Events and triggers

[sudoUgando]

What's the difference between pull_request and pull_request_target triggers, and why might using the wrong one be a security issue?

[EimanTahir027]

What's the difference between pull_request and pull_request_target triggers, and why might using the wrong one be a security issue?

pull_request vs pull_request_target
Both trigger workflows on pull request events, but they differ in what context the workflow runs under, which determines what permissions and secrets are available.

pull_request
The workflow runs in the merge commit of the PR's branch, using the workflow from the PR's base branch. It has read-only token permissions by default and cannot access secrets from the target repository.

Runs code from the fork/feature branch
Uses the workflow definition from the base branch (for security)
Token has read-only permissions
No access to repository secrets
pull_request_target
The workflow runs in the context of the base repository (the target branch), using the workflow from the base branch. This means it has full write permissions and access to secrets.

Runs code from the base branch (but the workflow file itself is from the base branch too)
Runs in the context of the base repository, not the fork
Token has write permissions by default
Full access to repository secrets
Why the wrong one is a security issue
The danger comes from pull_request_target being used carelessly. Here's how an attacker can exploit it:

A repository uses pull_request_target in a workflow that does something like:

yaml
on: pull_request_target
jobs:
build:
steps:
- uses: actions/checkout@v4
with:
ref: ${{ github.event.pull_request.head.sha }}
- run: npm install && npm test
An attacker opens a PR from a fork. The actions/checkout step checks out their fork's code.

The workflow runs in the base repository's context with access to secrets, but the attacker controls the code that gets executed (e.g., a malicious package.json or postinstall script)..

The root issue: pull_request_target gives the workflow privileged access (write token, secrets), but if you then check out and execute the PR's code, you're running untrusted code with trusted credentials.

Safe usage of pull_request_target: Only use it for workflows that do not check out and execute the PR's code — for example, adding labels, posting comments, or running checks that only look at PR metadata