diff --git a/.github/workflows/_ci.yaml b/.github/workflows/_ci.yaml
index e9427fc85..a75d27944 100644
--- a/.github/workflows/_ci.yaml
+++ b/.github/workflows/_ci.yaml
@@ -420,6 +420,7 @@ jobs:
         name: nsys-jax-unit-test-A100
     - name: Extract archives and execute install scripts
       run: |
+        pip install virtualenv # for install.sh
         for zip in $(ls *.zip); do
           ZIP="${PWD}/${zip}"
           pushd $(mktemp -d)
diff --git a/docs/nsys-jax.md b/docs/nsys-jax.md
index 4a66fa5ad..f26911d13 100644
--- a/docs/nsys-jax.md
+++ b/docs/nsys-jax.md
@@ -22,7 +22,7 @@ There are three convenient ways of running profile data analyses:
 - Manual execution; the output files of both `nsys-jax` and `nsys-jax-combine` include an installation script that sets
   up a local Python virtual environment including the `nsys_jax` library, Jupyter Lab, and an example notebook for
   Jupyter-based exploration of profile data (*e.g.* run this on your laptop and explore your profile data in a Jupyter
-  notebook)
+  notebook). This installation script depends on the `virtualenv` command.
 
 ## Installation
 The containers published from this repository (`ghcr.io/nvidia/jax:XXX`) have `nsys-jax` pre-installed; the recipes for