Claude Code Daily Briefing - 2026-02-28

Release Summary

VersionDateKey Changes
v2.1.632/28/simplify & /batch commands, HTTP hooks, worktree config sharing, massive memory leak fixes
v2.1.622/27Server-side KV cache improvement (caused P1 regression — stale context after compaction)

Full Release Notes

New Features & Practical Usage

1. HTTP Hooks — Send JSON to URLs Instead of Running Shell Commands (v2.1.63)

HTTP hooks have been added as a new hook type. Previously, hooks could only execute local shell commands. Now you can POST JSON to a specified URL and receive a JSON response. This enables integrations with Slack/Discord notifications, external CI/CD triggers, and custom audit log servers — all without writing shell scripts.

// .claude/settings.json example
{
  "hooks": {
    "PostToolUse": {
      "url": "https://your-server.com/webhook",
      "method": "POST"
    }
  }
}

GitHub Releases

2. /simplify & /batch — Bundled Slash Commands (v2.1.63)

Two new bundled slash commands have been added. /simplify refactors complex code into cleaner, more concise versions. /batch handles repetitive operations across multiple files in one go. Tasks that previously required lengthy prompts can now be triggered with a single command.

/simplify  # Refactor current context code for clarity
/batch     # Run repetitive operations across multiple files

GitHub Releases

3. Project Configs & Auto-Memory Shared Across Git Worktrees (v2.1.63)

Git worktrees from the same repository now share project configurations and auto-memory. Previously, each worktree had isolated settings, requiring developers to maintain separate CLAUDE.md files, auto-memory, and project configs for each worktree. Now, context learned in one worktree automatically applies to all worktrees of the same repository. This is particularly useful for parallel multi-branch workflows.

GitHub Releases

Developer Workflow Tips

v2.1.62 Stale Context Regression — Caution for Long Sessions

v2.1.62 (deployed 2/27) introduced a server-side KV cache improvement that increased cache hit rates but missed compaction-event invalidation. After context compaction, the model operates on stale cached state — falsely claiming files don’t exist, continuing pre-compaction task plans despite explicit user redirection, and requiring 3-5 interruptions instead of 1.

Mitigation:

  1. Update to v2.1.63 immediately — today’s release includes related fixes
  2. Reset long sessions periodically with /clear — prevents compaction from occurring
  3. Start a new session if the model behaves erratically — stale cache doesn’t affect fresh sessions

GitHub Issue #29230

v2.1.63 Fixes 8+ Memory Leaks — Major Stability Improvement for Long Sessions

v2.1.63 ships a sweeping batch of memory leak fixes that directly impact anyone running Claude Code for extended periods.

Fixed leaks include:

These fixes are especially impactful for users running subagents, teammates, or long-running sessions.

GitHub Releases

Security & Limitations

Trump Orders All Federal Agencies to Stop Using Anthropic — Supply Chain Risk Designation (2/27)

The Pentagon-Anthropic standoff escalated dramatically after yesterday’s deadline. President Trump ordered every federal agency to “immediately cease” using Anthropic technology, and the Pentagon officially designated Anthropic as a “supply chain risk” — a classification typically reserved for adversary nations. Defense contractors are now barred from deploying Anthropic AI in Pentagon work. Existing agency users have been given a 6-month phaseout period.

CEO Dario Amodei published a statement titled “Statement from Dario Amodei on our discussions with the Department of War,” reaffirming that while Anthropic has actively deployed Claude for intelligence analysis, operational planning, and cyber operations, its two red lines remain non-negotiable: fully autonomous weapons and mass domestic surveillance. “These threats do not change our position: we cannot in good conscience accede to their request,” Amodei stated.

Anthropic has announced it will legally challenge the supply chain risk designation.

Anthropic Statement | NPR | CNBC | Washington Post

Ecosystem & Plugins

Opt Out of claude.ai MCP Servers (v2.1.63)

A new environment variable ENABLE_CLAUDEAI_MCP_SERVERS=false lets you disable MCP servers provided by claude.ai in Claude Code. Useful for enterprise environments that restrict tools to approved MCP servers only, or for reducing unnecessary tool loading to improve context efficiency.

# Disable claude.ai MCP servers
export ENABLE_CLAUDEAI_MCP_SERVERS=false

MCP OAuth Manual URL Paste Fallback (v2.1.63)

When automatic localhost redirect fails during MCP server OAuth authentication, you can now manually paste the callback URL to complete authentication. This resolves authentication failures in SSH remote environments or setups with restricted port forwarding.

GitHub Releases

Community News

Minor Changes Worth Knowing