The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local shell commands (grep, cat, sed, node -p) to read documentation and type definitions from the installed @cloudcannon/sdk package. These operations are limited to the local filesystem and target vendor-provided resources to ensure the agent uses the correct method signatures for the installed version.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates reading content from external CloudCannon sites through methods such as site.getFile(path). This creates a surface for untrusted data ingestion. The skill mitigates this risk by requiring human confirmation for all mutating operations (site.delete, site.copy, commit) across all files, and by explicitly defining the 'rule that costs the most' as a mandatory check for the agent. Capabilities include file writes, network operations, and site management, which are sanitized via the SDK's internal REST client logic.

cloudcannon-sdk