2 min read AI-generated

Claude Code 2.1.247: A Feedback Draft and a Cost Check for the API

Copy article as Markdown

The new version brings a SendFeedback tool that prepares an error report automatically when something breaks, plus a cost optimizer for the Claude API and the usual long list of bug fixes.

Featured image for "Claude Code 2.1.247: A Feedback Draft and a Cost Check for the API"

Next item in the rapid cadence: Claude Code 2.1.247 is here. No big overhaul, but a few additions that smooth out the day-to-day.

The most noticeable new feature is the SendFeedback tool. When something goes wrong in a session, Claude can now prepare a feedback report for you. You look it over and send it via /feedback. Don’t want it? Turn it off with the feedbackDrafts setting. Handy, because you often report a bug at exactly the moment you have no patience to write up the context by hand.

For anyone working heavily through the API, /claude-api cost-optimize is the more interesting addition. The command takes an existing project, looks at its API spend, and works through the cost levers one by one — caching, token hygiene, batching, effort level, model choice. Always one measured change at a time, instead of blindly turning several knobs at once. On top of that, the /claude-api skill now covers the Admin API too: members, invites, workspaces, API keys, rate-limit reports, and more.

Organizations also get more control over the little tips that show up in the spinner while you wait. Through spinnerTipsOverride and a custom tipsFile, you can drop in your own hints. And on Bash permission prompts there’s now a nudge toward auto mode — including a one-keystroke “Yes, and switch to auto mode” option.

The rest is detail work, but the right kind. Sub-agents no longer die on a 404 during the first model call; they fall back to the session’s model chain instead. A hook that spews megabytes of error output can no longer overflow the conversation and wedge the session on “Prompt is too long.” And the Bash sandbox no longer accidentally deletes a dotfile-managed settings.json symlink. Fixes like these never make the marketing sheet — but they’re exactly what decides whether a tool grates on you in daily use.

None of it is spectacular. But the sum of cost transparency, less friction when reporting bugs, and a stack of quiet fixes is exactly what a tool you touch every day needs.

Sources: Claude Code Changelog · @anthropic-ai/claude-code on npm