GHSA-2MW5-23GM-PCCQ
Vulnerability from github – Published: 2026-09-02 23:45 – Updated: 2026-09-02 23:45Summary
OpenChoreo Workflow Plane templates were vulnerable to OS command injection because some developer-controlled workflow parameters were interpolated directly into shell program text executed through sh -c.
An authenticated user with permission to configure and trigger an affected workflow could supply crafted parameter values containing shell metacharacters. Because Argo substituted these values directly into the shell script before execution, the values could alter the script and execute arbitrary commands inside the workflow pod.
In affected configurations, some build and publish templates used privileged Podman containers without Kubernetes pod user-namespace isolation. Consequently, injected commands could run as UID 0 inside a privileged container, with UID 0 mapped into the host user namespace. This substantially increased the potential impact of the command-injection vulnerability.
The remediation ensures that workflow parameters are treated as data rather than executable shell syntax. Affected values are now passed through container.env and consumed as quoted shell variables. JSON-derived build environment variables and build arguments are passed as separate argument-vector entries instead of being reconstructed as whitespace-delimited shell strings.
Impact
An authenticated user with permission to trigger an affected workflow could execute arbitrary commands inside the corresponding workflow pod.
Depending on the affected template and its configuration, an attacker could potentially: - Read source code checked out by the workflow. - Read Git credentials, registry credentials, or other secrets mounted into the pod. - Read the pod's Kubernetes service account token. - Access mounted volumes and other build-time data. - etc
Some affected build and publish templates ran Podman using:
securityContext:
privileged: true
and did not previously enable pod user namespaces.
Without hostUsers: false, UID 0 inside the container is UID 0 in the host user namespace. In combination with privileged mode, this grants the container broad Linux capabilities and removes several normal container isolation restrictions.
This does not mean that every successful command injection automatically results in a container escape or complete node compromise. However, it materially increases the risk. Depending on the container runtime, available devices, mounted paths, kernel configuration, and node environment, an attacker may be able to access host resources or escalate the compromise to the underlying node.
For example, access to a host block device such as /dev/vda1 is only possible when that device is exposed or otherwise visible inside the container. The presence and naming of such devices are environment-specific and should not be assumed across all deployments.
Patches
The fixes have been applied to the 1.0.4, 1.1.4, and 1.2.0-rc.2 releases. The fix prevents OS command injection by removing direct interpolation of developer-controlled Argo parameters into shell scripts executed through sh -c. Affected values are now passed through container environment variables and consumed as quoted shell variables, while JSON-derived build arguments and environment variables are passed as separate argument entries instead of being reconstructed as shell command strings. The patched workflow templates also enable pod user namespaces for containers that still require privileged Podman by setting hostUsers: false. This maps UID 0 inside the pod to a non-root UID on the host when user namespaces are supported by the cluster. Privileged execution has also been removed from workflow templates where it is no longer required.
Users of the sample workflow templates should apply the corrected templates from the corresponding release branch: https://github.com/openchoreo/openchoreo/tree/release-v1.x/samples/getting-started/workflow-templates. Replace release-v1.x with the relevant release branch, such as release-v1.0, release-v1.1, or release-v1.2, and apply the applicable workflow templates to the cluster.
Users with custom workflow templates should also review and update those templates, as upgrading OpenChoreo does not automatically correct independently maintained custom templates containing unsafe shell interpolation.
Workarounds
If you cannot upgrade immediately: - Restrict permission to create, modify, or trigger workflows to trusted users only. - Do not allow untrusted users to control parameters used by checkout, build, publish, or workload-generation templates.
{
"affected": [
{
"package": {
"ecosystem": "Go",
"name": "github.com/openchoreo/openchoreo"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"fixed": "1.0.4"
}
],
"type": "ECOSYSTEM"
}
]
},
{
"package": {
"ecosystem": "Go",
"name": "github.com/openchoreo/openchoreo"
},
"ranges": [
{
"events": [
{
"introduced": "1.1.0"
},
{
"fixed": "1.1.4"
}
],
"type": "ECOSYSTEM"
}
]
},
{
"package": {
"ecosystem": "Go",
"name": "github.com/openchoreo/openchoreo"
},
"ranges": [
{
"events": [
{
"introduced": "1.2.0-rc.1"
},
{
"fixed": "1.2.0-rc.2"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-73667"
],
"database_specific": {
"cwe_ids": [
"CWE-78"
],
"github_reviewed": true,
"github_reviewed_at": "2026-09-02T23:45:26Z",
"nvd_published_at": "2026-08-13T22:17:28Z",
"severity": "HIGH"
},
"details": "### Summary\nOpenChoreo Workflow Plane templates were vulnerable to OS command injection because some developer-controlled workflow parameters were interpolated directly into shell program text executed through sh -c.\n\nAn authenticated user with permission to configure and trigger an affected workflow could supply crafted parameter values containing shell metacharacters. Because Argo substituted these values directly into the shell script before execution, the values could alter the script and execute arbitrary commands inside the workflow pod.\n\nIn affected configurations, some build and publish templates used privileged Podman containers without Kubernetes pod user-namespace isolation. Consequently, injected commands could run as UID 0 inside a privileged container, with UID 0 mapped into the host user namespace. This substantially increased the potential impact of the command-injection vulnerability.\n\nThe remediation ensures that workflow parameters are treated as data rather than executable shell syntax. Affected values are now passed through container.env and consumed as quoted shell variables. JSON-derived build environment variables and build arguments are passed as separate argument-vector entries instead of being reconstructed as whitespace-delimited shell strings.\n\n### Impact\nAn authenticated user with permission to trigger an affected workflow could execute arbitrary commands inside the corresponding workflow pod.\n\nDepending on the affected template and its configuration, an attacker could potentially:\n- Read source code checked out by the workflow.\n- Read Git credentials, registry credentials, or other secrets mounted into the pod.\n- Read the pod\u0027s Kubernetes service account token.\n- Access mounted volumes and other build-time data.\n- etc\n\nSome affected build and publish templates ran Podman using:\n\n```\nsecurityContext:\n privileged: true\n```\n\nand did not previously enable pod user namespaces.\n\nWithout `hostUsers: false`, UID 0 inside the container is UID 0 in the host user namespace. In combination with privileged mode, this grants the container broad Linux capabilities and removes several normal container isolation restrictions.\n\nThis does not mean that every successful command injection automatically results in a container escape or complete node compromise. However, it materially increases the risk. Depending on the container runtime, available devices, mounted paths, kernel configuration, and node environment, an attacker may be able to access host resources or escalate the compromise to the underlying node.\n\nFor example, access to a host block device such as /dev/vda1 is only possible when that device is exposed or otherwise visible inside the container. The presence and naming of such devices are environment-specific and should not be assumed across all deployments.\n\n### Patches\nThe fixes have been applied to the 1.0.4, 1.1.4, and 1.2.0-rc.2 releases. The fix prevents OS command injection by removing direct interpolation of developer-controlled Argo parameters into shell scripts executed through `sh -c`. Affected values are now passed through container environment variables and consumed as quoted shell variables, while JSON-derived build arguments and environment variables are passed as separate argument entries instead of being reconstructed as shell command strings. The patched workflow templates also enable pod user namespaces for containers that still require privileged Podman by setting hostUsers: false. This maps UID 0 inside the pod to a non-root UID on the host when user namespaces are supported by the cluster. Privileged execution has also been removed from workflow templates where it is no longer required.\n\nUsers of the sample workflow templates should apply the corrected templates from the corresponding release branch: `https://github.com/openchoreo/openchoreo/tree/release-v1.x/samples/getting-started/workflow-templates`. Replace release-v1.x with the relevant release branch, such as release-v1.0, release-v1.1, or release-v1.2, and apply the applicable workflow templates to the cluster.\n\nUsers with custom workflow templates should also review and update those templates, as upgrading OpenChoreo does not automatically correct independently maintained custom templates containing unsafe shell interpolation.\n\n### Workarounds\nIf you cannot upgrade immediately:\n- Restrict permission to create, modify, or trigger workflows to trusted users only.\n- Do not allow untrusted users to control parameters used by checkout, build, publish, or workload-generation templates.",
"id": "GHSA-2mw5-23gm-pccq",
"modified": "2026-09-02T23:45:26Z",
"published": "2026-09-02T23:45:26Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/openchoreo/openchoreo/security/advisories/GHSA-2mw5-23gm-pccq"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-73667"
},
{
"type": "WEB",
"url": "https://github.com/openchoreo/openchoreo/pull/4193"
},
{
"type": "WEB",
"url": "https://github.com/openchoreo/openchoreo/pull/4243"
},
{
"type": "WEB",
"url": "https://github.com/openchoreo/openchoreo/pull/4277"
},
{
"type": "WEB",
"url": "https://github.com/openchoreo/openchoreo/pull/4297"
},
{
"type": "WEB",
"url": "https://github.com/openchoreo/openchoreo/commit/017c3c6d8b27c21d11c8c2b43da1846aa7ae73b9"
},
{
"type": "WEB",
"url": "https://github.com/openchoreo/openchoreo/commit/65c081ff74618714cb0c82b5d0e0fad2c2cbc46b"
},
{
"type": "WEB",
"url": "https://github.com/openchoreo/openchoreo/commit/b274127a4342e5433c5035384cef57478b5b65ed"
},
{
"type": "WEB",
"url": "https://github.com/openchoreo/openchoreo/commit/fb2b659b9884eb45fa4f02b9cc7e89718a0276b7"
},
{
"type": "PACKAGE",
"url": "https://github.com/openchoreo/openchoreo"
},
{
"type": "WEB",
"url": "https://github.com/openchoreo/openchoreo/releases/tag/v1.0.4"
},
{
"type": "WEB",
"url": "https://github.com/openchoreo/openchoreo/releases/tag/v1.1.4"
},
{
"type": "WEB",
"url": "https://github.com/openchoreo/openchoreo/releases/tag/v1.2.0-rc.2"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H",
"type": "CVSS_V3"
}
],
"summary": "OpenChoreo: Authenticated OS command injection via OpenChoreo Workflow Plane templates enables code execution in privileged pods"
}
Sightings
| Author | Source | Type | Date | Other |
|---|
Nomenclature
- Seen: The vulnerability was mentioned, discussed, or observed by the user.
- Confirmed: The vulnerability has been validated from an analyst's perspective.
- Published Proof of Concept: A public proof of concept is available for this vulnerability.
- Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
- Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
- Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
- Not confirmed: The user expressed doubt about the validity of the vulnerability.
- Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.
The approach is described in our paper Mapping CVEs to MITRE ATT&CK Techniques: A Curated Gold-Set Classifier and the Limits of LLM-Assisted Label Expansion.