CVE-2026-44358: Espressif Shared GitHub DangerJS: Untrusted Search Path in DangerJS Action Entrypoint

Synopsis

An untrusted search path vulnerability affects Espressif Shared GitHub DangerJS, a reusable GitHub Actions CI workflow used by Espressif GitHub projects. The flaw exists in the action's entrypoint.sh, which resolves the DangerJS binary and Node.js modules from a caller-controlled workspace after copying fork content into it. A crafted fork pull request submitted against a repository using a pull_request_target workflow can substitute attacker-supplied code for the action's own code and execute it inside the action container. No fix version has been published yet; HarborGuard tracks the upstream advisory and will make a patched rebuild available as soon as one is released.

HarborGuard Coverage

Exploit Conditions

• Network reachabilityRequired

  The attacker submits a fork pull request over the network targeting a repository that processes pull_request_target workflows, so network-level access to the target repository's hosting service is necessary.

• AuthenticationNot required

  No authentication is required; creating a public fork and opening a pull request does not require any privileged account on the target repository.

• Victim interactionRequired

  A maintainer or CI system must trigger the pull_request_target workflow against the attacker's fork, meaning the victim's CI pipeline must process the malicious pull request.

• Attack complexityDetail

  Attack complexity is low; no race condition or special environmental state is needed, and the path substitution happens reliably whenever the workflow processes a fork checkout.

Blast Radius

• Attacker-controlled code executes inside the GitHub Actions runner container with the permissions granted to the workflow, which may include access to repository secrets.
• Secrets such as tokens, API keys, and credentials exposed to the workflow environment can be read and exfiltrated by the injected code.
• The attacker can modify build artifacts, tamper with release pipelines, or inject malicious content into outputs produced by the CI job.