Egor Urvanov

Glossary

76 terms · 8 sets

Agents and tools · 11

AI agent
An AI agent (LLM agent, agentic AI) is a program where an LLM works inside a harness: it picks its own steps, reads files, runs commands, calls tools and checks the result right inside your environment.AI agent = LLM + harnessMCP server and AI agent for 1CHow many AI agents you need
more 19Will AI agents take your jobAnalytical reports with an AI agentClaude in ChromeThe agent’s data plane and control planeIncident investigation with an AI agentMCP scopes in Claude CodeRequirements and meeting notes with AIQA work after support with an AI agentMCP for an AI agentClaude Code, Codex, Cursor and Hermes AgentHow to run and analyze interviews about AI useChatGPT Enterprise usage report for a companyHarness engineering in practiceApproaches to working with a harnessA reconcile loop for specsSpecs by aggregateAGENTS.md and CLAUDE.md, specs, rules and skillsWho to interview before buying AI toolsPrompts for work tasks
Staging environment
A separate copy of a system where you can try and break things without touching the live one (production).MCP server and AI agent for 1CClaude in ChromeQA work after support with an AI agent
LLM
An LLM (large language model) is a neural network trained on large volumes of text and code that continues text from a prompt; it reads and writes text in tokens.AI agent = LLM + harnessHarness engineering in practice
IDE
The program where code is written: editor, hints, debugger and project launcher in one window.MCP server and AI agent for 1CClaude Code, Codex, Cursor and Hermes Agent
VS Code
VS Code (Visual Studio Code) is a free, open-source code editor from Microsoft.
Terminal
A window where commands are typed as text, also called the CLI (command-line interface).Claude in ChromeClaude Code, Codex, Cursor and Hermes Agent
Cloud coding agent
A cloud coding agent is an AI agent that runs a task on a remote server in an isolated environment with a copy of your repo.MCP server and AI agent for 1CClaude Code, Codex, Cursor and Hermes Agent
Sandbox
A fenced place where an agent’s actions run: only allowed folders are visible and only allowed network hosts are reachable.AI agent = LLM + harnessMCP server and AI agent for 1CClaude Code, Codex, Cursor and Hermes Agent
more 2A reconcile loop for specsPrompts for work tasks
Subagent
A helper the main agent launches for a separate piece of work, such as a code search or a review.AI agent = LLM + harness

Coding agents · 8

Hermes Agent
A free, open-source agent from Nous Research that you install on your own server or computer and connect to any model, through a vendor API or a local one.Claude Code, Codex, Cursor and Hermes Agent
OpenCode
A free, open-source coding agent that runs in the terminal; there is also a desktop app and IDE extensions.Claude Code, Codex, Cursor and Hermes Agent
Gemini CLI
Google’s open-source coding agent for the terminal on Gemini models, released in June 2025.AGENTS.md and CLAUDE.md, specs, rules and skills
Antigravity CLI
Google’s coding agent for the terminal, tied to the Antigravity app.
Claude in Chrome
A Claude extension for the Chrome browser that lets an agent work in your browser: open pages, click, fill forms and read the console.Claude in ChromeThe agent’s data plane and control planeMCP scopes in Claude Code
more 2QA work after support with an AI agentChatGPT Enterprise usage report for a company

Companies · 9

Anysphere
The US company behind Cursor, a code editor with a built-in agent.
DeepSeek
A Chinese company that builds the DeepSeek language models with open weights: you can download them and run them on your own server.
Yandex
A Russian company that builds the Alice AI language models, and earlier the YandexGPT line.
JetBrains
A company founded in Prague in 2000 that makes the IntelliJ IDEA, PyCharm, WebStorm and other IDEs, plus the Kotlin language.
Microsoft
A US company that makes the Copilot assistant, built into Windows, Word, Excel, Outlook and Teams.AGENTS.md and CLAUDE.md, specs, rules and skills
Atlassian
An Australian company that makes Jira, Confluence and Bitbucket, the work systems of many development teams.MCP for an AI agent

Harness · 15

Hooks
Scripts the harness runs automatically at set moments of the agent’s work: before an action, after a file edit, on finish.AI agent = LLM + harnessClaude in ChromeHarness engineering in practice
more 1AGENTS.md and CLAUDE.md, specs, rules and skills
Agent loop
The repeating rhythm of an agent’s work: think, call a tool, look at the result, think again.AI agent = LLM + harness
ReAct
A way for an agent to work in which the model alternates reasoning and acting: it decides what to do, calls a tool, reads the result and reasons again.MCP for an AI agent
Agent memory
Notes an agent keeps between sessions: project rules, constraints it found, agreements.
Project bible
A description of how the project is built and why: system boundaries, agreements, principles and values: what matters more when choosing and what is never allowed.Will AI agents take your jobAGENTS.md and CLAUDE.md, specs, rules and skills
Agent rules
Short do and do-not instructions the harness loads into the agent context: always, or only when the agent touches matching files.MCP server and AI agent for 1C
Static checks
Tests, linters, builds and other automatic verifications that confirm the agent’s result before a person looks at it.
Orchestration
Managing several steps or several agents: who is responsible for what, in which order they work and how results are passed on, through files, a subagent result or cross-session messaging.AI agent = LLM + harness
Cross-session messaging
A way for parallel agent sessions to pass text to each other: one finds another by name and sends a message, and the other reads it on its next step.
Logs and tracing
A record of each agent step: what it read, which tools it called and what came back.AI agent = LLM + harness
CLAUDE.md
A file in the project root with standing instructions for Claude Code: project layout, build and test commands, conventions.AI agent = LLM + harnessWill AI agents take your jobHarness engineering in practice
more 1AGENTS.md and CLAUDE.md, specs, rules and skills

MCP and safety · 8

MCP client
The part on the agent side: it calls a system’s MCP server, asks for data and invokes tools.MCP for an AI agent
OAuth
Sign-in through the browser: when the agent connects to a service, the harness opens that service sign-in page, you type the password there and approve access, and the agent gets a token.Incident investigation with an AI agentMCP for an AI agent
Prompt injection
An attack where someone hides an instruction for the agent inside their own text: an email, a ticket, a web page.MCP server and AI agent for 1CThe agent’s data plane and control planeIncident investigation with an AI agent

Systems and services · 5

GitLab
A platform for code and merge requests.MCP server and AI agent for 1CMCP for an AI agent
Confluence
An Atlassian wiki: the team’s knowledge base.MCP for an AI agent

Specs and architecture · 15

Harness engineering
The general discipline of tuning a harness: what the agent knows about the project, which tools it has, what it may do and how the result is checked.AI agent = LLM + harnessHarness engineering in practice
Spec-Driven Development
An approach where a specification — what the system must do — is written first, and an agent writes the code and checks it against the spec.Harness engineering in practice
Spec Kit
A GitHub toolkit for Spec-Driven Development: the specify CLI installs agent commands that lead step by step: project principles (/speckit-constitution), spec (/speckit-specify), technical plan (/speckit-plan), tasks (/speckit-tasks), code (/speckit-implement) and a check against the artifacts (/speckit-converge), repeated until code and spec agree.AGENTS.md and CLAUDE.md, specs, rules and skills
RFC 2119
A standard that fixes the meaning of MUST, SHOULD and MAY in requirements.Specs by aggregate
Domain-Driven Design
A design approach where code is built around the business domain and a language shared with the business.Specs by aggregate
Aggregate
A group of related domain objects that changes as one unit through its root and holds its own business rules.How many AI agents you needA reconcile loop for specsSpecs by aggregate
more 1AGENTS.md and CLAUDE.md, specs, rules and skills
Reconcile loop
A loop in which a system keeps comparing the desired state with the actual one and brings the second to the first.A reconcile loop for specsSpecs by aggregate
Memory Bank
A memory-bank folder with six markdown files where the agent keeps project context between sessions: projectbrief (goals and scope), productContext (why the product exists), systemPatterns (architecture), techContext (stack), activeContext (current focus) and progress (what is done).AGENTS.md and CLAUDE.md, specs, rules and skills
C4 model
A way to draw architecture at four levels: context, containers, components, code.QA work after support with an AI agent
OSI model
A reference model that splits network communication into seven layers, from physical to application.Will AI agents take your job
Epimorphism
A matching where every part of the result comes from something in the source: no part is left uncovered.MCP for an AI agent

Model and prompts · 5

Hallucination
An LLM hallucination is a confident but invented answer from a model: a nonexistent fact, function or link.Requirements and meeting notes with AIMCP for an AI agentChatGPT Enterprise usage report for a company
© 2026 Egor Urvanov