Skip to main content
Loading navigation
Back to challenges

Model Hardware Standard Physical Device Safety Controller with Claude Agents SDK

Safely enforce hardware constraints for AI agents interacting with physical devices. Build a safe device controller using Claude Agents SDK and the Model Hardware Standard specification to validate actuator motion ranges and emergency shutdown triggers before command execution.

EngineeringHosted by Vera
Sign in free to start

What you are building

Develop an MHS safety execution harness using Claude Agents SDK that intercept physical hardware command calls and verifies safety boundary rules before execution.

How work is evaluated

Evaluation

Evaluates MHS command validation accuracy and interceptor safety enforcement.

Datasets

Public datasets, plus the private datasets you have access to.

How submissions are scored

Dimensions
2 scoring checks
Binary
2 pass or fail dimensions
Ordinal
0 scaled dimensions
Dimension 1

Safety Boundary Enforcement

Ensure out-of-bounds commands return approved == False.

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Dimension 2

Safety Compliance Score

Ratio of correctly validated vs misvalidated commands. • target: 1 • range: 0-1

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Learning goals

  • Implement the Model Hardware Standard (MHS) safety verification pattern.

  • Utilize Claude Agents SDK to structure safe multi-step hardware interaction loops.

  • Interceptors and validation wrappers for physical actuator inputs.

Resources

Dataset notes

MHS telemetry and command log suite containing valid and physical safety violating robotic arm instructions.

Challenge setup

Evaluates MHS command validation accuracy and interceptor safety enforcement.

Challenge input

JSON containing command, velocity, hardware_safety_limits

Claude Agents SDK

Anthropic agent SDK for Claude.

Evaluated output

JSON approved boolean, reason, executed_command

Evaluation checks
  • Ensure out-of-bounds commands return approved == False.
  • Ratio of correctly validated vs misvalidated commands. • target: 1 • range: 0-1
Configured evaluation evidence
  • Benchmark: Model Hardware Standard Compliance Suite
  • Safety Compliance Score target: 1
  • 1 public reference case
Configured runtime
  • Python execution harness
  • Python sandbox (unavailable on Versalist)

Shows the first task, up to three tools, and up to three evidence items.

View technical recipe

Configured tools

Action Space
  • Claude Code · Required
  • Anthropic · Optional
Observation
  • Claude Code · Required
Policy Serving
  • Claude 4.1 Opus · Optional

Evaluation contract

  • Safety Boundary Enforcement · Weight 1
  • Safety Compliance Score · Weight 1

Metrics

  • Safety Compliance Score · Target 1

Benchmarks

  • Model Hardware Standard Compliance Suite
  • robotics

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON containing command, velocity, hardware_safety_limits

    Output: JSON approved boolean, reason, executed_command

    Execution harness: Python

Evaluation tools

  • Claude Agents SDK · Agent Framework

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Model Hardware Standard Physical Device Safety Controller with Claude Agents SDK