Planning, Memory and Stopping · ai-skills

Tools, Workflows and Agents · Beginner to Intermediate

4topics
4lessons

An agent loop plans a task, acts with tools, and records what happened. Planning, external memory, and stopping rules keep long runs correct and bounded.

Concern	Rule
Plan	Split the task into verifiable steps with a done condition.
Memory	Keep state in notes outside the model; hold only the working slice in context.
Stopping	Bound steps, tokens, and repeats; stop with a named reason.

Agent loop with plan, act, observe, budget and loop checks, and stop.

One loop turn plans, acts, observes, checks budgets and repeats, and stops with a reason.

Lesson	What it covers
Task decomposition	Split tasks into verifiable steps and define done first.
Memory and summaries	External state, context growth, and safe summarisation.
Budgets and enforcement	Step and token budgets with a runnable enforcer.
Loop detection and stopping	Repeat detection with a runnable detector and stop rules.

Check yourself An agent repeats the same two tool calls six times with no new information. Which control answers it first?

Loop detection and stopping. The ABAB pattern triggers a stop, while the step and token budgets bound the cost of the run.

Up nextTask Decomposition and Success Criteria

Practice this lesson Report an error