The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The instructions specify bootstrapping external security tools such as trivy, nmap, and nuclei. This process typically involves downloading and executing installation scripts or binaries from remote community-maintained repositories at runtime.
[COMMAND_EXECUTION]: The skill utilizes powerful command-line interfaces (kubectl, aws CLI) to perform sensitive operations, including listing all cluster secrets (kubectl get secrets -A) and retrieving cloud identity information (aws sts get-caller-identity).
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill processes potentially untrusted data from external sources, including Kubernetes API responses, cloud metadata services (IMDS), and outputs from vulnerability scanners.
Boundary markers: The skill includes a 'Mandatory confirmation gate' that requires user validation for probe and exploit actions, which serves as a delimiter to reduce the risk of accidental obedience to embedded instructions.
Capability inventory: The agent has access to cluster-wide resources, cloud provider identities, and the ability to download/execute external tooling.
Sanitization: There is no evidence of sanitization or strict schema validation for the data ingested from the cluster or cloud APIs before it is incorporated into the agent's context.
[DYNAMIC_EXECUTION]: The 'bootstrap' commands for trivy and the nmap/nuclei ecosystem involve the dynamic acquisition and setup of executable environments based on the toolchain table in SKILL.md.

cloud-k8s