Complete for freeAdvancedBrowser simulator included

Course 25

Custom Agent Harnesses, Evaluations, and Multi-Agent Systems

Engineer secure agent runtimes, sandboxes, traces, evaluations, graders, routing, multi-agent roles, loop controls, approvals, rollbacks, adversarial tests, and measurable improvement cycles. The course combines structured lessons, guided browser concept simulations, and learner-created project files made with freely installable or open tools described in the course instructions. This guided page does not execute code, train models, run databases, create repositories, access hardware, or provide hosted infrastructure.

Free completion promise: The always-free pathway combines multi-agent concept simulations with learner-created harness, trace, grader-design, routing, and security files. The C4T page does not provide a runtime, repository, model, grader, or sandbox; paid model and evaluation services are optional.

Before you begin

Requirements & setup for Custom Agent Harnesses, Evaluations, and Multi-Agent Systems

Start with the complete free pathway below. Hardware, subscriptions, paid APIs, and cloud credits are never required.

Start here · free path

Your complete $0 setup

The always-free pathway combines multi-agent concept simulations with learner-created harness, trace, grader-design, routing, and security files. The C4T page does not provide a runtime, repository, model, grader, or sandbox; paid model and evaluation services are optional.

$0 required

Setup checklist

0 of 9 complete

This checklist is a temporary setup aid and resets when the page reloads. Saved course progress begins in the setup lessons below.

  1. Use the browser-based tools and simulations throughout this course. Paid software, APIs, cloud credits, GPUs, and physical equipment are optional.

    About 2 min

  2. Open the concept simulator and read its scenario, assumptions, and limitations.

    About 5 minOpen C4T Local Agent Harness Concept Simulator (opens in a new tab)

  3. Make a prediction, adjust the available controls, and compare the result with the explanation.

    About 5 min

  4. Record what changed and why; repeat with at least one different input.

    About 5 min

  5. For applied work, follow the course instructions to create and keep your own project files with a freely installable or open tool appropriate to the subject.

    About 5 min

  6. Open C4T Local Agent Harness Concept Simulator.

    About 5 minOpen C4T Local Agent Harness Concept Simulator guided alternative (opens in a new tab)

  7. Choose the browser-based guided pathway.

    About 5 min

  8. Complete this check: Submit the learner-created file or documented design evidence requested by the assignment; the C4T page does not verify local execution.

    About 5 min

  9. C4T Local Agent Harness Concept Simulator: Complete one prediction-and-explanation cycle and record the evidence you would carry into a learner-created project file. Optional local model runtime: Submit the learner-created file or documented design evidence requested by the assignment; the C4T page does not verify local execution.

    About 5 min

Software

2 items

C4T Local Agent Harness Concept Simulator

A free browser concept simulator with adjustable values, a prediction prompt, deterministic feedback, and a short explanatory trace. This guided page does not execute code, train models, run databases, create repositories, access hardware, or provide hosted infrastructure.

Course need
C4T Local Agent Harness Concept Simulator
Free

A free browser concept simulator with adjustable values, a prediction prompt, deterministic feedback, and a short explanatory trace. This guided page does not execute code, train models, run databases, create repositories, access hardware, or provide hosted infrastructure. Use keyboard controls and the text trace or written alternative when a visual or pointer interaction is not suitable.

  1. Open the concept simulator and read its scenario, assumptions, and limitations.
  2. Make a prediction, adjust the available controls, and compare the result with the explanation.
  3. Record what changed and why; repeat with at least one different input.
  4. For applied work, follow the course instructions to create and keep your own project files with a freely installable or open tool appropriate to the subject.
Open C4T Local Agent Harness Concept Simulator (opens in a new tab)

No-hardware path

Local Agent Harness Concept Simulator

Explore course concepts through adjustable values, prediction, deterministic feedback, and an explanatory trace. No hardware is required.

Optional paid alternatives

You do not need these options to complete the course.

Paid model API

Optional learner-managed benchmark outside the C4T site; credentials and cost limits remain the learner's responsibility.

Optional only—never required for course completion.

Hosted evaluation platform

Optional reporting comparison outside the C4T site; it is never part of the always-free path.

Optional only—never required for course completion.

Optional local model runtime

Optional learner-managed software for extending project files outside the guided concept simulator. It is not supplied, installed, connected, or run by this site.

Optional
C4T Local Agent Harness Concept Simulator guided alternative
Free

Use the built-in guided concept lab; native installation is an optional extension. The simulator, supplied traces, and written tables cover the assessed concepts without handling physical parts.

  1. Open C4T Local Agent Harness Concept Simulator.
  2. Choose the browser-based guided pathway.
  3. Complete this check: Submit the learner-created file or documented design evidence requested by the assignment; the C4T page does not verify local execution.
Open C4T Local Agent Harness Concept Simulator guided alternative (opens in a new tab)

No-hardware path

Local Agent Harness Concept Simulator

Explore course concepts through adjustable values, prediction, deterministic feedback, and an explanatory trace. No hardware is required.

Hardware

Nothing required

No physical hardware is needed for this course.

Accounts

Nothing required

No external account is needed for this course.

What you will be able to do

Learning outcomes

  • Implement an observe-plan-act-verify runtime with bounded tools, context, recovery, and traces.
  • Sandbox execution and enforce file, process, time, and network policy.
  • Build deterministic and carefully justified model-based evaluations resilient to gaming.
  • Coordinate supervised agents with routing, loop controls, human approvals, rollback, and security evidence.

Included downloads

Free starter assets

Original · synthetic · licensed

These small files contain no personal data, credentials, or secrets. Dataset fixtures are CC0 and starter code is MIT licensed.

Starter pack guide & licensesSee what each original synthetic dataset, fixture, and starter file contains.Bounded agent evaluation taskA fictional repository task with deterministic checks and no credentials.Recorded safe agent traceA simulated multi-agent trace that blocks an unsafe action.

Build real work

Assignments & capstone

Free tools accepted

Bounded single-agent harness

Build the observe-plan-act-verify loop with validated tools, context strategy, sandbox policy, retries, cleanup, and a complete trace.

The always-free pathway combines multi-agent concept simulations with learner-created harness, trace, grader-design, routing, and security files. The C4T page does not provide a runtime, repository, model, grader, or sandbox; paid model and evaluation services are optional. Submit a project file, text trace, screenshots with descriptions, or equivalent simulator evidence.

Adversarial evaluation and routing suite

Create versioned tasks, deterministic and justified model-based graders, hidden checks, attacks, metrics, and a multi-agent routing policy.

The always-free pathway combines multi-agent concept simulations with learner-created harness, trace, grader-design, routing, and security files. The C4T page does not provide a runtime, repository, model, grader, or sandbox; paid model and evaluation services are optional. Submit a project file, text trace, screenshots with descriptions, or equivalent simulator evidence.

Final assessment

Secure evaluated multi-agent engineering system

Design, deploy locally, attack, evaluate, and improve a system with supervisor, planner, worker, reviewer, permission boundaries, context strategy, communication, loop controls, approvals, rollback, deterministic and justified model-based evaluation, adversarial cases, injection and reward-hacking analysis, operational metrics, security reporting, and a final demonstration. Deliverables: Learner-created repository plan and bounded runtime design, Role, routing, memory, and communication design, Versioned evaluation tasks and graders, Adversarial, injection, and reward-hacking results, and Success, latency, token, cost, and security improvement report.

Prerequisites & pacing

Skills assumed

  • Safe tool-using agents, MLOps reliability, Git workflows, and software testing

Recommended prior courses

These are advisory, not locks. If you already have the skills listed above, enroll directly.

Choose your pace

  • self paced: 14 weeks at 6 hours per week
  • weekly: 14 weeks at 6 hours per week

Part of this learning path

  • 🛠️ Agentic Systems · step 5 of 5

    Build tool-using AI agents that are auditable, permission-bounded, and evaluated against adversaries.

Course created by Zafir Shamsi.