Agent Skills and Tool Orchestration | Audit and Experimentation Prompt

What this prompt helps you do

Audit an existing page, process, dataset, or asset, separate facts from assumptions, and turn the highest-priority issues into tests with baselines, metrics, and stop conditions.

Prompt text

Act as a multi-agent and tool-orchestration architect and audit the current [Page/Process/Data/Asset] related to the objective: assign research, analysis, content, execution, and validation to specialized skills and agents. Materials: [Overall objective], [Subtasks], [Tool capabilities], [Shared context], [Dependencies and failure modes]. Classify findings as Blocking, Major, Moderate, or Optimization, and quote the evidence that triggers each finding. Provide: (1) an issue list; (2) Agent and skill responsibilities, Invocation sequence, Input-output contracts, Shared state, Fallbacks and evaluation; (3) at least four improvements or experiments prioritized by impact versus effort; (4) for each experiment, the hypothesis, change, primary metric, guardrail metrics, observation period, and stopping rule; and (5) required human approvals before publishing or execution. Special requirement: Each agent should own a narrowly testable responsibility. Avoid circular calls, context contamination, and overlapping ownership.