2026-07-03 · Anthropic

v2.1.201

securitymodels

read at source ↗ github.com

v2.1.201

Source: Anthropic Claude Code Date: 2026-07-03 URL: https://github.com/anthropics/claude-code/releases/tag/v2.1.201

Summary

Single-line patch: Claude Sonnet 5 sessions no longer use the mid-conversation system role for harness reminders. A model-specific harness-hygiene fix — Sonnet 5 evidently mishandles or over-weights reminders injected mid-conversation under the system role, so those reminders now arrive via a different channel for that model.

Implications

Feeds agent-layer convergence and the two-clocks (closed) thread. Corroborates that Sonnet 5 (GA’d 07-01) is now the live default the harness is being tuned around — the harness is accumulating per-model quirks (cf. v2.1.201’s Sonnet-5-specific reminder handling), a sign the model landed as the working default fast. Minor on its own; tracked as part of the CC changelog arc.

← all signals