Claude Code 2.1.286 stops resume from losing your turns

Claude Code 2.1.286 shipped on September 30 with a fix for resumed sessions coming back with missing turns. If the earlier session crashed or was killed after a batch of parallel tool calls, claude --resume and --continue could restore the conversation without any of the turns that came after that batch.
Most of the rest of the release notes on GitHub are also fixes. They come as a main list plus separate blocks for the VS Code extension, cloud sessions and Claude Tag, the Slack integration. A smaller group of entries changes default behaviour.
Sessions that come back whole
Resumed sessions get a second fix. When a tool or hook returned an object, a number or a boolean instead of text, later requests failed with API 400 errors, and that included resumed sessions. That output no longer causes the errors.
Cloud sessions with very large histories sometimes never woke up, because the container was stopped while the transcript was still loading. A cloud session that had gone idle after Claude sent a message also gave no reply when you answered a question card or approved a tool call. A Remote Control message that arrived while Claude Code was exiting was marked delivered and then never answered. It now stays queued for the session’s next run.
Subagents started by the Workflow tool used to restart from their original prompt when a connection stalled for a few minutes mid-response. In the main window, typing /compact, /clear or /rewind while viewing a background agent’s or teammate’s transcript silently acted on the main conversation. A dialog now names the target and asks first. In Slack, Claude Tag sometimes started a thread over on a fresh machine and lost work it hadn’t pushed, when your reply answered a question it had just asked.
The gateway spend meter
The Claude apps gateway’s spend meter had two pricing bugs. It priced 1-hour prompt cache writes at the cheaper 5-minute rate. And on streamed turns that ran a server-side tool such as web search, it counted only the first model call’s input tokens.
The meter is also what clients see. Since version 2.1.268 in September, clients signed in to a gateway with pricing: set receive the same rates through managed settings, so /cost and telemetry match the spend meter. The /protocol page a gateway serves now matches what Claude Code sends today and tells implementers not to reject unknown input.
Secrets in logs and transcripts
Several fixes close gaps in secret redaction:
- MCP error messages showed a credential’s value when “Bearer” or “Basic” came before its key name.
- Percent-encoded Bearer tokens were only partly masked in error messages.
- Redacted logs and transcripts showed a secret whose key name had an invisible character inside, such as a zero-width space.
- Logs and transcripts showed part of a URL password containing punctuation such as ), quotes, ], & or a second @, or one that ran past a / to a bracketed host such as [::1] in an ssh URL.
- The session transcript in the zip that
/feedbacksaves to disk contained invalid JSON lines after redaction.
On the MCP side, connectors listed no tools for up to a day after their server dropped the older MCP handshake, and a repeat sign-in request from Claude replaced the pending sign-in link, which could stop that link from working.
New defaults
--bare now connects only the MCP servers named on the command line, sends the model no system reminders and starts no background tasks. Under --bare, a shell command that reaches its timeout stops instead of moving to the background.
Send now (ctrl+enter) moves a skill’s own shell command to the background instead of ending it, and in a subagent’s view it backgrounds the running command so your message is read right away. In VS Code, Stop and Escape end only the current turn; background agents keep running and can be stopped one by one from the agent map. If your project or user skills include one named verify, Claude is now told to run it right before committing, except for docs-only and tests-only commits. Plugin installs refuse npm sources that are git repositories or folders and install dependencies only from registry packages.
When several permission requests stack up, the prompt shows a count such as “2 of 5”. /hooks opens on one list of configured hooks grouped by event, so viewing a hook takes one Enter instead of three, and number keys no longer pick a theme or an output style. The VS Code extension adds bookmarks that keep Claude’s responses in a side panel.
Retries and fallbacks
If the Anthropic API refused the model your default or a model alias resolves to, every turn failed. Claude Code now retries once on the previous model of the same tier. Refusal and --fallback-model retries also failed when the fallback model couldn’t run fast; they now run at standard speed, with a one-time notice in interactive sessions. When a fallback drops the context window from 1M to 200K tokens, the fallback notice and the autocompact-thrashing error now say so. Autocompact thrashing is when compaction keeps firing because the conversation no longer fits.
Retry accounting changed as well. One limit now covers a whole model call, so with the default retry settings a failing call sends at most 14 requests.
Up next
FTC plans to make Anthropic and OpenAI executives testify
Per the New York Post, the FTC is drafting subpoena-like demands to force Anthropic and OpenAI executives to testify, right after their CEOs signed a White House pledge to self-regulate.
Related stories
- Claude Code 2.1.288 brings back the prompt you Ctrl+C’d
- Claude Code 2.1.285 lets admins lock down API providers
- Claude Code on desktop folds an iPhone Duo to test layouts
- Claude Code 2.1.318: login expiry warning, runner log change
- Claude Code 2.1.312 stops overwriting big CLAUDE.md files
- Claude Code 2.1.311 compacts past 200K instead of stopping
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

