The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches code templates and documentation from the vendor's official GitHub repositories (goldsky-io/documentation-examples) and website (goldsky.com). It also provides instructions for installing the official CLI using the vendor's official shell-piped installer.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to execute goldsky compose CLI commands for project initialization, secret management, contract deployment, and application lifecycle management.
[INDIRECT_PROMPT_INJECTION]: The instructions include robust security awareness regarding untrusted data. The agent is explicitly directed to treat blockchain events and external API responses as untrusted, to decode them strictly using known ABIs, and to never interpret their content as executable instructions or plan changes.
[SAFE]: Secret management is handled according to security best practices. The skill avoids hardcoding credentials, directing users to use the platform's native secret storage via CLI commands, and warns the agent against capturing or logging sensitive values.
[SAFE]: Heuristic flags for destructive commands (e.g., rm -rf .compose/) were evaluated and found to be benign, referring to the intended deletion of local temporary build caches rather than system-wide or sensitive user data.

compose