What changed
Claude Code 2.1.292 shipped on October 6, 2026. It is a large release, centered on three areas. First, new options for installing plugins and running sub-agents. Second, a set of permission and sandbox fixes, covering network (UNC) paths, links swapped in mid-read and a tampered settings cache. Third, fixes for carrying a session forward (scheduled tasks after /resume, /loop, plan mode restore) and for the plugin (mod) hooks engine. The items below are a selection of the changes with the most user impact.
New features
-
claude plugin install --marketplace <source>Installing a plugin from another marketplace used to take a separate
claude plugin marketplace addfirst.--marketplace <source>adds the marketplace if needed, under the same policy checks asclaude plugin marketplace add, then installs the plugin from it. -
effortparameter on the Agent toolClaude now runs a sub-agent at the effort level you ask for.
-
CLAUDE_CODE_OVERLOADED_RETRY_BASE_DELAY_MSSets a longer base delay for the backoff when retrying an overloaded (529) request.
-
Mod hook additions
prompt.autocompleteis an event a mod hooks to add its own rows to the prompt box’s autocomplete list.$.model.completegets prompt caching for mods:promptandsystemtake blocks of text, andcache: trueon a block caches the request up to it. Theagent.spawnmod hook now sees workflow agents, with their run and index, so a mod can refuse them. -
[Claude Tag] Edit button on the Allowed domains card
On a channel’s Configure page, Enterprise admins can open the access bundle that sets the channel’s domains.
-
[Code Review] PRs reviewed chart
Code Review analytics adds the period’s total with its change from the previous period, and a breakdown by repository.
Key improvements
-
Faster startup for
claude -pand SDK sessionsThe first turn no longer waits for HTTP and SSE MCP servers to answer
resources/list. -
Local (stdio) MCP servers that ignore the newer protocol check
After one slow connect they are remembered for 7 days and connected the older way without the wait.
-
Rendering speed of long bulleted or numbered replies
They stream, resize and re-open in the transcript (ctrl+o) much faster.
-
Ctrl+C draft recovery
A cleared prompt now stays reachable with Up after a slash command or a sent message.
-
Hook output handling
<system-reminder>tags written in a hook’s output are escaped before they reach Claude. -
Tool input handling
Grep accepts
file_pathforpath, and Write, WebFetch and Read ignore a few stray parameters instead of failing the call. -
Sandbox auto-allow
With strict sandbox mode set in user, managed or
--settingssettings, an interpreter command with an env var prefix likeFOO=bar python3 app.pyruns unprompted. -
Artifact tool listing
Claude now sees how many published artifacts you have and can list up to 200 at once instead of 50.
-
Other improvements
The steps shown when a marketplace declared in a settings file has a name that looks like an official Anthropic marketplace are improved. After a restart, cloud sessions tell Claude which stopped background agents it can resume by id. In claude.ai cloud sessions, when Claude in Chrome can’t reach the browser, Claude is now told it may continue with alternatives if the user prefers.
Changed behavior
-
Local (stdio) MCP servers negotiate protocol version 2026-07-28 by default
This applies on every install, including Bedrock, Vertex and Foundry;
MCP_PROTOCOL_NEGOTIATION=legacyopts out. -
claude plugin testfails instead of passing silentlyA failed
expectinside a hook the test registered, or a stub answer the engine refuses, now fails the test. -
Routine runs publish private artifacts without approval
Scheduled and Run now routine runs publish a new artifact only you can see without asking for approval; artifacts that request connectors or other access still ask.
-
Agent names are capped at 256 characters
A longer one is rejected, and a skill’s or a plugin file’s
namelonger than that is ignored. -
Clickable usage limit links
Usage limit messages now write claude.ai settings links with https:// so terminals and apps can make them clickable.
Bug fixes
Permissions & security
- Security: PreToolUse hook approvals and auto mode bypassing the permission prompt for file reads from network (UNC) paths.
- Sandboxed commands being able to read the staged file copies of
/ultrareviewuploads under~/.claude/seed-admin. - A notebook or PDF read on macOS and Windows being able to return a file outside what was approved, through a link swapped in mid-read.
- A managed sandbox read-deny path that appears or re-points mid-session not dropping project grants inside it or ending credential injection from files it covers.
- A tampered on-disk cache of server-managed settings being able to switch off or unseat the built-in policy plugin while the settings fetch failed.
rm -rfon the 8.3 short name or another alternate Windows spelling of the home folder or a drive not being treated as removing it.- Subagent definitions with
permissionMode: autoentering auto mode when auto mode is unavailable (disabled by settings, circuit breaker, or a model that doesn’t support it). - A skill’s or slash command’s
allowed-toolsrule coming back in a later turn when you leave auto mode or plan mode partway through that turn. - A compaction summary that repeated
/nameletting Claude invoke a skill that is reserved for the user. - A plugin’s
tool.checkhook answering allow running a tool that requires your answer (a question, a plan approval) without showing its dialog. - Tool calls made while the plugin hooks worker restarts being answered without the plugins’ permission hooks.
Network & MCP
NO_PROXYbeing ignored for Claude Code’s own API requests (sign-in, policy, feedback, artifacts) whenHTTPS_PROXYis set.- An MCP tool with a name longer than 128 characters making every request fail; that tool is now left out and an MCP error names it.
claude plugincommands such asmarketplace addandinstallrunning before an organization’s managed settings had loaded on a first run.
Resume & scheduled tasks
- Plan mode not being restored when resuming a session from the
claude --resumesession picker or with/resume. - Saved scheduled tasks created after
/resume,/branchor/clearnever firing, and saved tasks ignoring later creates and deletes after two writes to the tasks file milliseconds apart. - A background session’s
/loopsilently stopping when the session’s process restarted (for example after a crash), because its pending wakeup was lost. - One-shot
claude -pand Agent SDK runs stopping a background command 5 seconds after the final result, and one-shotclaude -pruns dropping a scheduled wakeup; both are now waited for. - The usage limit alert repeating once per background agent when agents failed on a limit that had already stopped the main conversation.
Tools & file reads
- Grep and Glob reporting no matches when the file or folder they were given could not be read; Claude now retries once or tells you.
- The Read tool returning only the first entry, with no error, when a PDF’s
pageswas a list such as “6,9,15”; it now returns an error saying to read each page or range separately. - @-mentioned text files over 256KB being left out silently; Claude is now told the file’s size and to read it in portions.
- Write, Edit, NotebookEdit and LSP rows, and single Read, Grep and Glob rows, hiding why a mod denied the call; the row now shows the reason.
Input & display
- Some pasted text reaching Claude as typed text when several pastes overlapped in one prompt.
- Vim mode leaving the cursor past the end of a line, j/k losing their column on shorter lines, and
f/t/F/T/;/,jumping to, or deleting up to, a match on another line of the prompt. - Fullscreen mode sending a full-screen clear on every window resize and Ctrl+L when iTerm2 is detected, which may be what filled iTerm2’s scrollback with stale pages.
- Fast typing, input-method text and decomposed accents being dropped while a prompt footer row was selected.
/add-dirpath box letting Shift+Enter or a paste add a line break, and treating fast-typed “tab”, “up” or “down” as those keys./remote-envreplacing your saved default environment when you pressed Enter right away; the list now opens on your default./bug,/shareand/feedback <text>starting over after Ctrl+O or Ctrl+Z while a report was being sent, and closing as cancelled after it had been sent.
Plugins & mods
- A mod’s guard hook with a
.catchbeing skipped silently for calls another mod’s hook makes beneath the guard’s own$call; its.catchis now asked. - A mod’s hook that called
next(e)and then failed while the turn was interrupted letting the call through; the call is now rejected. - A mod’s
config.set,state.set,env.setoragent.spawnhook that denies after callingnext(e)being answered as a refusal; the hook is now reported as failed, by name. - Plugin
tool.callhooks seeing some tool calls before misnamed parameters were repaired; a hook now sees the arguments the tool will run with. - An organization’s plugin being unloaded when it returned a
$name that a user-installed mod had added; the mod is now unloaded instead. - A plugin’s prompt drop or setting deny being ignored when its reason was longer than 4,096 characters.
- Plugins whose hooks module makes many
$.statecalls through one const taking minutes to load or validate, andclaude plugin validatelisting a matcher or state value for a hooks module that the engine reads from elsewhere. /theme, the/configTheme menu and the first-run theme step saving a theme before a plugin’sconfig.sethook was asked.
Cloud sessions, Remote Control & desktop app
- [Cloud sessions] Sessions with a large transcript sometimes asking for a permission again after it was approved, forgetting the thinking setting chosen in the client when the container restarted, and losing scheduled tasks and other queued notifications when a message was retried or edited.
- [Cloud sessions] Routine runs occasionally staying listed as running for hours after they had finished, and editing or duplicating a routine turning off its push notifications.
- [Cloud sessions] SVG, HEIC, TIFF and other less common image files failing to attach; they now attach as regular files.
- [Cloud sessions] Approval prompts offering “Always allow” for connector tools that an organization set to require approval; the choice had no effect.
- [Remote Control] The first message of a new Remote Control session started from claude.ai/code accepting only images; it now accepts PDFs and other files.
- [Remote Control] Viewers seeing an empty subagent pane for background subagents in sessions hosted by the desktop app or an IDE.
- Send now in the desktop app ending the subagent a turn was waiting on when another message was already queued.
- Cowork cloud sessions saying a proxy blocked artifacts when Anthropic couldn’t confirm the organization’s settings.
Claude Tag & Code Review
- [Claude Tag] Replies sent in a Slack thread while Claude was still on its first request there being held until that request finished, or missed when sent seconds apart.
- [Claude Tag] Claude posting a spend limit notice when the real cause was that your organization had run out of usage credits, and a session hanging when a permission prompt that can’t be answered from Slack was denied automatically.
- [Claude Tag] Slack threads woken only by GitHub pull request activity or a routine staying on their original model after an admin changed the default model.
- [Code Review] Reviews ignoring a CLAUDE.md’s rules when the pull request edits that file; reviews now use its version from the base branch.
- [Code Review] A queued review failing when the pull request moved to a new base branch and the old one was deleted; the commit is now re-queued for review.
Notes
- If you install plugins from other marketplaces,
claude plugin install --marketplace <source>adds the marketplace and installs in one step. - If a local (stdio) MCP server misbehaves after this update, it may be the new default protocol version 2026-07-28.
MCP_PROTOCOL_NEGOTIATION=legacyopts out. - If you hit overloaded (529) errors often,
CLAUDE_CODE_OVERLOADED_RETRY_BASE_DELAY_MSsets a longer base delay for the retry backoff. - Mod authors using
claude plugin testmay see tests that used to pass silently now fail. Checkexpectcalls inside hooks and stub answers. - Agent, skill and plugin file names over 256 characters are now rejected or ignored.
- Repositories using Code Review: pull requests that edit CLAUDE.md are now reviewed against the base branch’s version.