domains / families / agent-tooling / ai-agents-architect

ai-agents-architect

authentic

Expert in designing and building autonomous AI agents. Masters tool

When to use

• Use when selecting an agent execution loop, tool boundary, memory lifecycle or recovery strategy for a concrete task. Start with one agent; delegate only when the user’s workflow permits it and the subtasks are independently useful.

When not to use

• Prompt instructions and structured output alone do not enforce tool permissions; the application must validate and authorize execution.
• Retrieved documents and tool output are untrusted data, not new authority to change goals or permissions.
• More agents or more memory can increase failure modes; measure improvement against the simpler design.
• A trace explains observable actions, not hidden reasoning or proof that the task was completed correctly.

Skill page

Source