CVE-2026-23742: 
Skipper Ingress vulnerability analysis and mitigation

CVE-2026-23742 is a code injection vulnerability in Zalando Skipper, an HTTP router and reverse proxy widely used as a Kubernetes Ingress controller. The flaw exists in the Lua filter functionality, where the default configuration (-lua-sources=inline,file) prior to version 0.23.0 allowed untrusted users to create inline Lua scripts capable of reading the filesystem accessible to the Skipper process and extracting secrets from logs. All Skipper versions before 0.23.0 are affected. It was disclosed on January 16, 2026, and carries a CVSS v3.1 base score of 8.8 (High), assigned by GitHub, Inc. (GitHub Advisory, NVD).

The root cause is improper control of code generation (CWE-94) combined with execution with unnecessary privileges (CWE-250) and insufficiently protected credentials (CWE-522). By default, Skipper enabled the inline Lua source type, which permitted any user with the ability to create Lua filters — for example, via a Kubernetes Ingress resource annotation — to inject arbitrary Lua scripts that execute within the Skipper process context. These scripts can use Lua's file I/O capabilities to read any file accessible to the Skipper process (e.g., mounted Kubernetes secrets, environment variables, or configuration files) and expose the contents through logs or HTTP responses. The fix, introduced in commit 0b52894, adds a new --enable-lua flag (defaulting to false) that disables the Lua scripting engine entirely unless explicitly enabled by an operator (GitHub Commit, GitHub Advisory).

A low-privileged network attacker with the ability to create Lua filters (e.g., through Kubernetes Ingress resource creation rights) can execute arbitrary code with the privileges of the Skipper process, achieving high confidentiality, integrity, and availability impact. This enables reading sensitive files from the filesystem — including mounted Kubernetes secrets, TLS certificates, and API tokens — and extracting credentials that may appear in Skipper logs. Depending on the operational context and the privileges of the Skipper process, an attacker could also modify system state or facilitate lateral movement within the cluster (GitHub Advisory, Feedly).

There is no public proof-of-concept exploit code and no evidence of in-the-wild exploitation at this time. The EPSS score is approximately 0.041%, indicating a low probability of exploitation in the near term. The vulnerability is not listed in the CISA Known Exploited Vulnerabilities (KEV) catalog. Exploitation requires low privileges — specifically, the ability to create or modify Kubernetes Ingress resources or other Skipper filter configurations — making it most relevant in multi-tenant Kubernetes environments where such permissions are broadly granted. The vulnerability was responsibly reported by researchers identified as b0b0haha and moyushui (GitHub Advisory, Skipper Release).

1. Reconnaissance: Identify a Kubernetes cluster running Skipper as an Ingress controller with a version prior to 0.23.0. Confirm that the attacker has permissions to create or modify Kubernetes Ingress resources (e.g., via kubectl auth can-i create ingress).
2. Craft a malicious Lua filter: Write a Lua script that reads a sensitive file accessible to the Skipper process, such as a mounted Kubernetes secret or service account token. Example: local f = io.open('/var/run/secrets/kubernetes.io/serviceaccount/token', 'r'); local content = f:read('*a'); f:close(); ngx.log(ngx.ERR, content).
3. Inject via Kubernetes Ingress annotation: Create or annotate a Kubernetes Ingress resource with the malicious inline Lua filter using Skipper's filter annotation, e.g., zalando.org/skipper-filter: lua("<malicious_lua_script>").
4. Trigger execution: Send an HTTP request through the Skipper proxy that matches the annotated Ingress route, causing Skipper to execute the inline Lua script within its process context.
5. Exfiltrate secrets: Access Skipper's logs (e.g., via kubectl logs) to retrieve the sensitive file contents written by the Lua script, or observe them in HTTP response headers/bodies if the script is crafted to reflect data (GitHub Advisory, GitHub Commit).

• Logs: Skipper log entries containing unexpected file contents (e.g., Kubernetes service account tokens, certificate data, or environment variable values) logged at error or info level; log lines referencing file paths such as /var/run/secrets/, /etc/, or other sensitive mount points.
• Kubernetes: Ingress resources with zalando.org/skipper-filter annotations containing lua(...) filter definitions with inline scripts, especially those referencing io.open, os.execute, or similar Lua I/O functions; unexpected creation or modification of Ingress resources by low-privileged service accounts.
• Network: Unusual outbound connections from the Skipper pod to external IPs, potentially indicating data exfiltration via Lua HTTP calls.
• Process: Unexpected file access patterns by the Skipper process to sensitive paths (detectable via auditd or Falco rules monitoring file reads by the skipper binary) (GitHub Advisory).

Upgrade Skipper to version 0.23.0 or later, which disables the Lua scripting engine by default via the new --enable-lua=false default setting (Skipper Release). For operators who require Lua functionality, explicitly enable it with --enable-lua and restrict the source type to --lua-sources=file (rather than inline) so that only operators with filesystem access can provide Lua scripts. Additionally, apply Kubernetes RBAC controls to restrict Ingress resource creation and modification to trusted users only, and audit existing Ingress resources for suspicious inline Lua filter definitions (GitHub Advisory, GitHub Commit).

The vulnerability was covered by The Hacker Wire, which published an article specifically on the Skipper Lua filter LFI and secret disclosure issue (The Hacker Wire). The advisory was noted by INCIBE-CERT (Spain's national cybersecurity incident response team) and CCN-CERT, indicating recognition by European national security agencies. Social media activity was observed on Bluesky and Mastodon/Infosec.exchange, with automated CVE tracking accounts disseminating the advisory. Overall community reaction was measured, consistent with a vulnerability affecting a specialized infrastructure component with no active exploitation reported.