The Agent Skills Directory

[SAFE]: The skill operates entirely on local project documentation and does not perform any network requests or external data transfers. The use of local files for context is consistent with its stated purpose of architecture planning.- [COMMAND_EXECUTION]: While the skill has access to the Bash tool, its instructions are limited to project-specific tasks like updating session state files and does not involve executing untrusted or remote scripts.- [DATA_EXFILTRATION]: No evidence of credential harvesting or data exfiltration. The skill processes project files (such as GDDs and engine references) to create internal technical documents without sending data to external domains.- [PROMPT_INJECTION]: The instructions do not contain patterns for bypassing safety filters, overriding agent behavior, or extracting system prompts. It uses structured user interaction through the AskUserQuestion tool for all critical decisions.- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from project files (GDDs) to extract requirements. While this represents a theoretical attack surface, the risk is mitigated by the skill's structured processing phases and mandatory human-in-the-loop approval gates.

create-architecture