Follow Discord
Sweep 08 Oct 2026 · 18:53Z Build v2.1.295 516 read Stable v2.1.286 Latest v2.1.295 Next v2.1.295 Feeds RSS JSON llms.txt llms-full.txt Unofficial

Claude Code v2.1.292 ·

Protocol adds safety_stops counts, worker_epoch on interrupts and safeguard verdicts on turn handoffs

SDK results gain a safety_stops count, interrupts gain an internal worker_epoch, and turn handoffs can carry server-side safeguard verdicts

Group of 4 Use it now No documentation found New Features
JSON All of v2.1.292
Use it nowTier: how much it should matter to you
4Useful: my rating, 1 to 5
4Signal: worth watching, 1 to 5
SDKArea: what it touches
New FeaturesKind: in v2.1.292,
What probably matters to youSection of the release

What

The messages Claude Code exchanges with programs that host or control it gain several optional fields.

  • safety_stops: SDK result messages now carry a count kept by a new counter in the cost and usage tracker. It appears on every result type that shares the common fields. When a result already has safety_stops and the tracked cost is lower than the result's total_cost_usd, the merged value is the larger of the two counts.
  • worker_epoch: an internal field on the interrupt request, so an interrupt sent by the service cannot hit a later worker (the remote process running the session). A stale one can be refused with stale_worker_epoch. It is stripped from the interrupt event, as for_user_message_uuids already was.
  • The next_abort message to a remote worker now carries extra detail fields alongside the reason.
  • safeguard_results and safeguard_tool_use: new fields on the turn_handoff request, so a server-side classifier verdict can decide the single tool call to run.
  • run_id: the task notification message schema gains this optional field, filled in when sent, and background task records gain runId.

Why

Programs built on the SDK can see how many safety stops happened, with the count no longer under-reported, and can match notifications to a run. Remote sessions avoid a late interrupt stopping the wrong worker, and the server can now rule on a handed-over tool call.

Read from
Names in the bundlesafety_stops

See this entry in the whole of v2.1.292 →

Feedback