How does CrewAI's role/goal/backstory pattern shape the system prompt of each agent?

In CrewAI, every agent is built by filling three fields: `role`, `goal`, and `backstory`. CrewAI then compiles these into a structured system prompt before the agent runs. **Role** sets identity, what kind of agent this is ("Senior research analyst", "Python code reviewer"). **Goal** sets the objective, what this agent is trying to achieve ("Identify peer-reviewed sources on topic X", "Find bugs and security issues in the diff"). **Backstory** supplies implicit context that should color every response ("You have ten years of experience in market research and prefer primary sources to secondary summaries"). The framework templates these into a system prompt the LLM sees on every turn. You never hand-write the full prompt, you fill three slots and CrewAI assembles them in a consistent format across the crew. A `Crew` then orchestrates multiple agents using a Process policy, `sequential` chains them one after another, `hierarchical` adds a manager agent that delegates. The role/goal/backstory pattern is what gives each agent a distinct voice in that orchestration.

**CrewAI's role/goal/backstory** triple is structured prompt scaffolding. Instead of writing one ad-hoc system prompt per agent, you fill three named slots that map to identity, objective, and implicit context. The framework templates them into a consistent system prompt before any task runs. **Role** is the agent's identity, "Senior data analyst", "Adversarial code reviewer". **Goal** is what the agent is trying to accomplish on every task it receives, "Surface anomalies in the dataset and explain their likely cause". **Backstory** is the persona context that shapes voice and judgment, "You have a decade of experience in fraud detection and distrust convenient explanations". The payoff is consistency. Two agents built from the same template generate consistent system prompts, easier to A/B-test prompts, easier to swap models without rewriting per-agent prose, easier to reason about why one agent disagrees with another. The orchestration layer, `Crew(agents=[...], process=Process.sequential)` or `Process.hierarchical`, then threads task outputs between agents using the policy. The pattern is not magic, you could write the same prompt by hand. The win is that the framework owns the template, so a 10-agent crew has 10 consistent personas instead of 10 hand-rolled prompts that drift over time.