The glossary of coding agents: the vocabulary, defined.

Thirty-four terms you will meet around AI coding agents, in two groups: the methods and names for how people work with them, from agentic engineering to vibe coding, and the tool concepts, from CLAUDE.md to MCP. Each has a definition you can quote, its Polish name and its place on the ladder. Free to read.

Jump to a letter

34 terms in 8 letters

The full A to Z glossary in the docs has more entries.

Thirty-four terms, A to Z.

Each entry gives the definition, the Polish name and a place on the ladder. Open an entry for the detail, tool by tool where the tools differ.

Each method and name here is defined against vibe coding; the long-form guide to agentic engineering vs vibe coding takes the comparison further.

Show one kind of term

Show one ladder level

Levels are our reading of the free ladder guides: for a tool concept, the rung where they first have you set it up; for a method, the closest rung. The levels themselves are Dan Shapiro’s (opens in a new tab) (“The Five Levels”, danshapiro.com, 2026-01-23).

A

Agent orchestration

PolishOrkiestracja agentów

All levels

Agent orchestration is coordinating several coding agents on one body of work: split it, run the pieces in parallel or in sequence, check each, then merge. No single person coined the name; Anthropic described an "orchestrator-worker pattern" in June 2025. Orchestration multiplies whatever review discipline the human has: with a judge, tests and a merge gate it is a factory, and without them it is vibe coding at fleet scale.

Also searched as multi-agent orchestration, multi-agent development, multi-agent coding. Checked Vibe coding vs agentic engineering: 21 terms compared, 2026-10-10

Open the Agent orchestration entry

Agent Skills

PolishAgent Skills

All levels

Agent Skills are an open format for reusable instructions: a folder holding a SKILL.md whose name and description the agent always sees, and whose body loads only when the task matches. Claude Code, Codex and Cursor all read them, and the skills CLI installs one across agents with npx skills add OWNER/REPO.

Also searched as skills, SKILL.md, skills.sh. Checked Top practice skills guide, 2026-10-12

Open the Agent Skills entry

Agentic coding

PolishKodowanie agentowe

All levels

Agentic coding is a mode of work in which a coding agent plans a change, writes the code, runs it and iterates on the result in a loop, instead of suggesting the next line. No single coiner is established; Anthropic made it a product category with Claude Code in February 2025. The term says what the tool does and nothing about how carefully you check its output.

Also searched as agentic programming, AI agentic programming. Checked Vibe coding vs agentic engineering, 2026-10-10

Open the Agentic coding entry

Agentic engineering

PolishInżynieria agentowa

All levels

Agentic engineering is the discipline of directing fallible coding agents without lowering the quality bar, a name Andrej Karpathy popularised on 4 February 2026. Instead of asking whether an agent can write something, you place each loop on the autonomy ladder and ask what would move it up, such as a stop condition a machine can check instead of a human reading the diff.

Checked State of agentic engineering report, 2026-10-12

Open the Agentic engineering entry

Agentic software development

PolishAgentowe tworzenie oprogramowania

All levels

Agentic software development is an operating model in which autonomous agents work across the whole software development lifecycle, from planning to release, while humans stay accountable at the gates. No single coiner is established; Microsoft wrote of “agentic DevOps” in May 2025 and Forrester titled a 2026 report The State of Agentic Software Development. The unit is the lifecycle rather than a single coding session.

Also searched as agentic development, agentic software engineering. Checked Vibe coding vs agentic engineering, 2026-10-10

Open the Agentic software development entry

Agentic workflows

PolishAgentowe przepływy pracy

All levels

Agentic workflows are LLM applications in which the model works through a task in several passes, using reflection, tools, planning or other agents, instead of answering once. Andrew Ng popularised the name in 2024, and it reaches far beyond code: his own examples include writing text and answering questions. It sits on a different axis from vibe coding, which is about whether a human reads the result.

Also searched as agentic workflow, agent workflows, AI agentic workflows. Checked Vibe coding vs agentic engineering: 21 terms compared, 2026-10-10

Open the Agentic workflows entry

AGENTS.md

PolishAGENTS.md

Level 2

AGENTS.md is an open Markdown format for agent instructions: build commands, conventions and boundaries, kept at the repository root and, for scoped detail, in subdirectories. Codex reads it before doing any work. By default Claude Code reads it only when a project has no CLAUDE.md (v2.1.277+), and Cursor reads it in the project root and in subdirectories.

Also searched as AGENTS.md file, agents.md convention. Checked Cursor token management guide, 2026-10-12

Open the AGENTS.md entry

AI coding agents

PolishAgenci kodujący AI

All levels

AI coding agents are software in which a language model runs in a loop with tools: it reads a codebase, edits files, runs commands and checks results until a goal is met. The category has no single coiner; Simon Willison defines coding agents by one feature, that they can both generate and execute code. It names a tool, not a way of working.

Also searched as coding agents, AI coding agent, agentic coding tools, CLI coding agents. Checked Vibe coding vs agentic engineering, 2026-10-10

Open the AI coding agents entry

AI engineering

PolishInżynieria AI

All levels

AI engineering is the discipline, and the job, of building applications on top of readily available foundation models: the prompts, context, evaluation and fine-tuning around the model. Shawn “swyx” Wang popularised the role in June 2023, and Chip Huyen’s 2024 book defines the field. Its subject is AI inside the product. Using AI to write code is a different subject.

Also searched as AI engineer, LLM engineer, foundation model engineer. Checked Vibe coding vs agentic engineering, 2026-10-10

Open the AI engineering entry

AI pair programming

PolishProgramowanie w parze z AI

Level 2

AI pair programming is the framing of a coding assistant as a second programmer you work beside: the human steers and reads each suggestion as it appears. GitHub Copilot popularised the phrase “your AI pair programmer” on 29 June 2021, after Codota used it in 2017. On this site’s autonomy ladder it is Level 2.

Also searched as AI pair programmer, pair programming with AI, pairing with AI. Checked Level 1–2: Assisted and Paired Coding, 2026-10-12

Open the AI pair programming entry

AI SDLC

PolishAI SDLC

All levels

AI SDLC is the software development lifecycle rebuilt around coding agents: agents do the first pass at each stage, and named humans own the gates between stages. No single person coined the name, which is also written agentic SDLC or AI-native SDLC, and Anthropic’s playbook of 21 August 2026 treats the labels as one thing. Unlike vibe coding, it has stages and gates.

Also searched as agentic SDLC, AI-native SDLC, AI-SDLC, AI software development lifecycle, AI-native software development lifecycle. Checked Vibe coding vs agentic engineering: 21 terms compared, 2026-10-10

Open the AI SDLC entry

AI-assisted coding

PolishKodowanie wspomagane AI

All levels

AI-assisted coding is the umbrella term for any use of large language models and AI agents to help write software, from autocomplete and chat to coding agents. It has no single coiner and says nothing about who writes the code or who checks it. Simon Willison calls vibe coding “the less responsible subset” of it.

Also searched as AI assisted coding, AI-assisted programming, AI-assisted development, AI-assisted software development. Checked Level 1–2: Assisted and Paired Coding, 2026-10-12

Open the AI-assisted coding entry

AI-DLC

PolishAI-DLC

Level 4

AI-DLC is the AI-Driven Development Lifecycle, a methodology that Raja SP and his AWS team introduced on 31 July 2025. AI drafts a plan, asks clarifying questions and implements only after the team validates it, in sessions AWS calls “Mob Elaboration” and “Mob Construction”, and sprints become “bolts” of hours or days. AWS says these rituals keep AI’s suggestions from being blindly accepted; vibe coding accepts them unread.

Also searched as AIDLC, AI DLC, AI-Driven Development Lifecycle, AI-Driven Development Life Cycle, aidlc-workflows. Checked Vibe coding vs agentic engineering: 21 terms compared, 2026-10-10

Open the AI-DLC entry

AI-driven development

PolishProgramowanie sterowane przez AI

All levels

AI-driven development is a generic phrase for software work that AI leads rather than assists. It has no coiner and no canonical definition; its strongest anchor is AWS’s AI-Driven Development Lifecycle of 31 July 2025, which sets AI-assisted and AI-autonomous development against a third way. The phrase says who leads, not who checks, so it cannot tell disciplined AI-led work from vibe coding.

Also searched as AI driven development, AI-driven software development. Checked Vibe coding vs agentic engineering: 21 terms compared, 2026-10-10

Open the AI-driven development entry

AI-native development

PolishProgramowanie AI-native

All levels

AI-native development is an operating model in which a team redesigns its software process around AI instead of adding AI to the old one; Tessl’s phrase is “built-in, not bolted on”. Guy Podjarny introduced “AI Native Software Development” at Tessl’s launch on 6 July 2024. It describes how a team works, not one developer’s habit, and says nothing about autonomy.

Also searched as AI native development, AI Native Dev, AI-native software development, AI-native engineering. Checked Level 1-2 ladder guide, 2026-10-12

Open the AI-native development entry

Augmented coding

PolishKodowanie rozszerzone

All levels

Augmented coding is Kent Beck’s name, in use from April 2025, for programming with an AI “genie” while still caring about the code, its complexity and its tests. On 25 June 2025 he set it against vibe coding, where only the behaviour of the system matters: augmented coding keeps the values of hand coding, tidy code that works, and only the typing changes.

Also searched as AI-augmented coding. Checked Vibe coding vs agentic engineering guide, 2026-10-10

Open the Augmented coding entry

C

CLAUDE.md

PolishCLAUDE.md

Level 2

CLAUDE.md is Claude Code’s instruction file. It loads at session start from the project root and ~/.claude/CLAUDE.md, and from a subdirectory when the agent works there, and holds the commands, conventions and pitfalls Claude cannot infer from the code. /init drafts one, /memory edits it, and @path/to/file imports another file. It guides the model and enforces nothing.

Also searched as CLAUDE.md file, Claude Code memory file. Checked Cursor token management guide, 2026-10-12

Open the CLAUDE.md entry

Compound engineering

PolishCompound engineering

Level 3

Compound engineering is a method for working with coding agents in which every change leaves the next one easier: each loop of planning, work and review ends with a step that writes the lesson into files the agent reads next time. Kieran Klaassen of Every coined it on 18 August 2025, first as “compounding engineering”, and Every ships it as a plugin.

Also searched as compounding engineering, compound engineering plugin, Every compound engineering. Checked Level 4 ladder guide, 2026-10-12

Open the Compound engineering entry

Context engineering

PolishInżynieria kontekstu

All levels

Context engineering is the practice of deciding what a model sees at each step: instructions, tools, retrieved data and history. Tobi Lütke and Andrej Karpathy popularised the name in June 2025, though others used it earlier. In a coding agent it covers CLAUDE.md, AGENTS.md, files read and tool output. It governs what the agent reads and says nothing about checking its output.

Also searched as context engineering for agents, context engineering for coding agents, agentic context engineering. Checked Cursor token management guide, 2026-10-12

Open the Context engineering entry

Context window

PolishOkno kontekstu

All levels

A context window is the token budget a model can hold in one request. In a coding agent it is spent on the system prompt, CLAUDE.md or AGENTS.md, tool definitions, file reads, command output and conversation, all reread every turn. When it fills, the agent compacts older material and instructions it was following can quietly disappear. Its size depends on the model and the host.

Also searched as context length, token limit. Checked Context windows guide, 2026-09-26

Open the Context window entry

G

Git worktrees

PolishGit worktrees

Level 3

Git worktrees are extra working directories for one repository: each has its own files, index and checked-out branch or commit, and all of them share one .git history. Parallel agents each get one, so their edits, installs and tests never touch the same files. Claude Code starts one with --worktree, Codex with --worktree or /worktree, and Cursor documents Worktrees as isolated Git checkouts.

Also searched as worktree, git worktree. Checked Worktrees guide, 2026-09-26

Open the Git worktrees entry

H

Harness engineering

PolishInżynieria harnessu

Level 5

Harness engineering is the practice of turning each recurring agent mistake into a permanent control, such as an instruction, a tool, a test or a linter rule, so the mistake cannot recur. Mitchell Hashimoto named it in writing on 5 February 2026, and an OpenAI post put it in its title six days later. Unlike vibe coding, it adds checks that run even when nobody reads the diff.

Also searched as agent harness engineering, engineer the harness, Agent = Model + Harness. Checked Harness guide, 2026-10-12

Open the Harness engineering entry

Headless mode

PolishTryb headless

Level 4

Headless mode is running a coding agent without an interactive session, usually from a script, a git hook, a cron job or a CI step. Claude Code uses claude -p, Codex uses codex exec and Cursor’s CLI has print mode (-p). Nobody is there to answer a prompt, so permissions, output format and the check on the result have to be set on the command line.

Also searched as non-interactive mode, print mode, agent in CI. Checked codex exec guide, Codex 0.157.1, 2026-10-03

Open the Headless mode entry

Hooks

PolishHooki

Level 4

Hooks are scripts or handlers that a coding agent runs at fixed points in its loop, such as before a tool call or when a turn ends, whatever the model decides. On the events that can block, a hook stops the action, which a line in CLAUDE.md cannot. Claude Code 2.1.283 documents 33 hook events and Codex 0.157.1 documents 12; Cursor’s hooks can block too.

Also searched as agent hooks, Claude Code hooks, lifecycle hooks. Checked Security gates guide (Cursor docs), 2026-10-12

Open the Hooks entry

M

MCP (Model Context Protocol)

PolishMCP (Model Context Protocol)

All levels

MCP, the Model Context Protocol, is the open standard that lets a coding agent call external tools and read live data exposed by an MCP server, such as GitHub pull requests, Sentry errors, a browser or a database. Claude Code, Codex and Cursor are all MCP clients. Anthropic created MCP and donated it to the Agentic AI Foundation in December 2025.

Also searched as Model Context Protocol, MCP server. Checked MCP overview, 2026-09-26

Open the MCP (Model Context Protocol) entry

P

Plan mode

PolishTryb planowania

Level 2

Plan mode is a phase in which a coding agent reads the codebase and proposes a plan before it edits anything. Claude Code has it as the plan permission mode (Shift+Tab, /plan or --permission-mode plan), where edits stay blocked until you approve; Codex has /plan, and Cursor has Plan Mode.

Also searched as planning mode, plan first. Checked Build-stage guide, 2026-10-12

Open the Plan mode entry

Prompt engineering

PolishInżynieria promptów

All levels

Prompt engineering is the craft of writing and refining instructions for a language model so that it reliably produces the output you need. No coiner is established: the phrase was in use around GPT-3 in 2020. Since mid-2025 Anthropic, LangChain and Cognition have called context engineering its next step or its superset. It concerns the quality of the input; whether anyone reviews the output is a separate question.

Also searched as prompting, prompt design, prompt crafting, prompt programming, prompt engineer. Checked Prompt engineering for Claude Code, 2026-10-12

Open the Prompt engineering entry

S

Sandbox

PolishSandbox

Level 2

A sandbox is an operating-system boundary on what the commands an agent runs can read, write and reach on the network. It is separate from approval, which decides whether a command runs at all. Claude Code sandboxes Bash with Seatbelt on macOS and bubblewrap on Linux and WSL2; Codex offers --sandbox read-only, workspace-write and danger-full-access.

Also searched as sandboxing, sandbox policy, OS sandbox. Checked Permissions and sandboxes guide, 2026-09-26

Open the Sandbox entry

Slash commands

PolishKomendy slash

Level 2

Slash commands are controls typed with a leading / in the prompt of a coding agent: built-ins such as /clear, /compact, /model, /plan and /review, and your own commands, written as skills in Claude Code or kept in .cursor/commands/ in Cursor. Type / in Claude Code or Codex and the live list for your build appears; it is the only inventory that stays correct.

Also searched as /commands, custom commands. Checked Custom commands guide, Claude Code 2.1.283, 2026-10-03

Open the Slash commands entry

Software factory

PolishFabryka oprogramowania

Level 5

A software factory is a pipeline in which coding agents pull work from a queue, build and verify it, and open pull requests with little or no human reading the diff. It is the top of the autonomy ladder, Level 5, reached one loop at a time: a loop may run unattended only when an oracle the agent cannot edit decides it is done.

Also searched as software factories. Checked Level 5 guide, 2026-09-26

Open the Software factory entry

Spec-driven development

PolishProgramowanie sterowane specyfikacją

Level 4

Spec-driven development is a workflow in which a coding agent builds against a written specification that a human reviews before any code exists. No single person is credited with coining the name for the AI era: Kiro’s launch post used it on 14 July 2025 and GitHub’s Spec Kit followed on 2 September 2025. Vendors present it as the step after vibe coding, not its opposite.

Also searched as SDD, spec driven development, specification-driven development. Checked Spec-driven development guide, 2026-10-12

Open the Spec-driven development entry

Sub-agent

PolishSubagent

Level 3

A subagent is a helper agent that the main agent starts for one subtask, with its own context window, so the parent’s context stays small. It does not see your conversation history: Claude Code gives it its own prompt, the task message, your CLAUDE.md files and a git status snapshot, and it returns a summary. Claude Code ships Explore, Plan and general-purpose subagents; Codex has /subagents; Cursor has Subagents.

Also searched as subagent, sub agents. Checked Claude Code subagents guide, Claude Code 2.1.283, 2026-09-26

Open the Sub-agent entry

V

Vibe coding

PolishVibe coding

Off the ladder

Vibe coding is building software by prompting an AI model and accepting its code without reading it, a term Andrej Karpathy coined in an X post on 2 February 2025. The human judges the behaviour and skips the diff, and stops when “it mostly works”. Karpathy said it is “not too bad for throwaway weekend projects”; dictionaries now also use the word for any AI code generation.

Also searched as vibecoding, vibe-coding, vibe code. Checked Vibe coding vs agentic engineering guide, 2026-10-10

Open the Vibe coding entry

Vibe engineering

PolishVibe engineering

All levels

Vibe engineering is the name Simon Willison proposed on 7 October 2025 for the other end of the spectrum from vibe coding: seasoned professionals using LLMs and coding agents to work faster while staying accountable for the software they produce. On 23 February 2026 he added that “agentic engineering” was coming out on top as the name for it.

Also searched as vibe-engineering, vibe engineer. Checked Vibe coding vs agentic engineering guide, 2026-10-10

Open the Vibe engineering entry

Definitions come from the free glossary entries. Each entry shows the date it was last checked against the guides on this site. Terms the ladder guides do not tie to one rung read “All levels”, and a term the ladder does not hold at all reads “Off the ladder”.

The same idea has three names.

Claude Code, Codex and Cursor each name the same job differently. These are the lines from the entries above, side by side.

“Using several of them can feel messy. I switch between Claude Code, Cursor, VS Code, GitHub, and other agents, and often have to explain the same project or problem again.”
A reader, July 2026 survey
How Claude Code, Codex and Cursor each handle eight of the terms
Term Claude CodeCodexCursor
Project instructions Loaded at launch from the project root and ~/.claude/, and from a subdirectory when Claude reads a file there; the files add up and none overrides another.Project initialization guide, Claude Code 2.1.283, 2026-09-26 Reads AGENTS.md natively, and /init creates one.Codex AGENTS.md guide, Codex CLI 0.157.1, 2026-09-26 Applies instructions through Rules in .cursor/rules/. Cursor’s docs say it reads CLAUDE.md the same way it reads AGENTS.md and always applies it; the agent runtime in @cursor/sdk 1.0.32 (checked 2026-09-26) loads CLAUDE.md and CLAUDE.local.md only when third-party configuration loading is on, while it loads AGENTS.md unconditionally.Cursor token management guide, 2026-10-12
MCP claude mcp add --transport http NAME URL adds the server to your own local scope (in ~/.claude.json); --scope project writes the shared .mcp.json instead, which each teammate approves once and which shows as Pending approval until then.MCP overview, Claude Code 2.1.283, 2026-09-26 codex mcp add NAME --url URL writes ~/.codex/config.toml; a project .codex/config.toml is read only when the project is trusted.MCP overview, Codex CLI 0.162.1, 2026-10-12 An MCP client: the docs list MCP among its capabilities. Support for spec 2026-07-28 was not verified on 2026-09-26.Cursor docs, 2026-08-28
Hooks 33 hook events in 2.1.283; handlers can be shell commands, HTTP endpoints, MCP tools, prompts or agents. Exit code 2 blocks a PreToolUse call.Hooks guide, Claude Code 2.1.283, 2026-10-03 12 hook events; a hook needs persisted trust before it runs.Feature matrix, Codex 0.157.1, 2026-09-26 Hooks are spawned processes that exchange JSON over stdio, and they can observe, block or modify behaviour. Cursor’s hooks page lists events such as preToolUse, beforeShellExecution and afterFileEdit (checked 2026-10-12); the config file was not re-verified.Cross-tool hooks guide (Cursor docs), 2026-08-28
Sub-agent Built-in Explore, Plan and general-purpose subagents; custom ones are Markdown files in .claude/agents/. Each runs in its own context window and returns only a summary.Claude Code subagents guide, Claude Code 2.1.283, 2026-09-26 Subagents run inside one session: ask for them in the prompt and switch with /subagents; [agents] in config.toml caps concurrency.Codex multi-agent guide, Codex 0.157.1, 2026-09-26 Subagents are “specialized AI assistants that Cursor’s agent can delegate tasks to”; since 2026-08-19 a cloud subagent can run on its own virtual machine with an isolated copy of the project.Scoped subagents guide (Cursor docs), 2026-08-28
Sandbox Sandboxed Bash: Seatbelt on macOS, bubblewrap on Linux and WSL2, not native Windows. It covers shell commands only; Read, Edit and Write follow permission rules.Permissions and sandboxes guide, Claude Code 2.1.283, 2026-09-26 --sandbox read-only, workspace-write or danger-full-access; permission profiles (beta) are preferred, and the two systems do not compose.Permissions and sandboxes guide, Codex 0.157.1, 2026-09-26 Cursor’s Run modes page lists Auto-review (the recommended default since Cursor 3.6), Allowlist and Run Everything, and says Auto-review is not a security boundary (checked 2026-10-12). Sandboxing can be switched on per run through the SDK (local.sandboxOptions), and /sandbox is listed among the CLI slash commands. The default write boundary and network are not verified.Permissions and sandboxes guide (Cursor), 2026-09-26
Headless mode claude -p runs one task and exits, in Manual mode with nobody to approve; --output-format json returns one result object and --json-schema a validated one.Scripting guide, Claude Code 2.1.283, 2026-09-26 codex exec (alias e) prints the final message to stdout and never asks for approval; --json gives JSONL events and --output-schema FILE a schema-checked answer.codex exec guide, Codex 0.157.1, 2026-10-03 Print mode (-p, --print) in the Cursor CLI, whose command is agent (Cursor CLI docs, checked 2026-10-12), documented for GitHub Actions. Its machine-readable output options were not re-verified.Feature matrix (Cursor docs), 2026-08-28
Plan mode The plan permission mode: Shift+Tab, /plan or claude --permission-mode plan; edits stay blocked until you approve the plan.Agent vs ask mode guide, Claude Code 2.1.283, 2026-09-26 /plan in the TUI switches to Plan mode; approval flags such as -a on-request are a separate control and do not make the agent plan first.Build-stage guide, 2026-10-12 Plan Mode, which "creates detailed implementation plans before writing any code".Cursor docs, 2026-08-28
Git worktrees claude --worktree NAME (-w) creates .claude/worktrees/NAME/ on a new branch worktree-NAME.Worktrees guide, Claude Code 2.1.283, 2026-09-26 codex --worktree or /worktree starts a managed worktree under $CODEX_HOME/worktrees/ on a detached HEAD.Codex worktrees guide, Codex 0.157.1, 2026-09-26 Worktrees “let Agent work in isolated Git checkouts”, and you run and watch parallel agents from the Agents Window. Current setup steps were not re-verified.Level 3 guide (Cursor docs), 2026-08-28
Project instructions
Claude Code
Loaded at launch from the project root and ~/.claude/, and from a subdirectory when Claude reads a file there; the files add up and none overrides another.Project initialization guide, Claude Code 2.1.283, 2026-09-26
Codex
Reads AGENTS.md natively, and /init creates one.Codex AGENTS.md guide, Codex CLI 0.157.1, 2026-09-26
Cursor
Applies instructions through Rules in .cursor/rules/. Cursor’s docs say it reads CLAUDE.md the same way it reads AGENTS.md and always applies it; the agent runtime in @cursor/sdk 1.0.32 (checked 2026-09-26) loads CLAUDE.md and CLAUDE.local.md only when third-party configuration loading is on, while it loads AGENTS.md unconditionally.Cursor token management guide, 2026-10-12
Open the Project instructions entry
MCP
Claude Code
claude mcp add --transport http NAME URL adds the server to your own local scope (in ~/.claude.json); --scope project writes the shared .mcp.json instead, which each teammate approves once and which shows as Pending approval until then.MCP overview, Claude Code 2.1.283, 2026-09-26
Codex
codex mcp add NAME --url URL writes ~/.codex/config.toml; a project .codex/config.toml is read only when the project is trusted.MCP overview, Codex CLI 0.162.1, 2026-10-12
Cursor
An MCP client: the docs list MCP among its capabilities. Support for spec 2026-07-28 was not verified on 2026-09-26.Cursor docs, 2026-08-28
Open the MCP entry
Hooks
Claude Code
33 hook events in 2.1.283; handlers can be shell commands, HTTP endpoints, MCP tools, prompts or agents. Exit code 2 blocks a PreToolUse call.Hooks guide, Claude Code 2.1.283, 2026-10-03
Codex
12 hook events; a hook needs persisted trust before it runs.Feature matrix, Codex 0.157.1, 2026-09-26
Cursor
Hooks are spawned processes that exchange JSON over stdio, and they can observe, block or modify behaviour. Cursor’s hooks page lists events such as preToolUse, beforeShellExecution and afterFileEdit (checked 2026-10-12); the config file was not re-verified.Cross-tool hooks guide (Cursor docs), 2026-08-28
Open the Hooks entry
Sub-agent
Claude Code
Built-in Explore, Plan and general-purpose subagents; custom ones are Markdown files in .claude/agents/. Each runs in its own context window and returns only a summary.Claude Code subagents guide, Claude Code 2.1.283, 2026-09-26
Codex
Subagents run inside one session: ask for them in the prompt and switch with /subagents; [agents] in config.toml caps concurrency.Codex multi-agent guide, Codex 0.157.1, 2026-09-26
Cursor
Subagents are “specialized AI assistants that Cursor’s agent can delegate tasks to”; since 2026-08-19 a cloud subagent can run on its own virtual machine with an isolated copy of the project.Scoped subagents guide (Cursor docs), 2026-08-28
Open the Sub-agent entry
Sandbox
Claude Code
Sandboxed Bash: Seatbelt on macOS, bubblewrap on Linux and WSL2, not native Windows. It covers shell commands only; Read, Edit and Write follow permission rules.Permissions and sandboxes guide, Claude Code 2.1.283, 2026-09-26
Codex
--sandbox read-only, workspace-write or danger-full-access; permission profiles (beta) are preferred, and the two systems do not compose.Permissions and sandboxes guide, Codex 0.157.1, 2026-09-26
Cursor
Cursor’s Run modes page lists Auto-review (the recommended default since Cursor 3.6), Allowlist and Run Everything, and says Auto-review is not a security boundary (checked 2026-10-12). Sandboxing can be switched on per run through the SDK (local.sandboxOptions), and /sandbox is listed among the CLI slash commands. The default write boundary and network are not verified.Permissions and sandboxes guide (Cursor), 2026-09-26
Open the Sandbox entry
Headless mode
Claude Code
claude -p runs one task and exits, in Manual mode with nobody to approve; --output-format json returns one result object and --json-schema a validated one.Scripting guide, Claude Code 2.1.283, 2026-09-26
Codex
codex exec (alias e) prints the final message to stdout and never asks for approval; --json gives JSONL events and --output-schema FILE a schema-checked answer.codex exec guide, Codex 0.157.1, 2026-10-03
Cursor
Print mode (-p, --print) in the Cursor CLI, whose command is agent (Cursor CLI docs, checked 2026-10-12), documented for GitHub Actions. Its machine-readable output options were not re-verified.Feature matrix (Cursor docs), 2026-08-28
Open the Headless mode entry
Plan mode
Claude Code
The plan permission mode: Shift+Tab, /plan or claude --permission-mode plan; edits stay blocked until you approve the plan.Agent vs ask mode guide, Claude Code 2.1.283, 2026-09-26
Codex
/plan in the TUI switches to Plan mode; approval flags such as -a on-request are a separate control and do not make the agent plan first.Build-stage guide, 2026-10-12
Cursor
Plan Mode, which "creates detailed implementation plans before writing any code".Cursor docs, 2026-08-28
Open the Plan mode entry
Git worktrees
Claude Code
claude --worktree NAME (-w) creates .claude/worktrees/NAME/ on a new branch worktree-NAME.Worktrees guide, Claude Code 2.1.283, 2026-09-26
Codex
codex --worktree or /worktree starts a managed worktree under $CODEX_HOME/worktrees/ on a detached HEAD.Codex worktrees guide, Codex 0.157.1, 2026-09-26
Cursor
Worktrees “let Agent work in isolated Git checkouts”, and you run and watch parallel agents from the Agents Window. Current setup steps were not re-verified.Level 3 guide (Cursor docs), 2026-08-28
Open the Git worktrees entry

Each line carries the date it was checked, and each entry page repeats it. Tool behaviour changes often, so a line is only as current as its date.

Where these words get used.

Three free pages that put the vocabulary to work.

The full A to Z glossary in the docs has more entries.

In English and Polish, from Agent Mode to YOLO Mode. It comes with the Toolkit.

Buy the Toolkit Open the complete A to Z

$59.99 once for 12 months of access. Nothing renews on its own. 30-day refund on a first purchase.

Questions about the vocabulary.

Something else? Write to support@developertoolkit.ai.

Is CLAUDE.md the same as AGENTS.md?

Same job, different tools: Claude Code reads CLAUDE.md and Codex reads AGENTS.md. From v2.1.277 (both release channels on 2026-10-12) Claude Code also reads AGENTS.md, but by default only when no CLAUDE.md, .claude/CLAUDE.md or CLAUDE.local.md exists, so adding a CLAUDE.md silently hides your AGENTS.md. Keep the rules in AGENTS.md and make CLAUDE.md import it with @AGENTS.md, or set Project instructions in /config to claude-md-and-agents-md to load both.

Is a sandbox the same as an approval mode?

No. Approval, meaning Claude Code’s permission modes or Codex’s approval policy, decides whether an action runs and who says yes. The sandbox limits what a running command can read, write and reach. Setting approval_policy = "never" in Codex removes the prompts but does not widen or narrow the sandbox, so set both explicitly.

What is the difference between vibe coding and agentic engineering?

The difference is what checks the code. In vibe coding nothing does: the human watches behaviour and accepts the diff unread. Agentic engineering keeps the human accountable for a quality bar, with tests and evals checking where they can. Karpathy’s published transcript of his Sequoia Ascent talk, which says an LLM cleaned it up, puts it as “Vibe coding raises the floor. Agentic engineering is about extrapolating the ceiling.”

Where do the ladder levels come from?

From Dan Shapiro’s “The Five Levels: from Spicy Autocomplete to the Dark Factory” (danshapiro.com, 2026-01-23). In his names Level 2 is the junior developer, Level 3 the developer, Level 4 the engineering team and Level 5 the dark software factory. The level next to a term is our reading of the rung where the free ladder guides first have you set it up. A method sits at its closest rung, and a term the ladder does not hold reads “Off the ladder”.

Why are some Polish names the same as the English?

Product and file names are never translated, so CLAUDE.md, AGENTS.md, MCP and Agent Skills read the same in both languages. Common terms such as plan mode and context window have a Polish name.

Used by engineering teams at

  • Grupa Morizon-Gratka
  • Ringier Axel Springer
  • TasteRay

Creator of wondel.ai Skills — 2.3k+ stars on GitHub (opens in a new tab)

Does this page answer your questions?

One tap, no sign-in.

not at all fully

One meaning for every term.

Read the whole library in the words defined here, with the Toolkit, in English or Polish.

$59.99 once for 12 months of access. Nothing renews on its own. 30-day refund on a first purchase.