How to use the LatchBio Workflow Integration agent skill

What an agent skill is and what this one does

An agent skill is a folder of instructions and supporting files that an AI agent loads when a task matches its description. Without a skill, the agent relies only on general knowledge, which may be out of date for a specific platform.

This skill, from K-Dense-AI/scientific-agent-skills, gives the agent working knowledge of Latch. Once installed, the agent can write Latch workflows in Python, package Nextflow or Snakemake pipelines, configure task compute resources, move data in Latch Data, read Latch Registry tables, and launch or monitor runs. It routes each task to one of nine reference files so it reads only the material it needs.

• Create and maintain Python SDK workflows and task graphs
• Package and register Python, Nextflow, or Snakemake pipelines
• Configure CPU, memory, storage, GPU, caching, retries and timeouts
• Work with Latch Data and Registry projects, tables and records
• Launch and monitor workflows through Python or Latch MCP

Install the skill from the ZIP file

Download the ZIP file for this skill from this page and unzip it. The skill folder is named latchbio-integration and contains SKILL.md, a references folder, and a scripts folder with scripts/inspect_latch_sdk.py.

For a personal install that works in every project, place the folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder instead.

If you use the Claude apps rather than Claude Code, upload the skill under Settings → Capabilities → Skills.

• Personal install: ~/.claude/skills/latchbio-integration/
• Project install: .claude/skills/latchbio-integration/ inside your project
• Claude apps: Settings → Capabilities → Skills, then upload the skill

Set up the Latch environment the skill expects

The skill targets Latch SDK 2.77.1 and recommends Python 3.12 with the uv package manager. Python 3.9 or newer is the minimum. Create a virtual environment and install the pinned SDK, then sign in with the supported OAuth flow.

Docker is only needed for local image builds. Remote registration is the default. On Windows the source recommends WSL for the Linux workflow tooling.

• Requires Python 3.9+ (3.12 recommended) and the uv package manager
• Requires a Latch account and workspace with OAuth login
• Requires network access
• Docker is optional, for local builds only

Trigger the skill with natural requests

You do not need to name the skill. The agent loads it when your request matches the skill description, which covers authoring or deploying workflows, configuring resources or interfaces, moving data, integrating Registry, and launching or monitoring runs. Be specific about the pipeline, the workspace and the outcome you want.

Good requests name the thing you are working on. For example, ask the agent to create a Latch workflow from an existing script, or to find out why a registration failed. The skill also covers Nextflow and Snakemake, so you can mention those tools directly.

• Write a Latch Python workflow that reverse-complements a DNA sequence and register it to my workspace
• Package my Nextflow pipeline for Latch using its nextflow_schema.json
• Why does my latch register fail during staging?
• Move my FASTQ folder into Latch Data and list what is there
• Read the samples table in my Latch Registry project and build a samplesheet

Follow the safe workflow: stage, develop, then register

The skill recommends checking compatibility, defining typed inputs, configuring resources, validating in the execution image, registering deliberately, and only then launching. Staging and development happen before a real registration, so you can catch image problems early.

Registration and launches change remote state. The skill asks the agent to confirm before launching paid compute, especially GPU or large batch runs, and before deleting data or overwriting shared destinations.

• Stage and test: latch register --staging . then latch develop .
• Register deliberately: latch register --yes --open .
• Launch only after reviewing cost and parameters

Connect Latch MCP for interactive operation

Latch also offers a remote Model Context Protocol server at https://mcp.latch.bio/mcp. It lets compatible AI clients discover workflows and interact with Latch Data and executions through an OAuth-authorized tool surface. MCP uses its own OAuth login, separate from the SDK login, and actions launched through it incur the same normal Latch charges as the Console or SDK.

• Cursor: add the server URL in Tools & MCP(s), select LatchBio, then Connect
• Claude Code: claude mcp add --transport http latchbio https://mcp.latch.bio/mcp

Safety notes: scripts and network access

This skill runs scripts and uses the network. The included scripts/inspect_latch_sdk.py performs local imports only. According to the source it does not authenticate or make network requests. The rest of the workflow, including registration, data movement and launches, does talk to the Latch service, so it needs your account and network access.

Never paste OAuth tokens or secrets into chat. The skill tells the agent not to read or copy ~/.latch/token manually, and never to log secrets or signed URLs. Check the agent's proposed commands before approving them, especially any that delete data or launch paid compute.

• The inspection script runs local imports only and needs no login
• Registration, data changes and launches affect your remote Latch workspace
• LatchBio MCP credentials cannot be reused for SDK access
• Review cost and parameters before any launch