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

A community contributor built a nutrigenomics skill and ran it — from raw genetic data to personalised nutrition report with radar charts, heatmaps, and reproducibility bundle:

https://github.com/ClawBio/ClawBio/releases/download/v0.2.0/david-nutrigx-demo.mp4

PharmGx Reporter: 12 genes, 51 drugs, under 1 second

Manuel Corpas introduces ClawBio to Peter Steinberger at the UK AI Agent Hack (1 March 2026):

Click to watch on YouTube

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.

A skill is a domain expert's knowledge — frozen into code — that an AI agent executes correctly every time.

ChatGPT / Claude  = a smart generalist who guesses at bioinformatics
🦖 ClawBio skill  = a domain expert's proven pipeline that the AI executes

• 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.
• Modular: Each skill is a self-contained directory (SKILL.md + Python scripts) that plugs into the orchestrator.
• MIT licensed: Open-source, free, community-driven.

Ask Claude to "profile my pharmacogenes from this 23andMe file." It'll write plausible Python. But:

• It hallucinates star allele calls and uses outdated CPIC guidelines
• It forgets CYP2D6 *4 is no-function (not reduced)
• You spend 45 minutes debugging its output
• No reproducibility bundle. No audit log. No checksums.

ClawBio encodes the correct bioinformatics decisions so the agent gets it right first time, 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.

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

• Parses raw genetic data (auto-detects format)
• 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.

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+.

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 compare --demo            # Manuel Corpas vs George Church (10s)

python clawbio.py run pharmgx --input my_23andme.txt --output results/

pip install pytest
python -m pytest

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

ClawBio skills are also available through RoboTerri, our Telegram AI agent — named after Prof. Teresa K. Attwood, a pioneer of bioinformatics education and computational biology in the UK. Send a genetic data file or ask for a demo — get back a summary, full report, and figures directly in Telegram.

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

RoboTerri auto-detects file type (23andMe .txt, AncestryDNA .csv, VCF, FASTQ) and routes to the right skill via the Bio Orchestrator. You can also ask explicitly:

• "run pharmgx demo" — PharmGx with synthetic patient data
• "run equity demo" — HEIM equity score with demo populations
• "run nutrigx demo" — Nutrigenomics with synthetic genotypes

The integration uses the same clawbio.run_skill() API, so results are identical whether you run via CLI or Telegram. See 01-AGENTS/02-ROBOTERRI for the full agent source.

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.

We want skills from the bioinformatics community. If you work with genomics, proteomics, metabolomics, imaging, or clinical data — wrap your pipeline as a skill.

See CONTRIBUTING.md for the submission process and templates/SKILL-TEMPLATE.md for the skill template.

ClawBio was announced at the London Bioinformatics Meetup on 26 February 2026.

• Slides: clawbio.github.io/ClawBio/slides/
• Talk: 10 Tips for Becoming a Top 1% AI User — with live demos of all three MVP skills

If you use ClawBio in your research, please cite:

@software{clawbio_2026,
  author = {Corpas, Manuel},
  title = {ClawBio: An Open-Source Library of AI Agent Skills for Reproducible Bioinformatics},
  year = {2026},
  url = {https://github.com/ClawBio/ClawBio}
}

• 🦖 Slides: clawbio.github.io/ClawBio/slides/
• OpenClaw — The agent platform
• ClawHub — Skill registry
• HEIM Index — Health Equity Index for Minorities

MIT — clone it, run it, build a skill, submit a PR. 🦖