Skip to content
Get packs

Guide · Agent skill

How to use the Genomic Intelligence DNA Predictions agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

The genomic-intelligence skill lets an AI agent send a gene symbol, a genomic region, or a DNA/FASTA sequence to Genomic Intelligence's hosted DNA language models and return structured predictions. This guide explains what an agent skill is, how to install this one, how to ask for predictions in plain language, and what to check before you rely on the output. No GPU or model weights are needed on your machine.

Genomic Intelligence DNA PredictionsK-Dense-AI/scientific-agent-skills · MIT

Quick start

  1. 1Download and unzip. Download the skill ZIP from the page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder. Or upload it under Settings, Capabilities, Skills in the Claude apps.
  2. 2Check the requirements. Confirm you have Python 3.10 or newer with the requests library for REST examples, and network access to the Genomic Intelligence hosts.
  3. 3Choose MCP or REST. Use the keyless MCP server if your host supports MCP. Use the REST API if you need the raw envelope or a script, which requires a GI_API_KEY.
  4. 4Set the key if needed. Export GI_API_KEY in your environment for REST calls. Do not hardcode it or commit it to a repository.
  5. 5Ask with a sequence and a task. Give the agent a gene symbol, a region, or a DNA/FASTA sequence and name the prediction you want. Check the length against the task's limit.
  6. 6Check the context window. Compare your input length with the model's context window before you read the output, especially for enhancer predictions on short sequences.

What an agent skill is and what this one does

An agent skill is a folder of instructions and optional scripts that an AI agent loads when a request matches the skill's description. This skill's description covers DNA-based predictions, so the agent uses it when you mention a gene, a genomic region, or a DNA sequence and ask for a prediction.

Once loaded, the agent can request six kinds of prediction: promoter regions, splice donor and acceptor sites, enhancer activity, chromatin state, expression as log(TPM+1), and de-novo gene annotation. It can also run a composite workflow that finds the genes in a region and predicts each one's expression.

  • Promoter regions in a genomic region (promoter)
  • Splice donor and acceptor sites (splice)
  • Enhancer activity, developmental and housekeeping scores (enhancer)
  • Chromatin state across hundreds of tracks (chromatin)
  • Expression as log(TPM+1) from a sequence and cell-type context (expression)
  • De-novo transcript and gene annotation (annotation)

Install the skill from a ZIP file

Download the ZIP file for the skill from the page where you found this guide. Unzip it so that the skill folder, which contains SKILL.md and a references folder, sits inside your skills directory. For a personal install, use ~/.claude/skills/. For a single project, use .claude/skills/ inside the project folder.

If you use the Claude apps instead of a folder on disk, open Settings, go to Capabilities, then Skills, and upload the skill there.

  • Personal install: unzip into ~/.claude/skills/
  • Project install: unzip into .claude/skills/ in your project
  • Claude apps: Settings → Capabilities → Skills, then upload the skill

Set up access: the keyless MCP demo or a REST key

There are two ways the agent can reach Genomic Intelligence. The hosted MCP server at https://mcp.genomicintelligence.ai/mcp works without a key against a public demo tier that is rate- and concurrency-limited, so you can try it with nothing set. Use it when your agent host supports MCP.

The REST API at https://api.genomicintelligence.ai/v1 requires a key. Prediction and job requests send the key as a bearer token. Request a prediction key at contact@genomicintelligence.ai. Store the key in an environment variable and never paste it into a file you commit.

  • MCP demo: no key needed, subject to rate and concurrency limits
  • REST API: requires GI_API_KEY, sent as Authorization: Bearer <key>
  • Keys are scoped to a partner tier; a 429 response means you hit a cap, so back off and retry
export GI_API_KEY="gi_yourkeyhere" export GI_BASE_URL="https://api.genomicintelligence.ai"

Ask for a prediction in plain language

You do not need to name the skill or know the API. Describe the sequence and the prediction you want, and include the gene symbol, region, or sequence. The skill's trigger words include Genomic Intelligence, genomicintelligence.ai, promoter prediction, splice site prediction, enhancer activity, chromatin state, and sequence to expression.

Name the task you want so the agent picks the right model. If you only want to know which genes lie in a region, say so; if you want expression for each gene, ask for the composite find-genes-then-predict-expression workflow.

  • "Predict promoter regions in chr17:7,661,000-7,688,000 (GRCh38)."
  • "Find splice donor and acceptor sites in this FASTA sequence: ..."
  • "Predict expression as log TPM for the BRCA1 gene in breast epithelial cells."
  • "Annotate the transcripts in this 20 kb region of DNA."

Know the six tasks and their length limits

Each task has its own endpoint, its own minimum length, and its own options. The skill's table gives the accepted length for each task. Expression is the most constrained: it needs at least 9,198 bp and a tss_index unless the sequence is exactly 9,198 bp, plus a cell-type description.

Input shorter than a model's context window is still scored, but the window is padded. For enhancer, the floor is 50 bp while the context window is 249 bp, so short inputs are mostly padding. Compare your length with context_window_bp before you trust the result. Longer inputs are scanned in windows.

  • promoter: 300 to 500,000 bp
  • splice: 100 to 500,000 bp; strand-specific, so give the transcript orientation
  • enhancer: 50 to 500,000 bp
  • chromatin: 200 to 500,000 bp
  • expression: 9,198 to 500,000 bp
  • annotation: 1,000 to 500,000 bp; runs asynchronously

Run a REST prediction or an async job

For REST, each task has a literal endpoint such as /v1/tasks/promoter/predict. The request body contains sequence, plus optional sequence_name, model, and options fields. The response is a data and meta envelope. Omit the Prefer header for a synchronous 200 response. Send Prefer: respond-async to receive a 202 and poll GET /v1/tasks/jobs/{job_id}.

Synchronous JSON requests above the published sync limit return 413 sync_too_large. The limits are 200,000 bp for annotation and 50,000 bp for the composite workflow. Read x-sync-limit-bp in the OpenAPI document rather than relying on memorised numbers.

  • Endpoint pattern: POST /v1/tasks/{task}/predict
  • Async: add Prefer: respond-async, then poll /v1/tasks/jobs/{job_id}
  • OpenAPI contract: https://api.genomicintelligence.ai/v1/openapi.json

Safety, privacy and research-only use

The skill sends sequences to a third-party hosted service over the network. Do not use it for sensitive or confidential sequence data unless your data policy allows that. The skill also runs scripts and makes network calls, so review the files before you install them in a shared environment.

The skill is for research and development. It is not for clinical or diagnostic decisions. It does not do local alignment, variant calling, or file handling; for those tasks use a local tool such as BioPython or bcftools.

  • Requires Python 3.10+ with the requests library for the REST examples
  • Needs network access to api.genomicintelligence.ai and mcp.genomicintelligence.ai
  • Never hardcode GI_API_KEY; read it from the environment or a .env file
  • Use local-only tools for sensitive data

What people use it for

  • Scan a candidate promoter region for regulatory features before designing a wet-lab experiment.
  • Get model-based splice donor and acceptor predictions for a transcript sequence you already have.
  • Compare predicted enhancer activity across a set of short DNA sequences during early research.
  • Find the genes in a region and estimate their expression as log TPM for a chosen cell type.
  • Build a pipeline that calls the hosted REST API and handles async jobs and 429 retries.

Frequently asked questions

Is the Genomic Intelligence skill free to use?

The MCP demo tier works without a key but is rate- and concurrency-limited. The REST API needs a GI_API_KEY, which you request from contact@genomicintelligence.ai. This guide does not state pricing or licence terms beyond the skill's MIT licence, so check the official docs for commercial use.

Do I need a GPU or model weights to run it?

No. Predictions run on Genomic Intelligence's hosted service. Your machine only sends API requests, so you need Python with requests for REST examples and network access.

How do I install the genomic-intelligence skill?

Download the ZIP from the page and unzip it into ~/.claude/skills/ for a personal install, or into .claude/skills/ in a project. In the Claude apps, upload it under Settings, Capabilities, Skills.

Why does my REST request return 429?

A 429 means you hit a rate or concurrency cap for your key's tier. Back off, retry later, or ask Genomic Intelligence to raise your tier.

Can this skill call variants or align sequences?

No. It is for model inference from sequence. Use BioPython or bcftools for alignment, variant calling, and file handling.

Why is my enhancer prediction on a short sequence unreliable?

Enhancer inputs can be as short as 50 bp, but the model's context window is 249 bp. Shorter inputs are padded, so most of the score reflects padding rather than real sequence.

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

Use this skill if you are a researcher, student, or developer who wants quick, model-based regulatory predictions from DNA without installing heavy tools, and you can send your sequences to a hosted service. Pick a local tool such as BioPython or bcftools for alignment, variant calling, or any sensitive data, and never use the output for clinical or diagnostic decisions.

Download the skill