agentforce-architecture-analyze

This module is not overtly malicious (no direct network access, persistence, or hidden payload logic in the shown fragment). However, it creates a significant supply-chain/operational risk by executing an external CLI with arguments driven by external YAML recipes and runtime parameters, and by explicitly enabling secret disclosure behavior in the subprocess environment (`SF_TEMP_SHOW_SECRETS=true`). If the recipe directory/files are not strictly trusted and immutable, an attacker who can tamper with `CLI_DIR/<name>.yaml` can influence the invoked CLI’s behavior. Additionally, the effectiveness of redaction is dependent on helper functions not shown, and stdout is parsed as JSON without any redaction step, so sensitive output could still propagate into returned data.