Compare your current skills to what this opportunity needs—we'll show you what you already have and what could strengthen your application.
Job Description
Harness Engineer [33250] at Stealth Startup Harness Engineer [33250] at Stealth Startup in Menlo Park, California Posted in 1 day ago.
Type:
full-time
Job Description:
We're turning frontier model capability into a genuinely usable desktop agent product. Beyond the model itself, every system capability that lets the agent understand tasks, organize context, call tools, modify code, record its process, roll back state, and form training feedback belongs to the Harness. You'll join the Harness team to work on the core of our desktop agent product, landing model capability inside real software-development workflows. Responsibilities Work on core Harness R D, including the desktop workbench, agent loop, tool calling, context management, plugins/MCP/skills, trace, rollback, and training feedback. Own the design, implementation, testing, and iteration of React / Electron / Python / engineering-infrastructure modules. Design and improve the agent's code-change pipeline-Git control, diff/patch, staging, branch/worktree, change tracking, conflict handling, rollback, and restore. Collaborate with researchers to analyze the agent's successes and failures on real tasks, driving the Harness and model capability to evolve together. Dogfood the product on real internal development tasks, continuously surfacing and resolving usability, reliability, and developer-experience issues. Contribute to technical architecture, engineering quality, product experience, and project execution. Qualifications 2+ years of software development experience (flexible for exceptional candidates). Bachelor's degree or above from a strong university. Solid engineering ability, with deep proficiency in at least one of the following: Frontend. React, TypeScript, state management, complex interactions, desktop UI engineering. Desktop. Electron, main/renderer process communication, local file system, process management, cross-platform desktop apps. Backend. Python, async/process models, tool systems, server-side architecture, testing, and engineering practices. Engineering / version control. Git, branch/worktree, diff/patch, merge/rebase, conflict handling, code-change tracking, rollback/restore, CI/test pipelines. Not all four areas required-but you must be able to independently own one core area and read and collaborate on another. Heavy hands-on use of AI agent tools for real software development, with intensive experience using agent products.
Understanding of LLM/agent fundamentals:
LLM APIs, context window, agent loop, tool use, reasoning, planning, MCP, memory, subagents, etc. Good engineering habits-valuing maintainability, testing, observability, safety boundaries, and long-term evolution cost. Strong Chinese communication skills. Nice-to-haves Capability across multiple of React / Electron / Python / engineering at once. Familiarity with Git internals-object store, index, refs, worktree, patch apply, merge conflict, reflog-or experience building Git-based snapshot, rollback, audit, or sync systems. Experience with AI agents, code assistants, IDEs, developer tools, LLM applications, or automated testing/evaluation platforms. Heavy use of Claude Code, Codex, Cursor, Windsurf, Devin, OpenAI Agents SDK, LangGraph, the MCP ecosystem, or similar products/frameworks. Experience with plugin systems, MCP, LSP, sandboxes, terminal orchestration, trace/replay, workflow engines, or evaluation systems. Personal open-source work, open-source contributions, research experience, competition awards, or publications. Able to communicate in English with open-source or user communities. What We Value We value solid software-engineering ability first. The Harness is complex, but it ultimately comes down to React, Electron, Python, tool systems, file systems, processes, state, testing, Git control, and user experience. The ideal candidate need not know everything, but must be strong in at least one engineering direction and have real interest in and judgment about agent products.
