DEV Community

Cover image for Harness Engineering - Technology Landscape
Oresztesz Margaritisz
Oresztesz Margaritisz

Posted on • Edited on

Harness Engineering - Technology Landscape

#ai #sre #programming #agents

Harness engineering is the discipline of designing the environments, constraints, and feedback loops around AI coding agents that make them reliable at scale. The formula is: Agent = Model + Harness. Harness is everything that isn't the model: the infrastructure that governs how the agent operates, what it can access, and how it self-corrects.

In this article I'm collecting all the technologies I stumbled upon while researching harness engineering. The list is not exhaustive, but it should give you a good starting point to explore the landscape. This aims to be a living document, so I'll keep it updated.

high quality version of the technology landscape

SDD

Spec-Driven Development

  • OpenSpec - Spec-driven development for AI coding assistants
  • spec-kit - Toolkit to get started with Spec-Driven Development
  • BMAD - Breakthrough Method for Agile AI Driven Development

Coding Assistants

Coding assistants usable from a terminal

General Purpose

  • pi.dev - Minimal agent harness — adapt to your workflow, not vice versa
  • OpenCode - OpenCode is an open source agent that helps you write code in your terminal, IDE, or desktop.
  • NanoCoder - An open coding agent for your terminal, built by a community collective rather than a company. Bring your own model, keep your code on your machine, and owe nothing to anyone.
  • Qwen Code - An open-source AI agent for the terminal, optimized for Qwen series models

Local-First

  • Codehamr - A minimal, local-first coding agent.
  • Atomic Agent - A local-first AI agent that runs on your machine, with local or cloud models.

Orchestration

  • Symphony - Isolated autonomous implementation runs for teams
  • Conductor - Multi-agent workflows with GitHub Copilot SDK
  • OpenSquilla - Token-efficient AI agent with higher intelligence density
  • Fabro - Version-controlled workflow graphs orchestrating AI agents, commands, and human gates
  • smolagents - Barebones library for agents that think in code
  • OpenHive - Multi-agent harness for production AI workloads
  • bernstein - Audit-grade multi-agent orchestration, HMAC-chained audit log

AST and Code Parsers

  • CocoIndex Code - AST-based lightweight code search engine CLI; saves 70% tokens
  • CodeGraphContext - MCP server indexing local code into a graph database
  • ast-grep - CLI tool for code structural search, lint and rewriting in Rust
  • Graphify - Turn code and docs into queryable AI knowledge graphs

Sandboxes

  • NVIDIA OpenShell - Sandboxed agent runtime with hardware-enforced isolation and policy
  • forkd - fork() for AI agent microVMs; spawn 100 children in ~100ms
  • OpenSandbox - Secure, fast, extensible sandbox runtime for AI agents
  • Daytona - Elastic infrastructure for running AI-generated code; sub-90ms creation

Working with Skills

Skill Reference

  • Agent Skills - Standardized open format for extending AI agent capabilities via SKILL.md files

Skill Syntax Validation

  • skill-validator - Validates skill content against the Agent Skill specification

Skill Dependency Management

  • skills (Vercel Labs) - Open agent skills CLI; supports OpenCode, Claude Code, Codex, Cursor and 68+ more

Skill Security Scanners

  • DefenseClaw - Security governance for agentic AI; scan capabilities, inspect traffic, audit evidence
  • SkillSpector - Security scanner for AI agent skills; detects vulnerabilities before installation

Knowledge Base

LLM Wikis

Open Knowledge Format (OKF)

  • OKF Ecosystem - Open Knowledge Format ecosystem; tools, spec and docs for AI agent knowledge bases
  • okflint - Deterministic compliance linter for OKF bundles; profile-based rule enforcement
  • okf CLI - CLI for git-based OKF knowledge bases; keeps bundles fresh via git hooks

Technology Specifics

Token Control

  • rtk - CLI proxy reduces LLM token consumption by 60-90% on common dev commands; single Rust binary
  • ponytail - Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.
  • Headroom Desktop - MacOS menu bar app cuts Claude Code and Codex token costs by ~50%

Observability

  • Langfuse - Open source AI observability platform; LLM evals, metrics, tracing, prompt management
  • Phoenix - AI observability and evaluation by Arize; traces, evals, datasets
  • Claude Code - OTEL - How to enable and configure OpenTelemetry for Claude Code.
  • SigNoz - Coding agent observability - ClaudeCode observability with SigNoz and OpenTelemetry
  • agentops - Python SDK for AI agent monitoring, LLM cost tracking, benchmarking

References

Top comments (0)