Claude Code Draft
Anthropic · reviewed 22 November 2026 · its card
Claude Code lets a developer hand coding work to Claude: it is "an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools" (S1). Anthropic makes it, and its glossary says "Claude Code is the harness; Claude is the model inside it", an agent harness (S2). It runs as a terminal CLI, in VS Code, Cursor and JetBrains, in the Desktop app's Code tab, on the web and in the iOS and Android apps (S1). It is proprietary, used under Anthropic's consumer or commercial terms with a Claude subscription or a Console account (S1, S12). The latest release seen is 2.1.330, of 20 November 2026 (S3). It is a client. Its skills follow the Agent Skills specification. Its maker's own words for it are on the vendor card Claude Code (Anthropic).
- An instance of
- Agent harness, Client
Capabilities
| Capability | Value | Detail | Sources |
|---|---|---|---|
| Model | Yes | Claude models only, through the Anthropic API, Amazon Bedrock, Google Cloud, Microsoft Foundry or Claude Platform on AWS; Anthropic "doesn't support routing Claude Code to non-Claude models through any gateway" | S1 S13 |
| Session | Yes | Sessions resume (claude -c, /resume) or fork (--fork-session); transcripts kept in ~/.claude/projects/; cloud sessions keep running when the laptop closes |
S2 |
| Subagent | Yes | Subagents run in their own context window and return a summary; built-in Explore, Plan and general-purpose, custom ones in .claude/agents/; worktree isolation |
S2 S8 |
| Agent team | Partial | Agent teams: a team lead, a shared task list and mailbox messaging between teammates; experimental, off by default | S6 |
| Skill | Yes | Skills "follow the Agent Skills open standard", with Claude Code's own extras such as context: fork |
S9 |
| Instruction file | Yes | CLAUDE.md at project, user and managed level; AGENTS.md when there is no CLAUDE.md, from v2.1.277; .claude/rules/; auto memory in MEMORY.md |
S2 |
| Plugin | Yes | Plugins bundle skills, agents, hooks and MCP servers, installed from marketplaces | S10 |
| MCP | Yes | MCP client over HTTP, stdio, WebSocket and the deprecated SSE, at local, project and user scope | S11 |
| Execution environment | Yes | harness-sandbox: a Bash sandbox, off by default, Seatbelt on macOS and bubblewrap on Linux and WSL2, not native Windows; worktrees and dev containers; cloud sessions on Anthropic-managed or self-hosted infrastructure |
S7 S2 |
| Client | Yes | Terminal, IDE, desktop, web and mobile; Remote Control continues a local session from a phone or browser; agent view (research preview) shows background sessions | S1 S14 |
| Agent host | Unknown | The docs describe hosting only Claude Code's own sessions, in agent view and its background service | S14 |
| Channel | Partial | Channels: MCP servers that push events into a running session and can reply, such as Telegram, Discord, iMessage and webhooks; research preview | S5 |
| Cross-session messaging | Yes | ListAgents and SendMessage, over a local socket on one machine and through Remote Control to other machines and cloud sessions; v2.1.224 or later |
S4 |
| ACP | Partial | Only through a separate adapter: Anthropic's docs index has no ACP page, and agentclientprotocol/claude-agent-acp drives the Claude Agent SDK "from any ACP client" |
S16 S15 |
| AHP | Unknown | Not found in the docs index | S16 |
| A2A | Unknown | Not found in the docs index | S16 |
| Bot | Unknown | No such concept in the docs; auto memory is per repository, not a named identity. Scheduling: cloud scheduled tasks, Desktop scheduled tasks and /loop |
S2 S1 S14 |
State of play
How things stand, newest first, as the card records it.
- Version 2.1.330: recurring scheduled tasks and
/looprun for up to 30 days, up from 7. S3 - Versions 2.1.328 and 2.1.329: cross-session messaging reports held or dropped messages, and
/taskslists scheduled tasks,/loopwake-ups and forks. S3 - Version 2.1.326, a security-hardening release for the Bash sandbox, bypass mode and PowerShell. S3