🦖 ClawBio — Bioinformatics AI Agent Skills

v0.7.0 · Agent Skills

Genomics skills, in your editor

Every ClawBio skill is a plain Agent Skills folder: a SKILL.md contract, the script that executes it, demo data and tests. Claude Code installs the whole library as a plugin. Cursor, VS Code, Codex, Zed and any other editor that reads Agent Skills load the folders from ~/.agents/skills/. The runnable skills execute on your machine and hand back the report, the structured result, and a commands.sh with SHA-256 checksums, so the analysis can be replayed without the agent that produced it.

            > /plugin marketplace add ClawBio/ClawBio
          

Then /plugin install clawbio. Other editors: copy or symlink the skill folders you need into ~/.agents/skills/.

✅ 97 skills, 50 with a deterministic CLI entry point
🔒 Demo data included, your files only when you pass them
🏠 Local-first. No server, no hosted endpoint, by design

# Claude Code
/plugin marketplace add ClawBio/ClawBio
/plugin install clawbio

# Cursor, VS Code, Codex, Zed: Agent Skills
git clone https://github.com/ClawBio/ClawBio
cp -r ClawBio/skills/pharmgx-reporter ~/.agents/skills/

# Python, any language
pip install clawbio
clawbio run pharmgx --demo

The MCP server introduced in v0.6.0 is deprecated as of v0.7.0 and will be removed in v0.8.0. Existing clawbio mcp configurations keep working until then and print a notice on start. Why, and how to migrate.

🚀 What is new in v0.7.0 🏷️ Release notes 📄 MCP migration guide

The Problem

Reproducing bioinformatics analyses is broken

# Without ClawBio
$ git clone repo && cd repo
# Wrong Python version...
# Reference data link is dead...
# Paths hardcoded to /home/jsmith/data/
# Email first author. Wait 3 weeks.
# Give up.

✗ Dependency hell & broken environments
✗ Dead links to reference data
✗ Hardcoded paths & missing configs
✗ AI hallucinates star alleles & outdated guidelines
✗ No audit trail, no checksums

# With ClawBio
$ python skills/multiqc-reporter/multiqc_reporter.py --demo --output /tmp/multiqc_demo

✓ Demo report regenerated.
✓ Replay command saved alongside outputs.
✓ Checksums available for verification.

# Or with natural language:
$ openclaw "Profile my pharmacogenes"

Many skills write a reproducibility/ bundle with replay commands, environment metadata, and output checksums. Exact files vary by skill, and some replays still require local tools or original inputs.

Why ClawBio

Domain expertise, frozen into code

ChatGPT and Claude are smart generalists. ClawBio skills are domain experts' proven pipelines — executed correctly every time.

🏠

Local-first by default

Your genome is analysed on your machine. Skills that transmit data to external services are clearly labelled and run only when you invoke them.

🔁

Reproducible

Many skills export replay bundles with commands, environment metadata, and output checksums so runs can be rechecked without the original agent session.

🧩

Modular

Each skill is a self-contained directory (SKILL.md + Python scripts) that plugs into the orchestrator — or runs standalone.

⚖️

MIT Licensed

Open-source, free, community-driven. Clone it, run it, build a skill, submit a PR.

🌍

Equity-aware

Built-in support for underrepresented populations. HEIM diversity metrics baked into the roadmap.

🎯

Benchmark Validated

Public scientific-correctness leaderboard: 161/182 tests passing (88.5%) as of 26 Jul 2026, across 10 of 95 skills, every failure tagged. The suite currently reports an overall FAIL: nutrigx-advisor regressed from 10/10 to 0/10. AD ground truth (34 genes, 3 evidence tiers), mock API server for offline CI, swappable fine-mapping (ABF vs SuSiE), benchmark scorer with precision/recall/F1. View leaderboard · DOI

🦞

Built on OpenClaw

Powered by OpenClaw, the open agent platform that routes natural language to the right specialist skill. ClawBio is the bioinformatics catalogue any OpenClaw-compatible agent can call.

🔌

Claude Code Plugin

Install with two commands: /plugin marketplace add ClawBio/ClawBio then /plugin install clawbio. All skills instantly available.

Skills Library

97 skills · 92 with runnable demo data · 4,654 tests · 161 / 182 publicly benchmarked passing

Each skill includes demo data so you can try it immediately without your own files.

💊 PharmGx Reporter

Pharmacogenomic report from 23andMe/AncestryDNA: 12 genes, 31 SNPs, 51 drugs, CPIC guidelines.

MVP

🥦 NutriGx Advisor

Personalised nutrition report from consumer genetic data: MTHFR, folate, vitamin D, caffeine, lactose, omega-3, and more.

MVP

🔍 GWAS Lookup

Federated variant lookup across 9 genomic databases: GWAS Catalog, Open Targets, PheWeb (UKB, FinnGen, BBJ), GTEx, eQTL Catalogue, and more.

MVP

📊 GWAS Polygenic Risk

Calculate polygenic risk scores from DTC genetic data using curated PGS Catalog scores across 6 traits.

MVP

⚖️ Equity Scorer

Compute HEIM diversity and equity metrics from VCF or ancestry data: heterozygosity, FST, and population representation scores.

MVP

+ 92 more skills on GitHub · see which ones are publicly benchmarked →

In Production

ClawBio skills, running in conversation

ClawBio skills aren't just CLI tools — they power live AI agents that talk to real users every day.

RoboTerri

Telegram AI Agent

A bioinformatics intelligence agent on Telegram — inspired by Professor Teresa K. Attwood, a pioneer of bioinformatics education and research. Run pharmacogenomics reports from your own genetic data. Snap a photo of your medication, get personalised dosage guidance. Local-first, with reproducibility support where the underlying skill writes a replay bundle.

PharmGx Reporter Drug Photo Lookup NutriGx Advisor Genome Compare Equity Scorer Metagenomics GWAS Lookup scRNA Orchestrator

Repo	OpenClaw	Stars	Forks	Skills
ClawBio	✓	1,124	258	97
OpenClaw-Medical-Skills	✓	1,037	123	869
bioSkills	✓	331	52	425
LabClaw	✓	317	54	211
claude-scientific-skills	~	14.5k	1,585	250
Biomni	—	2,826	508	314

Recent Updates

Latest from the community

🐛

Fix(clinical-variant-reporter): parse ClinVar significance into terms instead of crashing on VEP's list. The clinical variant reporter was crashing when VEP returned ClinVar significance as a delimited list rather than a single term. Significance is now parsed correctly across all list formats. Contributed by @krudo-taco.

PR #382 · merged · 4 September 2026

🚀

Release v0.7.0: deprecate the MCP server, refresh the Claude Code plugin manifests. The MCP server is deprecated and will be removed in v0.8.0. ClawBio now installs as a Claude Code plugin or plain Agent Skills folder; all runnable skills are available via /plugin install clawbio. Contributed by @manuelcorpas.

PR #384 · merged · 3 September 2026

⚙️

Feat(gwas-prs): add reproducibility bundle. The gwas-prs skill now emits a full reproducibility bundle via the shared clawbio.common layer, recording input SHA-256 digests, portable replay commands, environment provenance, and output checksums, without storing genotype paths or raw trait queries. Contributed by @krudo-taco.

PR #379 · merged · 3 September 2026

🐛

Fix(pathway-enricher): submit gene lists as multipart form data. The pathway enricher skill was sending gene lists as URL-encoded parameters, which caused failures for large gene sets exceeding query string length limits. Gene lists are now submitted as multipart form data to support arbitrary input sizes. Contributed by @krudo-taco.

PR #381 · merged · 2 September 2026

🐛

Fix(gwas-prs): give curated panels honest identifiers. Curated PGS panel identifiers were using names that could be confused with real PGS Catalog IDs, causing downstream tools to attempt live API lookups against demo scores. Panel identifiers are now prefixed to make their demo-only status unambiguous. Contributed by @krudo-taco.

PR #380 · merged · 1 September 2026

🐛

Fix(rnaseq-de): verify LFC shrinkage matches the requested contrast. Log-fold-change shrinkage was being applied to the wrong contrast in some multi-factor designs, producing incorrect shrunken estimates in the differential expression output. Contributed by @krudo-taco.

PR #378 · merged · 31 August 2026

⚙️

CI: run vcf-annotator tests on the main allowlist. The vcf-annotator test suite is now included in the main CI allowlist, ensuring annotation correctness is checked on every push rather than only in scheduled runs. Contributed by @krudo-taco.

PR #370 · merged · 31 August 2026

⚙️

Feat: equity-scorer writes reproducibility bundle via clawbio.common. The equity-scorer skill now emits a full reproducibility bundle using the shared clawbio.common helpers, matching the pattern adopted across other MVP skills. Contributed by @camlloyd.

PR #352 · merged · 31 August 2026

🐛

Fix(catalog): sync nutrigx version. The catalog entry for nutrigx had a stale version string that did not match the implementation; now kept in sync automatically.

PR #377 · merged · 30 August 2026

🐛

Fix(nutrigx): correct rs4988235 lactose polarity and dominance. The lactose-persistence variant rs4988235 was encoded with the wrong risk polarity and an incorrect dominance model, leading to inverted dietary guidance for heterozygous carriers. Contributed by @krudo-taco.

PR #376 · merged · 30 August 2026

View all merged pull requests on GitHub →

The first bioinformatics-native
AI agent skill library

Genomics skills, in your editor

Get started in 30 seconds

The first bioinformatics-nativeAI agent skill library

Genomics skills, in your editor

Get started in 30 seconds

The first bioinformatics-native
AI agent skill library