The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/generate-canvas.mjs executes the kubectl binary to fetch resource data from the cluster. It also attempts to resolve the binary from ~/.agents/bin/kubectl, a path that should be verified for integrity in the environment.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes data from external sources, specifically Kubernetes cluster resource metadata, which is then summarized by the agent.
Ingestion points: Resource names, labels, and annotations fetched via kubectl and the project's .sealos/state.json file.
Boundary markers: The instructions lack explicit markers or warnings for the agent to treat fetched resource data as untrusted.
Capability inventory: The skill can execute shell commands (kubectl), perform file system writes (.sealos/canvas/), and host a local HTTP server (127.0.0.1).
Sanitization: The script specifically masks Secret values and ConfigMap data, and uses HTML/JSON escaping during template rendering to mitigate XSS and injection risks.

sealos-canvas