openchoreo-platform-engineer
Fail
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: HIGHCREDENTIALS_UNSAFEPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [CREDENTIALS_UNSAFE]: The file
resources/workflow-templates/generate-workload.yamlcontains a hardcoded OAuth client secret (openchoreo-workload-publisher-secret) and client ID. Although labeled as defaults for local k3d development, hardcoding credentials in template files is a high-severity security risk. - [PRIVILEGE_ESCALATION]: Several workflow templates, including
resources/workflow-templates/containerfile-build.yamlandresources/workflow-templates/generate-workload.yaml, requiresecurityContext: privileged: true. While often necessary for container-in-container build tools like Podman, this grants the container full access to the host node's resources and kernel. - [COMMAND_EXECUTION]: The skill instructions and recipes make extensive use of
kubectlandhelmcommands for managing Kubernetes resources and platform components, which involves high-level cluster permissions. - [EXTERNAL_DOWNLOADS]: The skill's workflow templates download container images from the GitHub Container Registry (
ghcr.io) and other public registries. Additionally,resources/workflow-templates/publish-image.yamlis configured to push images tottl.sh, an anonymous public ephemeral registry, which could result in unintended public exposure of application artifacts. - [INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection as it processes data from external Git repositories, specifically reading and parsing
workload.yamlfiles. Malicious content within a repository could potentially exploit the build pipeline or influence agent logic if not properly sanitized. - Ingestion points: Repository cloning in
checkout-source.yamland workload descriptor parsing ingenerate-workload.yaml. - Boundary markers: Not explicitly defined in the shell scripts.
- Capability inventory: Subprocess calls for
podman,kubectl,helm, andcurl; file-write operations to the shared workspace. - Sanitization: No explicit sanitization or validation of the repository content before processing.
Recommendations
- AI detected serious security threats
Audit Metadata