Claude Code v2.1.247: Cost Optimizer & Fixes

Claude Code v2.1.247 ships with a new SendFeedback tool, a CLI command to profile API spend, and a broad set of fixes affecting sub-agents, keyboard shortcuts, and cloud sessions. Here's what changed.

New: SendFeedback tool and feedback drafts

When something goes wrong in a session, Claude can now draft a feedback report for you to review and send from /feedback. You can turn this off with the feedbackDrafts setting.

New: /claude-api cost-optimize

Run /claude-api cost-optimize to profile an existing project's Claude API spend and work through cost levers—caching, token hygiene, batch, effort, model choice—one measured change at a time. This is a practical tool for anyone looking to reduce API costs incrementally.

Admin API coverage in /claude-api skill

The /claude-api skill now covers Admin API endpoints: organization members, invites, workspaces, API keys, rate limit reports, workload identity federation, and CMEK.

Spinner tips customization

Organizations can now rotate their own tips alongside built-in ones via spinnerTipsOverride, with {id, text, cooldownSessions, priority} entries, a tipsFile, and a label. Also added a tip on Bash permission prompts that offers a one-keystroke "Yes, and switch to auto mode" option.

Key bug fixes

Sub-agent fallback on model 404: Sub-agents now use the session's fallback model chain, and the error returned to the parent includes error type, status, request id, and model.
Fixed fast arrow-key + Enter sequences acting on the wrong row in history search, /config, /mcp, /skills, background tasks, and /model.
Keyboard shortcuts under non-Latin layouts: Ctrl shortcuts now work with Cyrillic and other layouts in kitty-protocol terminals.
Bash sandbox symlink preservation: Fixed cleanup deleting a dotfile-managed ~/.claude/settings.json symlink (nix/home-manager, stow).
/terminal-setup now merges into Zed's keymap.json instead of overwriting it.
/rename now warns when the session registry could not be updated.
/compact and "Summarize from here" in --agent sessions now use the conversation's own system prompt.
Background session fix: Rows now show failure reason instead of "opening…" forever when the terminal host dies.
Memory leak fixed: Unbounded growth when hook/background task output files can't be written—now notes where output was lost.
Cloud sessions: Fixed permission mode switch issues, and now report lost work when container restarts.
Remote Control sessions now report working-tree diff to connected clients.

Improvements and platform updates

Plugin marketplace hardening: Rejects names with control/invisible characters; escape-safe output.
MCP server connection errors: Claude is now told when a configured MCP server fails to connect (Bedrock, Vertex, Foundry, telemetry-disabled sessions).
Sonnet 5 auto-compact: Default now at full 1M context—auto-compact triggers at ~967K.

This release is substantial for teams running Claude Code in production—especially the cost optimizer and the sub-agent fallback fix. Check the full changelog for details.

📖 Read the full source: GitHub Claude-Code