Claude Code /goal Command: Set Completion Conditions and Let Claude Run Autonomously
Claude CodeClaude Code v2.1.139 introduced the /goal command, allowing users to define a specific completion condition and have Claude continue working autonomously across multiple turns until that condition is met. A live overlay panel tracks elapsed time, turn count, and token usage throughout the run, giving users visibility without requiring intervention. The command works in interactive mode, --print (-p) flag mode, and Remote Control.
Sources & Mentions
4 external resources covering this update
Claude Code /goal: Define Done, Then Let Go
Claude Code v2.1.139, released on May 11, 2026, adds the /goal command — a new way to hand off a task to Claude with a clearly defined finish line. Instead of supervising each turn or manually re-prompting when Claude pauses, users set a natural-language completion condition and Claude continues running autonomously until that condition is satisfied.
How /goal Works
After entering /goal and describing the desired end state — for example, "all failing tests pass" or "the feature branch is merged and CI is green" — Claude Code takes over. It works through as many turns as needed, making decisions and executing actions, while a live overlay panel remains visible in the terminal showing:
- Elapsed time since the goal was set
- Turn count — how many back-and-forth exchanges have occurred
- Token usage — a running tally of context consumed
This overlay gives users a lightweight view of progress without requiring them to track the session manually.
Where It Works
The /goal command is compatible across all three major Claude Code execution modes:
- Interactive mode — standard terminal sessions
-p/--printflag — non-interactive, scripted runs- Remote Control — sessions managed via the Remote Control API
This broad compatibility means /goal integrates into both human-in-the-loop workflows and fully automated pipelines.
Why This Matters
The /goal command addresses a fundamental friction point in agentic coding: knowing when to let Claude keep going. Without a declared goal, users must actively monitor sessions and decide when to re-engage. With /goal, the decision logic is explicit from the start — Claude checks its output against the condition and keeps going until it's satisfied or until the session hits a natural stopping point.
Combined with agent view (also introduced in v2.1.139), /goal enables a powerful pattern: dispatch several sessions with explicit goals, background them via /bg, and return to agent view to see which ones have completed.