AI Aulis

AI NEWS IN ENGLISH

The most important AI news every morning. AI writes and checks the stories against the original sources – check important facts in the sources.

Claude Code and Codex: the 30 most important features

The key features and /commands of two coding agents: what each one does, when it helps and from which version. Every item links to the maker's own documentation or changelog.

This week's changes in the tools

One fact per row, with its source. The rows are compiled overnight from our archive, and a row is shown only after its source and version number have been checked. Updated 19 November 2026.

  • Claude Code 2.1.327: /status lists hook matchers that never match. (github.com, 19 November 2026)
  • Claude Code / MCP 2.1.327: claude mcp add-json - can read the server's JSON settings from standard input outside a Claude Code session. (github.com, 19 November 2026)
  • Claude Code 2.1.327: A notification was added to resumed sessions for when previous messages cannot be loaded. (github.com, 19 November 2026)
  • Claude Code 2.1.327: /install-github-app asks for confirmation before modifying the repository. (github.com, 19 November 2026)
  • Claude Code v2.1.326: Fixed an issue with skills and slash commands loading in Code Mode's auto mode when an inline ! command requires approval. (github.com, 18 November 2026)
  • Claude Code v2.1.326: According to the release notes, Code Mode can be enabled while the agent teams feature is enabled. (github.com, 18 November 2026)
  • MCP PHP SDK v0.10.0: The SDK has been split into the Mcp\Schema, Mcp\Client and Mcp\Server namespaces, which changes most imports. (github.com, 17 November 2026)
  • MCP PHP SDK v0.10.0: HTTP sessions belong to the authenticated party that opened them. (github.com, 17 November 2026)

Claude Code

/compact

What it does: The command condenses the conversation to make room in the model’s context window.

When it helps: It is useful when a long conversation fills the context but the task needs to continue.

source: code.claude.com

/clear

What it does: The command begins a fresh conversation without carrying over the previous conversation’s context.

When it helps: It is useful when moving to a new task that does not need the earlier discussion.

source: code.claude.com

/model

What it does: The command switches the active model or opens the model picker.

When it helps: It is useful when choosing a model that suits the task’s speed and reasoning requirements.

source: code.claude.com

/review

What it does: The command checks code changes for bugs and runs the same review as /code-review.

When it helps: It is useful for finding problems before sending changes to others for review.

source: code.claude.com

/init

What it does: The command examines the project and drafts a CLAUDE.md file or suggests improvements to an existing one.

When it helps: It is useful when introducing Claude Code to a project whose working instructions have not been collected.

source: code.claude.com

CLAUDE.md-muisti

What it does: A CLAUDE.md file gives Claude persistent guidance about project conventions and working practices.

When it helps: It is useful for carrying test commands and coding conventions across sessions.

source: code.claude.com

Hookit (hooks)

What it does: Hooks trigger automatic actions at defined events, such as after a file edit.

When it helps: They are useful for automatically formatting code or sending a notification.

source: code.claude.com

Skillit (skills)

What it does: Skills package reusable instructions that Claude applies when relevant or when explicitly invoked.

When it helps: They are useful when the same checklist or multi-step procedure recurs across tasks.

source: code.claude.com

Aliagentit (subagents)

What it does: Subagents perform focused tasks in separate contexts and return their results to the main conversation.

When it helps: They are useful when extensive research or log inspection would clutter the main conversation.

source: code.claude.com

MCP

What it does: MCP connects Claude Code to external tools and data sources through a shared protocol.

When it helps: It is useful when a task needs information from an issue tracker, database or monitoring service.

source: code.claude.com

Käyttöoikeudet (permissions)

What it does: Permission rules determine which tool actions are allowed, blocked or require user approval.

When it helps: They are useful for controlling file changes and command execution according to the project’s needs.

source: code.claude.com

Suunnittelutila (plan mode)

What it does: In plan mode, Claude examines the project and proposes a plan before implementation is approved.

When it helps: It is useful for assessing the approach to a substantial change before editing code.

source: code.claude.com

/resume

What it does: The command reopens an earlier session from within the active session.

When it helps: It is useful when returning to unfinished work or switching between saved conversations.

source: code.claude.com

claude -p

What it does: The command executes a supplied task without an interactive chat interface and prints the response.

When it helps: It is useful for integrating Claude Code into scripts, data pipelines and automated builds.

source: code.claude.com

Vastaustyylit (output styles) since version 1.0.81

What it does: Output styles change how Claude Code responds, including explanatory and educational approaches.

When it helps: They are useful when you want explanations alongside code changes or opportunities to practise programming.

source: code.claude.com

Tilarivi (statusline) since version 1.0.71

What it does: A customizable status line adds your own status display to Claude Code.

When it helps: It is useful for keeping essential working information visible throughout a session.

source: code.claude.com

Codex

AGENTS.md

What it does: Codex loads persistent working instructions from AGENTS.md files before starting a task.

When it helps: It is useful for providing project test commands and coding conventions whenever work begins.

source: developers.openai.com

codex exec

What it does: The command runs a task without an interactive interface for use in scripts and automated workflows.

When it helps: It is useful for automating continuous integration checks or generating release notes.

source: developers.openai.com

Sandbox-tilat

What it does: Sandbox modes define file and network access boundaries for commands run by Codex.

When it helps: They are useful for restricting a task to reading files or editing the workspace.

source: developers.openai.com

Hyväksyntäkäytäntö (approvals)

What it does: The approval policy determines when Codex needs approval before performing an action.

When it helps: It is useful for retaining control over actions such as changes outside the workspace.

source: developers.openai.com

/review

What it does: The command reviews selected Git changes and prioritizes findings without modifying the working tree.

When it helps: It is useful for finding problems before submitting a pull request.

source: developers.openai.com

MCP

What it does: MCP gives Codex access to tools and information in other systems through server connections.

When it helps: It is useful for retrieving issue-tracking information or documentation directly while working.

source: developers.openai.com

Asetusprofiilit (profiles)

What it does: Profiles group settings into separate TOML files selected with --profile.

When it helps: They are useful for switching between configurations suited to coding and code review.

source: developers.openai.com

Verkkohaku (web search)

What it does: Web search brings online information into a task, and --search enables live results for that run.

When it helps: It is useful when a solution depends on the latest documentation or release information.

source: developers.openai.com

codex resume

What it does: The command reopens a saved conversation so work on an earlier task can continue.

When it helps: It is useful when returning to unfinished work after a break or closing the CLI.

source: developers.openai.com

/model

What it does: The command changes the model or its reasoning effort during an interactive CLI session.

When it helps: It is useful for matching response speed and reasoning depth to the task.

source: developers.openai.com

/compact

What it does: The command condenses the chat to reclaim space in the context window.

When it helps: It is useful for continuing a long session while retaining the conversation’s key points.

source: developers.openai.com

config.toml

What it does: config.toml stores Codex settings such as the default model, approval policy and MCP servers.

When it helps: It is useful for sharing the same defaults between the CLI and IDE extension.

source: developers.openai.com

IDE-laajennus

What it does: The extension brings Codex into the code editor and uses open files and selected code as context.

When it helps: It is useful for editing and reviewing code in the same view.

source: developers.openai.com

/status

What it does: The command reports session settings, including the active model and approval policy, plus token consumption.

When it helps: It is useful for checking the working configuration and remaining context capacity.

source: developers.openai.com

How this page is made

The feature list is compiled from the makers' official documentation, and every link has been checked. The weekly changes are updated every night: Codex reads the archived sources on Claude Code, Codex and MCP, and a row is accepted only if its source is in the archive and the version number appears in the source text. Only the makers' own pages and repositories count as sources.

checked 19 November 2026