Sandbox security hardened and 16 fixes land across hooks, voice mode, and sessions; new /plugin persistent state and custom /model picker entries round out the release.
--resume sessions feel incomplete or conversation history is missing after resuming large sessionsStopFailure hook event that fires when the turn ends due to an API error (rate limit, auth failure, etc.)${CLAUDE_PLUGIN_DATA} variable for plugin persistent state that survives plugin updates; /plugin uninstall prompts before deleting iteffort, maxTurns, and disallowedTools frontmatter support for plugin-shipped agentsANTHROPIC_CUSTOM_MODEL_OPTION env var to add a custom entry to the /model picker, with optional _NAME and _DESCRIPTION suffixed vars for displaygit log HEAD failing with "ambiguous argument" inside sandboxed Bash on Linux, and stub files polluting git status in the working directorycc log and --resume silently truncating conversation history on large sessions (>5 MB) that used subagentsdeny: ["mcp__servername"] permission rules not removing MCP server tools before sending to the model, allowing it to see and attempt blocked toolssandbox.filesystem.allowWrite not working with absolute paths (previously required // prefix)/sandbox Dependencies tab showing Linux prerequisites on macOS instead of macOS-specific info.git, .claude, and other protected directories being writable without a prompt in bypassPermissions mode--worktree flag not loading skills and hooks from the worktree directoryCLAUDE_CODE_DISABLE_GIT_INSTRUCTIONS and includeGitInstructions setting not suppressing the git status section in the system promptANTHROPIC_BETAS environment variable being silently ignored when using Haiku modelsset -g allow-passthrough onsandbox.enabled: true is set but dependencies are missing — now shows a visible startup warning