Interview OS
Back to Context Engineering

Library

Agent loops & guardrails

Plan, act, observe, repeat.

Agents

Overview

Agent loops let a model plan, act, observe, and repeat until a goal is met. Understanding the reason-act cycle and its stopping conditions is key to building agents that make progress without spiraling.

How it works

Agents
PlanActObserveGuardGoalplanTool/stepcallResultfeedbackLimits
ClientServiceEdgeData

Step by step, with examples

  1. 1

    Goal

    • Break the task into steps.
  2. 2

    Tool/step

    • Take one action.
  3. 3

    Result

    • Read the outcome and reflect.
  4. 4

    Limits

    • Cap steps; validate; add human-in-loop.
    • Example: ReAct loop

Overview

Agents iterate: plan a step, call a tool, observe the result, and decide the next move — bounded by step limits and validation.

Common pitfalls

  • No loop/step cap
  • Acting on unverified output
  • No human-in-the-loop for risky actions

Where this content comes from

For full transparency, this content is curated and verified from these sources:

Frontier-lab prompting & agent guidesRetrieval-augmented generation literatureOppZen-authored context-engineering playbooks