[Hands-on] Rebuilding Claude Code's Harness

2026-07-19 · Source: Daily Dose of Data Science · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Intermediate, long

Summary

The article details rebuilding Claude Code's agent harness using an open-source stack, CrewAI, emphasizing that the "harness"—the surrounding code handling planning, tool execution, memory, and safety—is crucial for advanced coding agents, rather than solely the LLM. The author systematically reconstructed six key layers: the core agent loop, tool integration, planning capabilities, subagent orchestration, sandboxing for secure execution, and persistent memory. Each layer was mapped to corresponding CrewAI features, demonstrating that much of this complex machinery is available as configuration. While CrewAI simplifies many aspects, critical engineering tasks like prompt design, execution environment setup, and precise tool selection remain. The reconstructed harness successfully resolved 3 failing tests in a BankAccount class, achieving 5 passing tests, mirroring Anthropic's internal evaluation methods for coding agents.

Key takeaway

For AI Engineers building robust coding agents, recognize that the surrounding harness, not just the LLM, dictates performance and safety. You should prioritize implementing layers like planning, subagents, sandboxing, and persistent memory. CrewAI offers much of this as configuration, streamlining development. However, your expertise in prompt engineering, execution environment setup, and precise tool selection remains critical for optimal agent behavior and cost efficiency.

Key insights

The agent harness, not just the LLM, defines advanced coding agent capability.

Principles

• Agent capability stems from its harness, not just the model.
• Planning and reasoning prevent context rot in complex tasks.
• Sandboxing and permissions ensure agent safety.

Method

The article describes a step-by-step process of building an agent harness: core loop, tools, planning, subagents, sandboxing, and memory, using CrewAI for implementation.

In practice

• Configure CrewAI with "planning=True" for task roadmaps.
• Use hierarchical crews to manage agent context.
• Integrate E2B tools for sandboxed code execution.

Topics

• Agent Harness
• CrewAI
• LLM Agents
• Multi-Agent Systems
• Sandboxing
• Persistent Memory

Code references

• crewAIInc/crewAI
• patchy631/ai-engineering-hub

Best for: AI Engineer, Machine Learning Engineer, AI Architect

Related on AIssential

• Building a Coding Agent From Scratch — A coding agent's harness, not the LLM, is the primary driver of its performance and engineering value.
• A harness for every task: dynamic workflows in Claude Code — Claude Code's dynamic workflows enable on-the-fly multi-agent orchestration to overcome single-context AI limitations.
• Components of A Coding Agent — Agentic harnesses significantly enhance LLM coding capabilities by integrating tools, context, and memory management.
• Your coding agent doesn't always follow your rules — Talha Sheikh, Checkout.com — Model capability does not guarantee reliability; robust verification harnesses are essential for trustworthy AI agent outputs.
• Harness Engineering: The Missing Architectural Layer Between Powerful Models and Reliable AI Agents — The system around an AI model, not just the model itself, determines agent reliability and capability.
• The Model Is the Easy Part: Why Every AI Agent Needs a Harness — The model is the brain; the harness is the body, enabling LLMs to act in the real world.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Daily Dose of Data Science.