Anthropic’s Claude Code 2.1.286 brings a sizeable batch of fixes and workflow changes, including stronger redaction of secrets in logs and transcripts. The update also changes how remote-control sessions respond when an organisation switches the feature off, and how background agents behave in the VS Code extension.
Anthropic Watch analysis
What happened
The release notes, published on 30 September, describe changes across Claude Code, its VS Code extension and cloud sessions. For users, several stand out: secrets in MCP errors and saved logs are now more thoroughly masked; remote-control sessions disconnect when organisational policy turns the feature off; and failed model calls now share a retry limit, which the notes say means at most 14 requests with the default settings.
In VS Code, Stop and Escape now end only the current turn, leaving background agents running until stopped separately. The release also fixes cases where resumed sessions lost turns after parallel tool calls, and where remote-control messages could be marked delivered without receiving an answer.
Why it matters
These are the unglamorous fixes that decide whether an agent fits into everyday development or adds a fresh collection of things to supervise. Better secret masking matters when errors and transcripts are saved; clearer remote-control behaviour matters when an organisation changes its rules. The retry limit, meanwhile, gives developers a concrete ceiling to account for when calls fail.
Our read
This is a practical maintenance release with a few changes that deserve attention beyond the version number. Teams using remote control should note the policy change, and developers relying on automatic retries should check whether the default limit suits their workflows. For the rest, fewer lost turns and less accidental secret leakage are welcome improvements. Nobody puts those on a launch poster, which is probably a good sign.
What to watch
- Whether teams need to adjust workflows for the new retry limit.
- How the remote-control policy change behaves in organisations that manage access centrally.
- Whether the fixes reduce lost messages and session-recovery problems in day-to-day use.
Discussion spark: Should coding agents keep background work running when a user presses Stop, or should Stop mean stop everything?
Sources and evidence
- v2.1.286 (30 September 2026, 19:10 UTC)
Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.