Free tools
GitHub Actions Workflow Checker
Paste a workflow file from .github/workflows/. The page reads its triggers, jobs and steps and flags what an attacker looks for first in a pipeline, each with the step it came from, what an attacker does with it and what to change.
No network. Nothing you enter leaves this tab. The workflow is parsed and checked in your browser, nothing is stored, and nothing about what you paste is counted or logged.
Results
What triggers the workflow, each finding and the step it came from appear here.
Reading the results
- pull_request_target, workflow_run and the pull request head
- Both events run in the base repository with its secrets and a token that can usually write, even for a pull request from a fork. That is safe while the job only reads the base branch. Checking out or fetching the pull request head and then building or testing it runs the attacker's code with that access, which is how most Actions takeovers start. issue_comment jobs that check out refs/pull/N/head are the same case.
- Script injection
- GitHub replaces ${{ }} before the shell or script starts, so an issue title, pull request body, branch name or commit message becomes part of the code. Pass the value through env and quote it in the shell ("$TITLE"), or read context.payload inside actions/github-script. Fields GitHub names as untrusted are rated as such; other fields that end in body, title, name, ref or message are rated one step lower.
- Pinning to a commit SHA
- A tag or branch can be moved to any commit by whoever controls the action repository, so the next run picks up their code. A full 40-character SHA cannot be moved. GitHub-owned actions/* and github/* are rated lower than third-party actions, because the mechanism is the same but the odds are not.
- Token permissions and OIDC
- Without a permissions key the GITHUB_TOKEN gets the repository or organization default, which this file cannot show. write-all gives every scope. id-token: write lets any step mint an OIDC token for the repository; grant it only on the job that logs in to a cloud or signs an artifact.
- Secrets
- GitHub masks the exact value of a secret in logs, not an encoded or partial copy. A secret in a job or workflow env reaches every step in the job, including third-party actions, and a secret passed to an action pinned by tag goes to whatever that tag points to on the next run.
- Self-hosted runners, GITHUB_ENV and persist-credentials
- A self-hosted runner on a public trigger lets outsiders run code on a machine in your network. A line written to GITHUB_ENV or GITHUB_PATH changes the environment of every later step, so outside text there can set LD_PRELOAD or BASH_ENV. actions/checkout keeps the job token on disk unless persist-credentials is false, where later steps can read it.
- Compromised actions
- Three published incidents are known to this page: tj-actions/changed-files (GHSA-mrrh-fwg8-r2c3, March 2025), reviewdog/action-setup and the actions that call it (GHSA-qmg3-hpqr-gqvc, March 2025), and aquasecurity/trivy-action and setup-trivy (GHSA-69fq-xp46-6x23, March 2026). Each finding gives the affected versions and the window to check in your run logs.
What this check cannot see
A workflow file is one input to what a pipeline can do. This page reads only the file you paste, so it cannot see:
- Repository and organization settings: the default GITHUB_TOKEN permissions, approval for fork pull requests, and whether Actions may create or approve pull requests.
- Whether the repository is public, and who can open a pull request, comment on an issue or start a manual run.
- Branch protection, rulesets and environment protection rules, such as required reviewers, that gate a job.
- What the actions it calls actually do: their code, their own dependencies, and the actions they call by tag, as the reviewdog compromise showed.
- Reusable workflows and composite actions this file calls, which run with its token and its secrets.
- Values that arrive through step outputs, artifacts, caches or files an earlier step wrote, which can carry attacker text this check does not trace.
- The cloud side of OIDC: the trust policy that decides which repository, branch or environment can assume a role.
- Which secrets exist, where they are scoped (repository, environment or organization) and what they unlock.
- What a self-hosted runner can reach on the network, whether it is ephemeral, and which repositories its runner group admits.
- Advisories after this page was written: the compromised-action check knows three incidents and is not a feed.
A clean result means none of the patterns this page knows appear in the workflow. It is not proof the workflow is safe: what it can really do depends on everything above, and on the repository settings and the actions it calls.