The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill references authentication via environment variables (CC_ACCESS_KEY_ID, CC_ACCESS_KEY_SECRET, CLOUDCANNON_API_KEY) and a local auth.json file. It contains no hardcoded credentials and explicitly instructs the agent not to read, print, or exfiltrate these secrets. It also recommends against the use of .env files for local development to keep credentials outside the working directory.
[EXTERNAL_DOWNLOADS]: The skill utilizes the @cloudcannon/cli package, which is downloaded via npm or executed through npx. This is a legitimate vendor resource necessary for the skill's functionality and traces back to the vendor's own namespace.
[COMMAND_EXECUTION]: The skill frequently executes CLI commands to interact with CloudCannon services. It includes a high-priority safety rule requiring explicit user confirmation before executing any command that modifies remote sites, updates build configurations, or commits changes to a git repository.
[DYNAMIC_EXECUTION]: The skill provides scripts that use node -e to parse a local JSON file (documentation.json) shipped with the CLI package. This is used to dynamically discover command flags and ensure accuracy with the installed version, minimizing risk by referencing a local, vendor-provided file.
[INDIRECT_PROMPT_INJECTION]: The skill accesses untrusted data from the CloudCannon API (files, logs, and form submissions) which could theoretically contain malicious instructions.
Ingestion points: Data enters the agent's context through commands like sites files get, orgs list, and builds print-logs.
Boundary markers: The instructions mandate user confirmation for all writing actions, providing a manual checkpoint to review generated changes.
Capability inventory: The agent can perform network requests (via the CLI), commit to git repositories, and modify build configurations.
Sanitization: There is no automated sanitization of the retrieved data, but the agent is explicitly warned to review pending changes using list-edits before committing them to a live repository.

cloudcannon-cli