Claude Code v2.1.280 Ships Opus 5.5 as Default With 1M Context

Claude Code v2.1.280 is out, and the headline change is a model swap: claude-opus-5-5 is now the default Opus model. Anthropic lists it as 1M context at $4/$20 per Mtok, with cache reads at $0.20/Mtok.

New env var for MCP descriptions

MCP tool descriptions and server instructions were previously capped at 2,048 characters per server. You can now override that with CLAUDE_CODE_MAX_MCP_DESCRIPTION_LENGTH, which applies to every MCP server in the session.

Symlink write handling

Writes through a symlinked path were being judged by their in-tree spelling. That's fixed: the prompt now names where the write actually lands, and acceptEdits, allow rules, and auto mode no longer approve a write that lands outside the tree.

Auto mode stops looping

• When a safety check declined to review an action, auto mode retried it repeatedly. Now the action is denied once, with a note that retrying won't help.
• When a safety check gave no answer, auto mode denied actions repeatedly without pause. Retries now back off, and the turn stops with a message after ten consecutive denials.

Tool-call validation

Write calls no longer fail validation when a model sends path, file_text, file_content, or a stray description instead of file_path and content.

Keybinding and terminal fixes

• Pressing Ctrl+C or Ctrl+D twice in most dialogs (/model, /effort, /config, /status, /usage, /plugin, /sandbox, /permissions, /artifacts, /mobile, /login, /upgrade, /usage-credits, /install-github-app, /setup-bedrock, /setup-vertex) used to quit Claude Code instead of closing the dialog.
• A stray n no longer closes dialogs and a stray y no longer confirms them — Enter and Esc accept and cancel. To restore the old behavior, bind y/n to confirm:yes/confirm:no in keybindings.json.
• Text fields in dialogs were losing typed letters, digits, or Space to a keybinding on that key.
• The prompt line on Windows terminals stayed scrambled after invisible characters were removed on Enter; the screen is now repainted so you review the exact text that gets sent.
• The invisible-character cleanup was stripping the zero-width non-joiner that Persian and Arabic use to attach suffixes to Latin words or numbers.

Voice dictation and caching

• Ctrl+C now stops voice dictation (previously the prompt cleared but the mic kept recording); Esc cancels while a transcript is processing; held Space no longer starts dictation from the transcript view or vim NORMAL mode.
• A model switch made from a host app (Claude Desktop, VS Code, SDK) while Claude is working no longer causes a prompt-cache miss on the next prompt.
• Resumed fork subagents now re-send their original tool list instead of rebuilding it, which had broken prompt caching for that agent.
• Subagent hand-back messages no longer show an internal provenance preamble when expanded outside verbose mode.

Plugin and skill fixes

• installed_plugins.json no longer keeps the install-time commit after updating a plugin from a GitHub repo or git URL tracking a branch or tag.
• Skills in ~/.claude/skills/ were being moved to ~/.claude/skills/.trash/ when a manifest.json in that folder listed their names.
• A skill you switched off now shows a dim ◯ instead of the same red ✘ used for a plugin that failed to load in /plugin and /skills.

Observability and UI

The hook_execution_complete OpenTelemetry event now includes hook output sizes and the number of oversized outputs saved to a file. Mouse support extends further in fullscreen mode: the wheel scrolls the /skills list and skill state options in /plugin are clickable. Home/End now work in the /config settings list and selection lists like /model, /memory, and permission prompts. Various border, highlight, and icon-consistency bugs across /mcp, /plugin, and /workflows were also patched.

If you run Claude Code in CI or with MCP servers, the auto-mode retry changes and the MCP description override are the two items worth reading twice.