astorlm

Agent Harness Patterns

An agent is a language model inside a loop with tools. Each level is a pattern that explains one piece, in order, from scratch.

Level 0 ← → to move · Enter to go in

Your toolkit

The pieces before any agent: the model, the system prompt, the messages you resend, and a first tool.

Coming soon

How AI agents work, from scratch

An AI agent is a language model inside a loop with tools: it reads the request, asks for a tool, gets the result back, and decides what to do next until the job is done. Agent Harness Patterns explains that loop one piece at a time, in 17 short patterns that build on each other.

Each pattern shows the problem, an animated trace of an agent solving it, and the code twice: from scratch in TypeScript and Python, and with astorlm, the open-source TypeScript agent SDK the patterns come from.

The fundamentals track

0
Your toolkit
The pieces before any agent: the model, the system prompt, the messages you resend, and a first tool.
1
What is an agent?
The difference between calling a model, building a workflow, and letting the model decide what to do.
2
Agent Loop
Call the model, run the tools it asks for, hand back the results, and repeat until it's done.
3
Designing a tool
Name, description and schema: all the model ever knows about a tool.
4
When to stop
end_turn, turn limits and terminal tools: the ways to end the loop.
5
Errors in the loop
A failing tool doesn't break the agent: the error goes back as a result and the model recovers.
6
Hooks
The points where your code steps into the loop: before each turn, each model call, each tool run.
7
The backpack fills up
What happens when the history no longer fits the context window, and how to compact it.
8
On-demand skills
A light catalog always at hand, and the full manual only when needed.
9
Memory
Remembering across sessions: saved history, files and search by meaning.
10
Fresh laps
When the job outgrows one session: a fresh agent each lap, with the progress saved in files.
11
Observability and evals
See what the agent did, what it cost, and whether it did it well.
12
Plan and reflect
Think the plan through before touching anything, and review the result before handing it over.
13
Human in the loop
Ask for permission, take corrections mid-way, and escalate when in doubt.
14
Security and sandboxing
Lock up what the agent runs, and distrust what it reads.
15
Subagents
Hand a subtask to another agent with its own clean context.
16
Proactive agents
Agents that wake up on their own: a heartbeat, a cheap check first, and a message only when it matters.

Build it with astorlm

The agent loop as a library: an embeddable, SDK-first agentic runtime for TypeScript.

pnpm add astorlm

Using an AI coding assistant? Point it at llms.txt or read the guide for AI agents.