Claude Code adds hooks, security hardening

Anthropic is turning Claude Code from a smart terminal assistant into a more governable platform: plugins get durable state, hooks get cleaner failure signals, and the security model gets less leaky.

• –`StopFailure` gives hooks a deterministic way to react when an API error ends a turn, which should make retries and guardrails less brittle.
• –`${CLAUDE_PLUGIN_DATA}` makes plugin state survive updates, and the uninstall prompt is a good sign that persistence is being treated as a user-visible security boundary.
• –Fixes for `deny` rules, protected directories, and absolute-path writes are the most consequential changes here; they close holes that matter once people trust an agent with real repos.
• –Line-by-line streaming, tmux passthrough, and resumable large sessions make the tool feel more responsive in day-to-day terminal use.
• –Voice-mode, WSL2, and VS Code fixes round out the release by smoothing the rough edges that power users actually hit.