Claude Code 2.1.152–2.1.153: Skills Get Richer, /code-review Gets Teeth
Claude Code versions 2.1.152 and 2.1.153 ship a focused set of improvements aimed at skills authors and power users. The most impactful change in 2.1.152 is that /code-review --fix now applies its review findings directly to the working tree — previously the review output was informational only. The updated behavior means correctness bugs and simplification suggestions identified at a chosen effort level are automatically turned into edits, with /simplify now routing through the same path. This transforms code review from a read-only report into an actionable pass.
Skills and slash commands gain a new disallowed-tools frontmatter key. When a skill sets this, the listed tools are removed from the model's available tool set for the duration of the skill's execution — useful for constraining what an agent can do when operating in a specialized mode (for example, a read-only analysis skill that should never be able to write files). The companion /reload-skills command re-scans skill directories without restarting the session, so skills installed mid-session by a hook become available immediately. SessionStart hooks can now also trigger skill reloads by returning reloadSkills: true, closing the automation loop.
Version 2.1.152 also introduces a MessageDisplay hook event that lets hooks transform or suppress assistant message text as it is being displayed — enabling use cases like filtering out verbose tool output for specific workflows, or adding post-processing to Claude's responses before they reach the user. The /model command now persists the selected model as the default for new sessions when pressed without a modifier; pressing s in the picker switches models for the current session only.
Version 2.1.153 (released May 28) fixes a regression where stateful MCP servers without the optional GET SSE stream reconnect-looped on tools/list (introduced in 2.1.147), and resolves a bug where subagent frontmatter MCP servers were ignoring --strict-mcp-config and enterprise managed MCP policies.
OpenAI Codex Goal Mode Reaches GA, Appshots Launches on macOS
OpenAI shipped two significant Codex updates on May 21, 2026. Goal Mode — the long-running autonomous task feature — has graduated from experimental to generally available across the Codex app, IDE extension, and CLI. Instead of sending Codex a discrete instruction and waiting for completion, Goal Mode lets you assign an objective that the agent pursues across session breaks, token budget resets, and interruptions, with progress tracked in dedicated goal storage. The CLI (0.133.0) enables Goals by default, with progress visible across active turns.
Appshots is now available on macOS: pressing both Command keys simultaneously sends the frontmost app window to Codex as a screenshot plus any accessible text content (including content outside the visible scroll area). This makes context transfer from other applications — browser tabs, Figma screens, terminal output, Slack threads — a keyboard shortcut rather than a copy-paste workflow. The feature integrates naturally with Goal Mode, where referencing external application state is a common task. Codex CLI 0.134.0 (May 26) adds history search across local conversation history, per-server MCP environment targeting, and read-only MCP tool concurrency support when the server marks tools as safe for parallel execution.
GitHub Copilot Moves to AI Credits Billing June 1
GitHub Copilot individual plans are being restructured effective June 1, 2026, introducing usage-based billing across Free, Pro, Pro+, and a new Max tier. All paid plans shift from flat-rate access to a hybrid model combining base credits (matched dollar-for-dollar to the subscription price) and a flex allotment (a variable buffer that adjusts as AI model economics change). Code completions and next-edit suggestions remain unlimited on paid plans without consuming credits.
The plan tiers break down as follows: Pro at $10/month includes $10 base credits plus a $5 flex allotment ($15 total monthly usage); Pro+ at $39/month includes $39 base plus $31 flex ($70 total); Max at $100/month provides $100 base plus $100 flex ($200 total). One AI credit equals $0.01 USD, with consumption calculated from input, output, and cached tokens at published API rates per model. Existing Pro and Pro+ monthly subscribers will receive the expanded usage automatically without any action required. The flex allotment is designed to absorb model pricing changes — as frontier models get cheaper to run, the flex buffer can grow without a subscription price change.
