Claude Code Projects opens to Pro and Max; Haiku 5.5 ships

The week's changes cluster around two concerns: who gets access, and what stops a misbehaving rule. Claude Code Projects moved out of waitlist status for Pro and Max on October 10, but Projects is not available to Team and Enterprise, and it cannot be reached from the CLI, VS Code or JetBrains — only claude.ai/code and the desktop and mobile apps. New projects default every thread to Opus at high effort, which consumes plan limits faster.

On the pricing side, Haiku 5.5 landed on October 7 with a 1M-token context and 128k maximum output, priced at one-tenth of Haiku 4.5 for prompts up to 100k tokens. Anthropic continues to recommend Sonnet 5.5 and Opus 5.5 for complex agentic coding, positioning Haiku 5.5 for narrower work such as compaction, summarization and subagents. Sonnet 5.5's cache-read price also fell that same day, from $0.20 to $0.10 per million tokens.

The patch sequence from 2.1.289 to 2.1.296 reads as a sweep of enforcement gaps: environment-variable prefixes and variable assignments bypassing Bash deny/ask rules, symlink and image-path reads slipping past Read deny, and hooks that failed open. In 2.1.292, one entry is explicitly marked "Security" — PreToolUse hook approvals and auto mode bypassing the permission prompt for file reads from network (UNC) paths. The new onFailure option, available from 2.1.295, lets a failing or timing-out hook block the action rather than let it proceed.

Some fixes will produce more prompts: wildcard-expanded arguments in rg and git grep, pyright, some ps forms, and certain commands that assign to BASH_ARGV0. Anthropic's Usage Policy update, announced October 8 and effective November 12, is described as mostly a clarification of existing rules, but adds that users are responsible for the actions of agents they build or deploy.