317 ·GitHub
What is this?
A complete course (written in Russian) on harness engineering: how to design the system around an AI coding agent - instructions, tools, environment, state and verification - so it writes code reliably.
A complete course (written in Russian) on harness engineering: how to design the system around an AI coding agent - instructions, tools, environment, state and verification - so it writes code reliably.Is this a course, a tool, or a library?It is a course, not a software library: 11 theory blocks, 14 modules, 8 hands-on projects, 14 labs, 60 problems with solutions, plus an executable test suite and a library of ready templates.
It is a course, not a software library: 11 theory blocks, 14 modules, 8 hands-on projects, 14 labs, 60 problems with solutions, plus an executable test suite and a library of ready templates.How do I get started with it?Start with the 30-minute quick start, before any theory: add an AGENTS.md (50-150 lines), a featurelist.json and a PROGRESS.md to your repo, plus one 'make check' command that returns green or red.
Start with the 30-minute quick start, before any theory: add an AGENTS.md (50-150 lines), a featurelist.json and a PROGRESS.md to your repo, plus one 'make check' command that returns green or red.What do I need to know before taking it?Git, the command line, and some hands-on experience with at least one coding agent. Machine learning knowledge is not required at all.
Git, the command line, and some hands-on experience with at least one coding agent. Machine learning knowledge is not required at all.Is this about the Harness.io CI/CD platform?No - despite the constant name confusion, it is not about the commercial Harness.io CI/CD platform; it covers harness engineering for AI coding agents. Note also that the material is written entirely in Russian.
No - despite the constant name confusion, it is not about the commercial Harness.io CI/CD platform; it covers harness engineering for AI coding agents. Note also that the material is written entirely in Russian.Who is this course for?Developers whose coding agents give unstable results, those who want an agent to work for hours across sessions without supervision, and people building internal tooling or platforms around agents.
Developers whose coding agents give unstable results, those who want an agent to work for hours across sessions without supervision, and people building internal tooling or platforms around agents.What exactly is a 'harness'?Everything outside the model weights that determines how much of its capability actually shows up in your project: instruction files (AGENTS.md, CLAUDE.md, .cursorrules, skills), available tools (shell, tests, linter, browser, MCP), the runtime environment, session state, and feedback loops such as verification commands and E2E tests.
Everything outside the model weights that determines how much of its capability actually shows up in your project: instruction files (AGENTS.md, CLAUDE.md, .cursorrules, skills), available tools (shell, tests, linter, browser, MCP), the runtime environment, session state, and feedback loops such as verification commands and E2E tests.Why does the course say upgrading the model is usually the wrong fix?Its core thesis is that model capability and execution reliability are two different things: the same model in a bare environment and in a well-built harness gives very different results, so upgrading the model is the most expensive and usually least effective way to fix an agent.
Its core thesis is that model capability and execution reliability are two different things: the same model in a bare environment and in a well-built harness gives very different results, so upgrading the model is the most expensive and usually least effective way to fix an agent.How do I check that my harness actually works?The course offers a cold-start test: open a fresh agent session, explain nothing verbally, and ask five questions - what the system is, how to run it, how to verify it, which rules must not be broken, and where the work currently stands. An executable harness test suite is included as well.
The course offers a cold-start test: open a fresh agent session, explain nothing verbally, and ask five questions - what the system is, how to run it, how to verify it, which rules must not be broken, and where the work currently stands. An executable harness test suite is included as well.What are similar projects?learn-harness-engineering, vibe-coding-cn, superpowers-zh, skills, agency-agents, Orbital, hello-agents, advanced-context-engineering-for-coding-agents
learn-harness-engineering, vibe-coding-cn, superpowers-zh, skills, agency-agents, Orbital, hello-agents, advanced-context-engineering-for-coding-agentsWhich topics is it in?graph-engineering, e2e-testing, agentic-coding, state-management, gemini-cli, context-engineering, curriculum, best-practices, coding-agents, course, prompt-engineering, cursor
graph-engineering, e2e-testing, agentic-coding, state-management, gemini-cli, context-engineering, curriculum, best-practices, coding-agents, course, prompt-engineering, cursor