№ 2002026-11-16

Claude Report — 2026-11-16


🚀 Anthropic Official

Claude Code v2.1.325 (Nov 13) — permission and sandbox hardening

  • Sandbox and permissions:
    • Sandboxed Bash commands now follow the normal permission flow by default. Set sandbox.autoAllowBashIfSandboxed: true to keep running them without prompts.
    • Read-only git commands now ask first in a repository that isn't the project's or a trusted one. Name it with --add-dir to skip the prompt.
    • Scheduled tasks and /loop ask before running a skill that declares allowed-tools or hooks. Add Skill(name) to permissions.allow to run it unattended.
    • The sandbox refuses writes to .claude settings, skills, agents, commands and hooks in added directories.
    • Isolated worktree sessions and agents are refused git bisect run, git maintenance register|start and scalar set-up commands.
    • Sandbox-launched programs now run without Claude Code's sign-in tokens and API keys.
    • Fixes several gaps: compound Bash commands that ran without asking, allow rules that ran a different program than the one named, deny and ask rules that did not apply to some commands, and relative rules such as Read(./.env) that stopped covering the project after a cd.
    • Fixes a zstd-compressed response decoding issue.
  • Plugins:
    • Plugin MCP bundles (.mcpb, .dxt) now unpack under the plugins directory instead of .mcpb-cache.
    • Plugins in prependPlugins or appendPlugins count as the organization's for remote marketplaces that managed settings declare. pluginDigests is deprecated and ignored.
    • Plugin loading skips some unsafe paths and links, and claude plugin validate reports them.
    • .claude/skills folders in --plugin-dir plugins no longer load from a session started above them.
    • Marketplace ref accepts the full ID of any commit on a branch or tag. Clients before this version can't clone such a pin.
    • Plugin marketplace and plugin downloads use a CONNECT tunnel when ALL_PROXY is the only proxy set.
    • Plugin prompts and warnings now say that plugins can run code on your computer.
  • Subagents and context:
    • Subagents cut off at the output token limit are asked for the whole answer again, more briefly.
    • Fixes subagents with their own model being moved to the plan-mode model when the setting was opusplan or haiku.
    • CLAUDE_CODE_MAX_CONTEXT_TOKENS now also sets the context window for unrecognized model IDs starting with claude-.
    • Fixes sessions on many third-party gateways and Foundry stopping with a context length error every turn instead of compacting.
    • Fixes a saved sonnet[1m] model setting being rewritten to a model ID the API rejects.
  • MCP: Claude is told when an MCP task is waiting for your answer and can point you to /tasks.
  • VS Code:
    • Adds prompt suggestions: Tab accepts the next one, and promptSuggestionEnabled: false turns them off.
    • Adds a Console sign-in that creates no API key.
    • Adds a field to paste the redirect URL when signing in to an MCP server in SSH, WSL and container windows.
  • Other:
    • /cost, --max-budget-usd and the cost and token metrics now count a mod's $.model.complete and $.model.classify calls.
    • claude remote-control exits with status 4 when it stops by itself and 69 when it can't register.
    • Fixes recent turns going missing from a resumed session after an unclean exit.

🔌 Claude Code Plugins

Plugin tracking skipped: plugin cache unavailable this run.

🛠️ Skills

No relevant news today.

🤖 Agents & Subagents

No relevant news today.

🔗 MCPs & Integrations

No relevant news today.

💡 Community — Workflows & Ideas

No relevant news today.

📰 Quick Mentions

  • Claude API release notes — The newest entry is the Nov 12 self-hosted Managed Agents launch. Nothing is dated Nov 13–16.
← back to archiveend of edition № 200