AIToday
Large Language ModelsAI Coding AssistantsZenn AI/MLPublished: Oct 17, 2026, 10:00 JST

/goal run ends only on ✔ Goal achieved check

/goal run ends only on ✔ Goal achieved check

3 Key Points

  1. What happened

    The author walked through /goal, output styles, /keybindings, debugging settings via /status, and commit attribution; the /goal demo stopped with "✔ Goal achieved (36s · 3 turns · 2.8k tokens)" only after all three issues and pytest exit 0.

  2. Why it matters

    The condition is checked each turn by a separate small model that judges from what appeared in the conversation, so results must be surfaced in the chat — a practical constraint for anyone handing off multi-step work.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Context & Analysis

The piece is a hands-on walkthrough rather than an announcement. Each of the five items was tried in a small practice repository, and the author reports what came back on screen. For /goal, the demonstration deliberately limited the run to one fix per turn so the segmentation was visible; the evaluating model returned "Goal not yet met… continuing" after the first fix, then accepted the run after three fixes with 3 passed and exit=0. The author adds that /goal does not change permission mode, so unattended runs should use auto mode.

The settings section is the most operational. When a setting doesn't take effect, /status lists the files that were read under "Setting sources"; if the file is absent it wasn't read, if present but wrong you should suspect an override (local > project > user), and a ⚠ points to a formatting error that claude doctor can identify. In the worked example, a local file with an array matcher sat alongside the team's settings and produced a warning; deleting the model line and changing the matcher to the string "Edit|Write" restored the team model and cleared the warning.

On commits, the default Co-Authored-By line added by Claude Code is separate from the key signature. Emptying the attribution entries removes it, while subject-line conventions go in CLAUDE.md — which the author notes is not guaranteed to be followed, so anything that must always run belongs in hooks.

FAQ
How does /goal decide when to stop?
After each turn, a separate small model checks whether the end condition is met; if not, Claude starts another turn, and if met, it stops. The checking model judges only from what appeared in the conversation, so results need to be shown in the chat.
Does pressing Esc cancel a /goal?
No. Esc only stops the current work and leaves the goal in place; the way to remove it is /goal clear.
How do I remove the Co-Authored-By line from Claude Code commits?
Set "commit" and "pr" to empty strings under "attribution" in .claude/settings.json; empty removes them and text replaces them. The author confirmed the signature line disappeared while a feat: prefix was added via CLAUDE.md.

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleClaude Opus 5.5 multi-agent tips: let managers delegate