   Scorer     Advisor    Profiler        PCA    ...
  │                                                         │
  └─────────────────┬──────────────────────────────────────┘
                    │
             ┌──────▼──────┐
             │  Markdown    │  ← report + figures + checksums
             │  Report      │     + reproducibility bundle
             └─────────────┘

Each skill is standalone — the orchestrator routes to the right one, but every skill also works independently. The clawbio.run_skill() API is importable by any agent (RoboTerri, RoboIsaac, Claude Code).

See docs/architecture.md for the full design.

ClawBio is designed to be discovered and used by AI coding agents, not just humans.

Agents can also run python clawbio.py list to discover available skills programmatically.

Open skill requests (PRs welcome):

See Contributing a Skill above for the submission process.

• London Bioinformatics Meetup (26 Feb 2026): project announcement. Slides.
• UK AI Agent Hack, Imperial College London (1 Mar 2026): introduced ClawBio to Peter Steinberger, creator of OpenClaw. Video.
• DoraHacks Demo Day, Imperial College London (7 Mar 2026): live demo of pharmacogenomics, intelligent routing, multi-channel agents, and Drug Photo. Video.

ClawBio follows Semantic Versioning. The current release is v0.4.0. See CHANGELOG.md for a full history of additions and breaking changes.

• Try RoboTerri: t.me/RoboTerri_bot -- query a real genome on Telegram, no install needed
• Slides: clawbio.github.io/ClawBio/slides/
• Tutorial: Install your own RoboTerri
• OpenClaw: github.com/openclaw/openclaw -- the agent platform
• ClawHub: clawhub.ai -- skill registry
• HEIM Index: heim-index.org -- Health Equity Index for Minorities

MIT. See LICENSE.