The Agent Skills Directory

[SAFE]: The skill functions as a template for environment configuration. It instructs the agent to assist the user in mapping specific model providers and slugs to defined roles such as 'code delegates' or 'explorers'.
[DATA_EXPOSURE]: The skill mentions interacting with local configuration files like opencode.jsonc and reading from ../poteto-mode/references/hosts/. These actions are consistent with the skill's stated purpose of host-level model configuration and do not involve unauthorized access to sensitive user data or secrets.
[COMMAND_EXECUTION]: The instructions describe using host-specific commands (e.g., /model in Claude Code) to apply settings. These are built-in platform commands used for their intended purpose.
[INDIRECT_PROMPT_INJECTION]: The skill involves processing user-provided model names. While this is an ingestion point for external data, the scope is limited to configuration strings and does not present an exploitable surface for prompt injection or context poisoning.

setup-models