Claude Code v2.1.265: Plugin Directory Support, 1GB Tool Result Cap, and More
Anthropic shipped Claude Code v2.1.265, a maintenance release with several notable additions and a long list of bug fixes aimed at plugin management, telemetry, and reliability.
Key Additions
- Plugin folder support – You can now point
--plugin-dirat a folder containing multiple plugins. Each child folder with amanifestloads, and plugins added or removed while running are picked up automatically. - 1 GB cap on tool results – Results saved to disk are now limited to 1 GB. If a result is truncated, the in-conversation preview indicates that the saved file was truncated.
- Unified telemetry – Claude Desktop and Cowork now send
user.emailanduser.groupsthrough the Claude apps gateway, matching terminal sessions.
Bug Fixes Worth Noting
- Prompt-cache reuse fixes – Resuming a foreground-spawned subagent no longer changes its tool list and system prompt prefix, preserving cache reuse. Similarly, agent teammates and resumed subagents keep
SubagentStarthook context and preloaded skills in the prompt prefix. - Resume after crash during tool call – The last prompt is no longer rewritten, and the interrupted tool call is kept and marked as interrupted.
- Command-line fixes –
/model opusplan[1m]no longer fails with "Model not found." Also,/modelnow reports if saving the default model fails, instead of falsely claiming success. - Syntax highlighting – Fixed a bug that occasionally dropped a character after Ruby
?, Erlang$, or Perl$sigils in permission prompts. - Terminal UI – Fullscreen transcript no longer jumps a row when suggestion lists open/close. Two-key shortcuts now wait up to 3 seconds for the second key (important in tmux) and show a notice on timeout.
- Security – Plugins with backslashes in paths no longer bypass symlink containment checks on macOS/Linux. Plugin dirs starting with two dots are now handled correctly.
- Remote Control & background sessions – End-of-turn signal is now sent after the final message. Background (
--bg) sessions are less likely to be retired mid-turn near idle timeout. - IDE/SDK sessions – VS Code and SDK sessions no longer require re-login when a session closed during a token refresh.
- Plugin and marketplace –
/pluginDiscover/Browse andclaude plugin list --json --availablenow show descriptions for marketplace plugins whose metadata only exists inplugin.json. - Workflow runs – Resuming after a container restart now gives a clear error if the run journal is missing, instead of rerunning every agent.
- API skills – The
claude-apiskill's error-code guide now correctly states model access failures return 404 and unavailable beta headers return 400 (not 403). - Non-interactive sessions – A
cdnow persists across turns in-pmode with stream-json input, Agent SDK, and cloud sessions.
This release focuses on hardening the developer experience, especially for teams relying on prompt caching and long-running sessions.
📖 Read the full source: GitHub Claude-Code
👀 See Also

Claude Code gains TLA+ model checking via tla-mcp MCP server
tla-mcp is a new MCP server that lets Claude Code call the TLA+ model checker tla-rs as a first-class tool — validate specs, run bounded checks with counterexample traces, and replay scenarios from the chat.

Privacy-First MCP Server Directory Launches with Documented Data Handling Policies
A new directory at toolora.dev/mcp-hub lists MCP servers with documented data handling policies, including local vs hosted classification, what data each tool transmits, and whether accounts are required. The creator also provides a browser test method to verify privacy claims.

agent-recall: Local SQLite MCP for Persistent Claude Code Memory
agent-recall is an MCP server that gives Claude Code persistent memory across sessions using a local SQLite file. It provides 9 MCP tools for saving entities, relationships, and observations, with LLM-summarized briefings at session start instead of raw data dumps.

Open-Source Job Search Framework Built in Claude Code
A developer created an open-source job search framework in Claude Code that handles structured self-profiling, automated job portal scraping, fit evaluation, and a drafter-reviewer agent pipeline for tailored applications. The system stops before submission and requires manual review.