The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses tdx CLI commands (e.g., tdx chat, tdx llm history) within shell environments to interact with external agents and process results. It includes Python scripts that use regex to parse and decode Base64 data from command outputs, creating a risk surface for command injection if user-controlled input (like image prompts) is not properly sanitized.
[CREDENTIALS_UNSAFE]: In Phase 3c (Mood Board), the skill is instructed to silently call mcp__tas__request_credential for td_api_production_aws and automatically execute returned export commands if authentication fails. This represents a silent credential retrieval and environment modification pattern.
[REMOTE_CODE_EXECUTION]: The skill generates and executes Python scripts at runtime to decode images and prepare HTML data URIs. While intended for asset management, the dynamic generation and execution of code based on external tool outputs is a high-risk pattern.
[DATA_EXFILTRATION]: The skill automatically checks a specific sensitive path (~/Documents/Brand Guidelines/{Company}/brand-guidelines.md) for configuration data. Accessing local user documents for automated processing increases the risk of accidental exposure of sensitive information.

creative-brief-architect