You said you want shared terms for all of this. These are the ones the site uses, so that when a page says "channel" it means one specific thing. The first group is the newer vocabulary. The second is the established set, kept to one line each.

## Newer

**Agent view.** `claude agents`. Every running session on one screen, with the ones that need your input marked. [Agent view](https://code.claude.com/docs/en/agent-view)

**Cross-session messaging.** Sessions on the same machine, or your sessions on other machines, can list and message each other. [Cross-session messaging](https://code.claude.com/docs/en/cross-session-messaging)

**Dynamic workflow.** A script Claude writes that orchestrates many subagents, rerunnable. For audits, migrations, cross-checked research. [Workflows](https://code.claude.com/docs/en/workflows)

**Channel.** An MCP server that pushes messages from Telegram, Discord, or iMessage into a running session and carries the reply back. Needs claude.ai auth, so not on Bedrock. [Channels](https://code.claude.com/docs/en/channels)

**Routine.** A scheduled or event-triggered run in the cloud, so it runs when your Mac is off. Distinct from desktop scheduled tasks, which run on your machine. [Routines](https://code.claude.com/docs/en/routines)

**Auto memory.** Notes Claude writes for itself about your corrections and preferences, per repository, loaded every session. Separate from CLAUDE.md, which you write. [Auto memory](https://code.claude.com/docs/en/memory#auto-memory)

**Prompt cache.** The cached prefix of a session. A model switch or an effort change starts a fresh cache and forces a full uncached turn. [Prompt caching](https://code.claude.com/docs/en/prompt-caching)

**Fork.** A subagent that inherits the parent's whole conversation instead of starting clean. [Subagents](https://code.claude.com/docs/en/sub-agents#fork-the-current-conversation)

**Output style.** A system-prompt swap. Concise is the one you want. [Output styles](https://code.claude.com/docs/en/output-styles)

**Goal.** `/goal`. A completion condition a separate evaluator re-checks after every turn. [Goal](https://code.claude.com/docs/en/goal)

## Established

**Subagent.** A separate Claude with its own context and tool list, defined in `.claude/agents/`, reporting a summary back. [Subagents](https://code.claude.com/docs/en/sub-agents)

**Skill.** A `SKILL.md` that packages a workflow or domain knowledge, loaded on demand or invoked as `/name`. [Skills](https://code.claude.com/docs/en/skills)

**Hook.** A shell command at a fixed lifecycle point. Runs every time, unlike CLAUDE.md. [Hooks](https://code.claude.com/docs/en/hooks-guide)

**Compaction.** Summarising older conversation to free context. `/compact <instructions>` to steer it. [Context window](https://code.claude.com/docs/en/context-window)

**Headless.** `claude -p "..."`, no interactive prompt, for scripts and schedulers. [Headless](https://code.claude.com/docs/en/headless)

**Worktree.** A second checkout of the same repo so two sessions do not collide. `--worktree`. [Worktrees](https://code.claude.com/docs/en/worktrees)

**Permission mode.** How much Claude asks before acting. Auto mode is the default on paid claude.ai plans; Manual is the default on Bedrock. [Permission modes](https://code.claude.com/docs/en/permission-modes)

**launchd.** macOS's service manager. Starts things at login, restarts them when they die. The reason an agent survives a reboot. [launchd.info](https://www.launchd.info/)

**Tailnet.** Your private Tailscale network. [Tailscale](https://tailscale.com/kb/1151/what-is-tailscale)