Claude Code v2.1.77 Release: Token Limits, Sandbox Controls, and Bug Fixes

Token Limits and Sandbox Controls
Claude Code v2.1.77 increases the default maximum output token limits for Claude Opus 4.6 to 64k tokens, with the upper bound for both Opus 4.6 and Sonnet 4.6 models raised to 128k tokens. The release adds an allowRead sandbox filesystem setting that re-allows read access within denyRead regions, providing finer-grained control over file permissions.
Command Improvements
The /copy command now accepts an optional index parameter: /copy N copies the Nth-latest assistant response. The /fork command has been renamed to /branch, though /fork remains as an alias. The Agent tool no longer accepts a resume parameter; developers should use SendMessage({to: agentId}) to continue previously spawned agents, and SendMessage now auto-resumes stopped agents in the background instead of returning errors.
Performance and Memory Fixes
Startup on macOS is approximately 60ms faster due to reading keychain credentials in parallel with module loading. The --resume feature is up to 45% faster for fork-heavy and very large sessions, with about 100-150MB less peak memory usage. Background bash tasks are now killed if output exceeds 5GB, preventing runaway processes from filling disk space. The release fixes memory growth in long-running sessions from progress messages surviving compaction and resolves an auto-updater issue that caused overlapping binary downloads when the slash-command overlay repeatedly opened and closed, which could accumulate tens of gigabytes of memory.
Permission and Security Fixes
Fixed "Always Allow" on compound bash commands (e.g., cd src && npm test) saving a single rule for the full string instead of per-subcommand, which led to dead rules and repeated permission prompts. Fixed PreToolUse hooks returning "allow" bypassing deny permission rules, including enterprise managed settings. Fixed CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS not stripping beta tool-schema fields, which caused proxy gateways to reject requests.
Terminal and UI Improvements
Fixed Backspace and Delete keys not working in vim NORMAL mode. Fixed status line not updating when vim mode is toggled on or off. Fixed hyperlinks opening twice on Cmd+click in VS Code, Cursor, and other xterm.js-based terminals. Fixed background colors rendering as terminal-default inside tmux with default configuration. Fixed iTerm2 session crash when selecting text inside tmux over SSH. Fixed clipboard copy silently failing in tmux sessions; copy toast now indicates whether to paste with ⌘V or tmux prefix+]. Fixed ← / → accidentally switching tabs in settings, permissions, and sandbox dialogs while navigating lists.
Other Notable Fixes
- Fixed
--resumesilently truncating recent conversation history due to a race between memory-extraction writes and the main transcript - Fixed Write tool silently converting line endings when overwriting CRLF files or creating files in CRLF directories
- Fixed cost and token usage not being tracked when the API falls back to non-streaming mode
- Fixed Bash tool reporting errors for successful commands when the system temp directory path contains spaces
- Fixed paste being lost when typing immediately after pasting
- Fixed Ctrl+D in
/feedbacktext input deleting forward instead of the second press exiting the session - Fixed API error when dragging a 0-byte image file into the prompt
- Fixed Claude Desktop sessions incorrectly using the terminal CLI's configured API key instead of OAuth
- Fixed
git-subdirplugins at different subdirectories of the same monorepo commit colliding in the plugin cache - Fixed ordered list numbers not rendering in terminal UI
- Fixed a race condition where stale-worktree cleanup could delete an agent worktree just resumed from a previous crash
- Fixed input deadlock when opening
/mcpor similar dialogs while the agent is running - Fixed IDE integration not auto-connecting when Claude Code is launched inside tmux or screen
- Fixed CJK characters visually bleeding into adjacent UI elements when clipped at the right edge
- Fixed teammate panes not closing when the leader exits
- Fixed iTerm2 auto mode not detecting iTerm2 for native split-pane teammates
Additional Improvements
Improved Esc to abort in-flight non-streaming API requests. Improved claude plugin validate to check skill, agent, and command frontmatter plus hooks/hooks.json, catching YAML parse errors and schema violations. Sessions are now auto-named from plan content when you accept a plan. Improved headless mode plugin installation to compose correctly with CLAUDE_CODE_PLUGIN_SEED_DIR. Shows a notice when apiKeyHelper takes longer than 10s, preventing it from blocking the main loop. For VSCode integration, improved plan preview tab titles to use the plan's heading instead of "Claude's".
📖 Read the full source: GitHub Claude-Code
👀 See Also
Transformer Language Model Runs Locally on Stock Game Boy Color
Andrej Karpathy's TinyStories-260K model runs on a stock Game Boy Color via a custom ROM, using INT8 fixed-point math and bank-switched cartridge memory for weights and KV cache.

Claude Code System Prompts v2.1.51/52: New Prompts, SDK Updates, and GA Features
Claude Code system prompts v2.1.51 and v2.1.52 add six new prompts, update SDK/API references across seven languages, and promote code execution and memory to GA. The Python Agent SDK has been reworked with async changes and new interfaces.

Claude-Code v2.1.79 adds remote control, fixes subprocess hangs, and improves memory usage
Claude-Code v2.1.79 introduces a /remote-control command for VSCode to bridge sessions to claude.ai/code, fixes claude -p hanging in subprocesses, and reduces startup memory usage by ~18MB. The release also adds a --console flag for Anthropic Console authentication and improves API timeout handling.

Attentional Gating: The Challenge of Selective Forgetting in AI Memory Systems
A developer building a five-layer memory system for an OpenClaw bot identifies a key limitation: current approaches focus on recall but lack mechanisms for suppressing irrelevant information during focused tasks, similar to human attentional gating.