The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation section in SKILL.md includes a command curl -s https://fluxcd.io/install.sh | sudo bash. This pattern involves fetching a script from a remote URL and executing it directly in the shell. While the source domain fluxcd.io is the official site for the Flux CD project, this method of installation bypasses local package management and integrity checks.
[PRIVILEGE_ESCALATION]: The installation instructions for the Flux CLI suggest using sudo to execute the installation script, which runs the remote code with administrative privileges on the host system.
[INDIRECT_PROMPT_INJECTION]: The skill documents how to configure Flux to automatically ingest and deploy resources from external, potentially untrusted sources such as Git repositories, Helm repositories, and OCI registries. This creates a vulnerability surface where malicious instructions could be embedded in the data the skill processes.
Ingestion points: Sources are defined via flux create source commands targeting Git, Helm, OCI, or S3-compatible buckets, as well as flux push artifact for OCI registries.
Boundary markers: The skill does not define or suggest the use of specific boundary markers or "ignore instructions" blocks for the data pulled from these external sources.
Capability inventory: The skill enables the agent to execute cluster management commands via the flux CLI, create secrets, and modify deployment states across the cluster.
Sanitization: The skill relies on the underlying Flux controllers and user-configured security (like SOPS for secret decryption) rather than providing explicit sanitization or validation steps for the ingested manifests.
[COMMAND_EXECUTION]: The skill provides extensive examples of shell commands for the flux CLI, which are intended to be executed by the agent to manage Kubernetes resources. This includes operations that can significantly alter the state of a cluster.

flux-cli