> ## Documentation Index
> Fetch the complete documentation index at: https://docs.galagos.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Custom Skills: write your own

> Capture your lab's preferred protocols as reusable Skills in plain English — from chat. The agent saves the text, applies it automatically, and you can promote it to your personal library.

Custom Skills let you encode your lab's preferred analysis approaches in plain English. Instead of repeating the same tool choices, parameter thresholds, and interpretation rules in every prompt, you write them once as a Skill — and the agent applies them automatically every time the Skill is enabled.

A Skill is not code. It is a short, written brief. *"Whenever I upload gene expression data, log-transform counts and filter genes with zero reads in more than half the samples"* is a complete, working Skill.

## What a Skill can encode

A Skill can describe any combination of:

* **Tool preferences** — *"use STAR for alignment, not HISAT2"*.
* **Parameter settings** — *"DESeq2 with FDR 5%, |log2FC| > 1, and shrinkage on"*.
* **Pipeline steps** — *"align, quantify with featureCounts, then run DESeq2"*.
* **Output formatting** — *"export supplementary tables as TSV with gene symbols, not Ensembl IDs"*.
* **Interpretation guidance** — *"flag genes in the top 100 hits that overlap our internal cancer panel"*.

You can mix any of these in a single Skill. The agent reads the brief and decides how to apply it.

## When to write a custom Skill

Custom Skills earn their keep when a workflow is repeatable. Common cases:

> [accordion] Lab-standard RNA-seq pipeline
>
> *"For every bulk RNA-seq dataset, align with STAR, quantify with featureCounts, and run DESeq2 with shrinkage enabled and an FDR threshold of 5%. Always produce a volcano plot with the top 20 genes labelled."*

> [accordion] In-house variant filtering
>
> *"For variant calling, always use GATK HaplotypeCaller followed by the lab's hard filters: QD \< 2.0, FS > 60.0, MQ \< 40.0. Annotate with SnpEff and exclude variants in repeat regions."*

> [accordion] Consistent single-cell preprocessing
>
> *"For Scanpy workflows, exclude cells with fewer than 200 genes or more than 5% mitochondrial reads. Use Leiden resolution 0.5 for initial clustering. Always produce a UMAP coloured by cell type."*

> [accordion] Reading papers consistently
>
> *"When I share a PDF paper, extract the methods, the figures, and the supplementary tables. I do not need the introduction or the discussion unless I ask."*

## Writing a Skill from chat

The fastest path is to ask the agent to write one for you. It already has the context.

**1. Open the Skills panel**

Click the wrench icon in the project's right sidebar.

**2. Click Create**

The Skill creator opens as a chat prompt.

**3. Describe what you want in plain English**

Tell the agent what the Skill should do, when it should apply, and any specific parameters or output preferences. One paragraph is usually enough.

**4. Iterate**

The agent drafts the Skill. Read it. Ask for tweaks — *"add a sentence about the FDR threshold"*, *"don't mention specific gene symbols, keep it general"* — until the brief matches what you want.

**5. Save**

Click **Save**. The Skill is attached to the current project and toggled on. The next prompt will use it.

> [!TIP]
>
> Save Skills *after* you have run the analysis you want to capture, not before. The agent writes a better Skill when it knows what worked.

## Promoting to your personal library

A Skill you write inside a project is **project-local** by default — it lives in that project's Skills panel and isn't visible elsewhere. To make it reusable across every project you create, click **Save to My Skills** on the expanded Skill description. It will appear in your personal library and pre-attach when you start similar projects.

## Sharing Skills with collaborators

You can export a Skill as a small file and import it into another workspace, or share it with a collaborator. The text is plain — no opaque binary format, no hidden state — so reviewers can read what they're being asked to attach.

> [!NOTE]
>
> Shared team Skill libraries are on the way. Workspace-level pinned and required Skills already let owners and admins curate a baseline for the whole team — see [Workspaces](https://docs.galagos.ai/concepts/workspaces.md).

## A few tips

* **Brief like a new lab member, not a computer.** Tell the Skill *why* and *when*, not just *what*. *"Use limma-voom because we always have low replicate counts"* is more useful than *"use limma-voom"*.
* **One paragraph beats a checklist.** Skills are read by an agent that handles judgement — over-specifying produces brittle output.
* **Test on a real project.** Run a familiar analysis with the Skill enabled and read the Steps list. If the agent isn't doing what you intended, refine the brief.
* **Update when your defaults change.** A Skill is not a contract; edit it as the lab's practice evolves.

## Related pages

> [card] [Skills overview](https://docs.galagos.ai/skills/overview.md)
>
> How the Skills panel works inside a project.

> [card] [Workspaces](https://docs.galagos.ai/concepts/workspaces.md)
>
> Pinned and required Skills at the workspace level.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.