Skip to main content

Actions Warden

Read-only auditor for risky or injected GitHub Actions workflow config.

Current release: 0.1.6

After the 2026 wave of repository-theft attacks, a common post-compromise move is: steal a token, then inject or tamper with a repo's .github/workflows/ so CI exfiltrates secrets or runs attacker code. Actions Warden scans those workflow files for the patterns that enable it.

It does not execute workflows, contact GitHub, modify files, or prove a pipeline is safe. No telemetry, no data storage, no automatic remediation — alerts only.

Part of the Dragon Lady open security kit used alongside package/host scanners during the August 2026 npm/keyv (ChainDrop / Shai-Hulud) emergency tooling updates: when tokens may have been stolen, also audit CI workflows.

Install

pipx install actions-warden
# or
pip install actions-warden
# upgrade:
pipx upgrade actions-warden
# or: pip install -U actions-warden

Python 3.9+. No runtime dependencies.

Usage

actions-warden --version
actions-warden /path/to/repo
actions-warden /path/to/repo --json
actions-warden /path/to/repo --report report.json

It scans .github/workflows/*.yml|*.yaml, composite action.yml|action.yaml, package.json scripts, and .sh files. You can also point it at one of those files. Dependency, build, and Git object directories remain outside the scan.

Version 0.1.6 adds local checks for wfx-loot, including the npm-script delivery path that can run without a workflow change. See the campaign response guide for history checks and response limits.

Exit codes:

  • 0: no blocking workflow risks found
  • 1: usage/runtime error or incomplete scan
  • 2: blocking workflow risks found (suitable as a CI gate)

Missing or unsupported targets fail explicitly. Unreadable files/directories, invalid text, and candidate size limits set coverage.complete: false with path/reason entries in coverage.errors. Blocking findings still return 2; otherwise incomplete scans return 1, never 0. Completeness covers only the scanner's selected file types and directory policy, not every repository file.

--report creates a new report exclusively (mode 0600 on POSIX). Existing files, symlinks and hard links are refused, including aliases of scanned inputs. Write failures return 1; a partial newly created report may remain and must not be treated as a successful result. Choose a fresh output path for each run.

What It Flags

Rule Severity What it catches
secret-exfiltration critical a secret reference alongside an outbound network command
bulk-secret-context high serialization of the whole secrets context, including toJSON(secrets)
wfx-payload-markers critical combined payload configuration/staging markers in workflows or shell scripts, including configurations with no callback address
wfx-callback-indicator high a reported campaign address alongside a callback or network command
wfx-npm-hidden-helper high an npm script invokes a hidden .ci-xxxxxx.sh helper matching the reported filename pattern
untrusted-input-injection high attacker-controllable github.event.* / head_ref interpolated into the workflow (shell injection in run steps)
remote-code-in-run high a downloaded script piped straight into a shell
pull-request-target-head-checkout high pull_request_target running with secrets while checking out PR-controlled code ("pwn request")
checkout-unsafe-pr-opt-out high actions/checkout explicitly setting allow-unsafe-pr-checkout on privileged PR-adjacent triggers
self-hosted-on-untrusted medium self-hosted runner reachable by external pull requests
permissions-write-all medium write-all token permissions
oidc-with-write medium OIDC id-token: write combined with contents: write
deployment-trigger-review medium workflows triggered by deployment / deployment_status that need GitHub workflow execution protections reviewed
deployment-trigger-privileged high deployment-triggered workflows that also reference secrets, OIDC, or write permissions
unpinned-action low third-party action pinned to a mutable tag/branch instead of a commit SHA
hijacked-actions-cool-tag critical actions-cool/issues-helper or actions-cool/maintain-one-comment on a tag, branch, or the known hijacked commit. Stop using the tag. Do not revoke tokens until the monitor and wiper are understood
memtensor-sckit-stage0 critical workflow text contains both sckit and stage0. Treat the runner and publish tokens as exposed. Do not run the step. Notify-only

The rules are conservative. A finding means "review this workflow," not "this repo is compromised."

GitHub announced safer actions/checkout@v7 defaults for common pull_request_target pwn-request patterns on 2026-06-18, with supported major tag backports planned for 2026-07-16. Workflows pinned to a minor, patch, or full SHA need an explicit upgrade to receive that behavior, and any allow-unsafe-pr-checkout opt-out should be treated as a deliberate high-risk review item.

GitHub's workflow execution protections public preview also lets organizations restrict which actors and event types can trigger workflows. Treat deployment / deployment_status workflows as review items, and treat deployment-triggered workflows that use secrets, OIDC, or write permissions as blocked until the repository or organization restricts that trigger.

Why text-based, not YAML-parsed

A hostile workflow can be written to parse in surprising ways. Actions Warden inspects what is actually on disk rather than a parser's normalized view, and stays dependency-free. The tradeoff is coarser context: some file_all rules flag co-occurrence within a file rather than within a single job.

Scope Limits

This is a narrow CI/CD source scanner. It does not scan installed dependencies or packages (see a dependency/supply-chain scanner for that), does not resolve reusable or remote workflows, and will not catch every possible injection or obfuscated payload.

The new shell and npm checks are structural heuristics, not shell execution or a full shell parser. A filename match needs review; it does not establish theft. The scanner does not fetch Git history or query GitHub, cloud accounts, or past network traffic. A clean current tree cannot clear a reverted attack. Skipped candidate symlinks, malformed package metadata, and unreadable/oversized candidates produce incomplete coverage rather than a successful clean result.

Metadata

Release files for actions-warden 0.1.6

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for actions-warden 0.1.6
File Size Uploaded Provenance Actions
actions_warden-0.1.6.tar.gz 25.7 kB With provenance Details for actions_warden-0.1.6.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for actions-warden 0.1.6
File Interpreter ABI Platform Actions
actions_warden-0.1.6-py3-none-any.whl Python 3 none any Details for actions_warden-0.1.6-py3-none-any.whl

Total release size: 42.0 kB

Release files / actions_warden-0.1.6.tar.gz

Download URL actions_warden-0.1.6.tar.gz
Size 25.7 kB
Tags Source
SHA-256 checksum
How to use checksums
bdd514f526fcb49dbe67f9003a776c90e6a908f0d1fab3e4c412fe805b52b227
BLAKE2b-256 checksum
How to use checksums
450afbf05012a6470f2239c172f412b73a4d077dde0fbed4ac0da58f1b3aef98
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 14, 2026.

Transparency log

Release files / actions_warden-0.1.6-py3-none-any.whl

Download URL actions_warden-0.1.6-py3-none-any.whl
Size 16.3 kB
Tags Python 3
SHA-256 checksum
How to use checksums
31532a05302d43ba517687e0868272e26e81a47a21695d1b27ecdb8495784518
BLAKE2b-256 checksum
How to use checksums
18dad0b2d95b91cd889aabed3ba1fa3b4b60e1e683b4645edd56199deec6c110
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 14, 2026.

Transparency log

Security for actions-warden 0.1.6

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for actions_warden-0.1.6.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 14, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page