Guide · Agent skill
How to use the LaminDB biological data curation agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The LaminDB skill teaches an AI agent how to curate biological datasets with LaminDB, an open-source data management tool. It covers registering files as tracked artifacts, validating tables and AnnData objects against typed schemas, standardizing terms with Bionty ontologies, and recording which scripts produced which outputs. This guide explains what the skill does, how to install it, how to ask for it in plain language, and what to check before you trust its output.
LaminDB Biological Data CurationK-Dense-AI/scientific-agent-skills · MIT Download the skill Quick start
- 1Download the ZIP. Download the skill ZIP from the page where you found this guide.
- 2Unzip into a skills folder. Unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder.
- 3Create an empty project environment. Make a new directory, create a virtual environment and install the pinned LaminDB and Bionty packages with uv.
- 4Initialize a local instance. Run lamin init with a storage path and a name, enabling the bionty module.
- 5Ask for a curation task. Describe the dataset, the schema you need and the output key in plain language.
- 6Review validation and lineage. Check the validation result, the stored schema and the run record before you rely on the artifact.
What the LaminDB skill makes the agent do
An agent skill is a folder of instructions and reference files that an AI agent reads when a task matches the skill's description. This one is for managing biological datasets and models with LaminDB. Once it is active, the agent follows a defined workflow instead of improvising.
The workflow is: confirm the instance and storage with lamin info, define typed Feature records and a Schema, review ontology sources and standardize only reviewed synonyms, run the work in a tracked script, validate, save outputs as artifacts, and query metadata before loading content.
- Registers files and DataFrames as artifacts with keys
- Validates tables and AnnData objects against schemas
- Annotates terms using Bionty ontologies
- Records lineage: which run used which inputs and created which outputs
Install the skill from a ZIP or the Claude apps
Download the skill ZIP from the page where you found this guide, then choose one of two routes. For a personal setup, unzip it into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder.
In the Claude apps, you can instead upload the ZIP under Settings, then Capabilities, then Skills.
- Personal: unzip into ~/.claude/skills/
- Project only: unzip into .claude/skills/ inside the project
- Claude apps: Settings → Capabilities → Skills, then upload
Set up a safe local environment before running examples
The skill's worked example runs in a new, empty project directory with its own virtual environment. Local SQLite needs no login. Public ontologies and remote storage need network access, and private instances need credentials.
The skill says to use a new local development directory for experiments, so a trial does not change an existing instance or a shared cache.
- Use an empty project folder and a dedicated environment
- Check the target with lamin info before writing anything
- Keep API keys and database passwords out of outputs
uv venv --python 3.12
uv pip install 'lamindb==2.10.0' 'bionty==2.5.0' 'ipython==9.17.1'
source .venv/bin/activate
lamin init --storage ./storage --name biology-demo --modules biontyTrigger the skill with natural requests
You do not need to type the skill name. Describe the biological data task in plain words that match the skill's scope: registering datasets, curating DataFrame or AnnData metadata, querying annotated artifacts, or tracking a workflow.
Good requests name the data, the schema you want, and the output key. Vague requests such as clean my data may not match this skill, which is built for LaminDB-based curation rather than general cleaning.
- Register this AnnData file as an artifact with key scrna/batch.h5ad and tell me its schema
- Check this table against a schema requiring gene_count and condition, and reject extra columns
- Find the artifacts curated under the qc_metadata schema and show their keys
- Track this script as a run so the output records its inputs
Read a worked example: validate and save a table
The skill's example defines two features and a strict schema, validates a small DataFrame, saves it as a Parquet artifact, and checks the round trip. Validation raises an error on failure, so it is not used as an if condition.
This check covers table structure and types only. It does not decide whether the counts meet your assay's quality standards. Add assay-specific checks such as units, allowed ranges, missing values, duplicate sample IDs and batch balance before publication or downstream analysis.
import pandas as pd
import lamindb as ln
ln.track(params={"analysis": "QC metadata curation"})
count = ln.Feature(name="gene_count", dtype=int).save()
condition = ln.Feature(name="condition", dtype=str).save()
schema = ln.Schema(
name="qc_metadata",
features=[count, condition],
maximal_set=True, # reject extra columns
).save()
df = pd.DataFrame({
"gene_count": [120, 130],
"condition": ["control", "treated"],
})
curator = ln.curators.DataFrameCurator(df, schema)
curator.validate() # raises on invalid data; do not use as an if condition
artifact = curator.save_artifact(key="experiments/qc.parquet")
assert artifact.schema == schema
pd.testing.assert_frame_equal(artifact.load(), df)
ln.finish()Know the safety boundaries
The skill runs scripts that create and save records to the configured storage, and it uses the network for public ontologies and remote storage. Treat it as a tool that writes data. Do not point it at a production instance with write access unless someone reviews the changes.
The skill states that registration does not establish scientific validity or FAIR compliance. Assay-specific checks must be added by hand.
- Run scripts only in a local development directory first
- Review lamin info output before any write
- Use workload identity or named environment variables for authenticated work, not pasted secrets
What people use it for
- Registering a sequencing or experiment table as a versioned artifact with a stored schema
- Curating AnnData metadata against controlled vocabularies from Bionty ontologies
- Tracking which script and inputs produced each output for a reproducible analysis
- Querying saved artifacts by key or schema before loading only the content you need
Frequently asked questions
How do I install the LaminDB agent skill?
Download the skill ZIP from the page where you found this guide. Unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps, upload the ZIP under Settings, then Capabilities, then Skills.
What does the LaminDB skill require?
It needs Python in the range the skill lists, plus the lamindb and bionty packages, IPython, pandas and AnnData, installed with uv or pip. Local SQLite needs no login, while public ontologies and remote storage need network access and private instances need credentials.
Is the LaminDB skill free to use?
The directory listing gives the licence as MIT, but the SKILL.md front matter in the source names Apache-2.0. Check the licence file in the source repository before reuse. The source does not mention pricing.
Does the skill check that my data is scientifically correct?
No. Validation checks structure, types and schema membership only. The skill says registration does not establish scientific validity or FAIR compliance, so add assay-specific checks yourself.
Why does validation fail and how do I fix it?
Validation raises an error when a table does not match its schema, such as a wrong type or an extra column when extras are forbidden. Repair the data or the schema, then validate again. Do not treat the call as a true or false test.
How is this different from general data cleaning tools?
This skill is built around biological metadata, typed schemas, Bionty ontology annotation and provenance. If you only need general-purpose cleaning with no biological metadata or lineage requirements, the listing says the setup overhead is high and another tool may suit you better.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Choose this skill if you run a bioinformatics or lab data workflow that needs reproducible, traceable curation, or if you are building a LaminDB lakehouse. Pick something else if you only need general data cleaning, expect automatic scientific QC, or plan to write to a production instance without review.
Download the skill