The Agent Skills Directory

[SAFE]: The skill facilitates a collaborative and guided documentation process with explicit human-in-the-loop approval gates at every stage. No malicious patterns, obfuscation, or safety bypasses were detected.\n- [COMMAND_EXECUTION]: While the Bash tool is authorized in the YAML frontmatter, the instructions only specify its use for invoking project-specific sub-commands (e.g., /architecture-decision, /gate-check) which are typical for this development environment. There are no patterns of arbitrary command execution or privilege escalation.\n- [DATA_EXFILTRATION]: The skill operates entirely on local project data. It reads from and writes to specified project directories (docs/, design/, production/) and lacks any network capabilities (curl, wget, etc.) or exfiltration patterns.\n- [PROMPT_INJECTION]: The skill uses instructional language consistent with its intended purpose as a technical director agent. It does not attempt to override system prompts or bypass LLM safety filters.

create-architecture