Skip to main content
The /already skill lets Claude Code pull work from Already sessions and execute it directly in your terminal. Record a voice note, attach a screenshot, or type a task — then run /already and Claude handles the rest.

Prerequisites

  • Claude Code CLI installed (npm i -g @anthropic-ai/claude-code)
  • Already MCP server configured in Claude Code (see MCP overview)
  • A Personal API Key (PAK) or agent API key from Already

MCP server configuration

Add the Already MCP server to your Claude Code config. Put this in ~/.claude.json or your project’s .claude.json:
Replace the key and team ID with values from your Already account settings.

Installation

Copy the skill file to your Claude Code skills directory.
Installs for all projects.
Restart Claude Code after installing. The /already command will appear in your slash-command list.

Usage

Three modes depending on the arguments you pass.

List mode

Lists the 20 most recent sessions with status, priority, title, assignee, and labels. No work is performed.

Auto mode

Picks up the most recent todo session, marks it in_progress, reads all context (notes, comments, attachments), executes the work, then updates the status and adds a completion comment.

Specific mode

Same as auto mode but targets a specific session by its UUID. Use this after running /already list to pick a particular task.

Workflow

What happens when the skill executes work on a session:
1

Fetch context

Reads the session’s title, description, notes (including voice transcriptions), comments, and attachment list.
2

Download attachments

Screenshots, PDFs, and other files are downloaded to /tmp and viewed so Claude has full visual context.
3

Execute work

Based on the gathered context, Claude performs the requested task (writing code, fixing bugs, creating files, etc.).
4

Update status

The session is marked done or needs_feedback and a summary comment is added.

Examples

Fix a bug from a voice note — Record a voice note in Already describing a bug. The session is created with your transcription in the notes. Run /already and Claude reads the transcription, understands the issue, fixes it, and marks the session done with a comment explaining the fix. Work through a backlog — Run /already list to see what’s pending, then /already <id> to pick a specific task, or just /already repeatedly to work through todos in order. Screenshot-driven UI changes — Attach a screenshot to a session with annotations showing what to change. Claude downloads and views the screenshot, then implements the UI changes to match.