Plugin Marketplace Installation and Source Verification
When you want to install plugins recommended by your team or maintained by third parties, command-line operations become cleaner and safer. claude plugin install now supports the --marketplace <source> flag, where <source> is the marketplace source to install the plugin from. If the marketplace has not been added locally, it performs the same security policy checks, automatically adds it, and completes the installation directly. On initial startup, plugin commands wait until organization-managed policies are fully loaded before proceeding. If a configured marketplace name resembles an official source, the interface also provides clear identification guidance.
FEATURE
Subagent Thinking Depth and Execution Limits
When you assign background tasks to subagents for heavy workloads, you can allocate their effort and permissions more precisely. The Agent tool adds an effort parameter, allowing you to specify the thinking depth invested by a subagent. Agent names are capped at 256 characters and rejected if they exceed this limit, while skill or plugin file name properties exceeding this length are ignored. If auto mode is disabled by configuration, triggers a circuit breaker (a safety mechanism that disables execution when protection thresholds are reached), or is unsupported by the model, subagents configured with permissionMode: auto will no longer enter auto mode. If the main conversation stops because the usage quota (the allowed usage ceiling) is exhausted, background agents encountering the same limit will not trigger repeated alerts one by one.
FEATURE
Plugin Feature Events and Testing Mechanisms
When customizing plugin modules, you can provide input suggestions and see clearly why tools were blocked. The prompt.autocomplete event allows modules to contribute completion options to the input box. The prompt and system fields of $.model.complete accept text blocks; setting cache: true on a block caches (temporarily stores for reuse) requests up to that block. The agent.spawn hook (a small program triggered by events) adds workflow agents along with their run execution and index information, allowing modules to reject startup. Write, Edit, NotebookEdit, LSP rows, and single Read, Grep, and Glob rows now display why a module rejected the call. When running claude plugin test, failed expect checks inside test-registered hooks or an engine rejection of stub responses (substitute answers used for testing) will fail the test instead of passing silently.
BUGFIX
Plugin Interception Hooks and Permission Logic Fixes
When plugins intercept operations, failure reasons and confirmation prompts should not vanish. When the hooks (small programs triggered by events) config.set, state.set, env.set, and agent.spawn reject after calling next(e), they now report the named hook failure; if they fail upon turn interruption after the call, the call is rejected. The /theme command, /config Theme menu, and initial theme setup now query config.set before saving. Approvals from tool.check no longer bypass question or plan approval prompts. Rejection reasons exceeding 4096 characters for dropped prompts or rejected settings are no longer ignored. tool.call now sees arguments with corrected parameter names. When another module's hook initiates a call under a guarded $ invocation, the guard's .catch handler is now consulted. The <system-reminder> tag output by hooks is now escaped so Claude receives it as plain text.
BUGFIX
Plugin Module Loading Performance and Worker Process Fixes
When loading complex plugins or restarting background worker units, you might have experienced multi-minute delays or duplicated startup jobs. Hook modules making frequent $.state calls through the same const binding no longer slow down loading and validation. claude plugin validate no longer outputs matchers or state values read elsewhere by the engine; reading state through top-level var declarations is now rejected if the variable is re-declared or reassigned. Plugin-provided $ methods no longer restart the hook source, which previously caused guarded hook .catch handlers to rerun infinitely. When replacing hook (a small program triggered by events) worker units, startup prompts, commands, or subagents are no longer queued redundantly. After an organization plugin reloads or another plugin crashes, returning a $ method name already added by a personal module now unloads the personal module instead. During restarts, plugin interface calls still pass through hooks added by other plugins, and tool calls still pass through permission hooks.
FEATURE
Sandbox Isolation Mechanisms and Sensitive Path Protection
When running commands in a sandbox (an environment with restricted access), you can no longer read the /ultrareview upload copy in ~/.claude/seed-admin. If managed or co-located user read-restricted paths appear or are repointed mid-session, internal project authorizations are revoked, and credential (proof of authentication) injection for covered files is stopped. Reading notebooks or PDFs on macOS and Windows no longer accesses unapproved files due to mid-operation symlink swapping. When fetching configuration fails, tampered cache (temporary storage) cannot disable or replace built-in policy plugins. The 8.3 short names and other alternate Windows spellings of home directories or drives are now recognized as deletion targets for rm -rf. PreToolUse hook approvals and auto mode no longer bypass permission prompts for reading UNC network files. After enabling strict sandbox mode in user settings, managed settings, or settings specified via --settings, interpreter commands prefixed with environment variables (runtime configuration values passed to programs), such as FOO=bar python3 app.py, no longer trigger confirmation prompts.
BUGFIX
Session Resumption and Background Scheduling Fixes
When you reopen a previous conversation to resume work or leave loop commands running in the background, restored modes and schedules should not silently break. Returning to a session from the claude --resume menu or /resume correctly restores plan mode. Schedules created and saved after /resume, /branch, or /clear no longer fail to trigger permanently; task files written twice within milliseconds no longer ignore subsequent additions and deletions. One-shot claude -p and the Agent SDK (a toolkit allowing programs to invoke agents) no longer terminate background commands 5 seconds after the final result, waiting for them instead; only one-shot claude -p additionally fixes missed scheduled wakeups and waits for wakeups. After a background session process restarts, /loop no longer silently halts due to lost wakeups.
FEATURE
File Reading, Search Retries, and Large Attachment Handling
When you ask Claude to search code, browse documents, or reference files in prompts, responses from various reading tools are much clearer. When Grep and Glob encounter unreadable files or directories, they no longer falsely report no results; Claude retries once or notifies you. Mentioning text files exceeding 256KB with @ is no longer silently ignored, as Claude receives the file size alongside instructions to read in chunks. When Read parses PDFs, specifying discrete page lists such as "6,9,15" for pages no longer returns only the first page, but errors and requests reading page-by-page or by range. When Read deny rules are configured and the working directory is under a symlink (a shortcut pointing to another location), @ references to nonexistent files no longer erroneously display "could not be examined". Grep accepts file_path in place of path; Write, WebFetch, and Read ignore a few extra parameters rather than failing the call.
FEATURE
Prompt Commands and Operation Permission Management
When switching operation modes, changing directories, or managing existing outputs during a conversation, stale records should not suddenly revive tool restrictions. After leaving auto mode or plan mode mid-session, allowed-tools rules from skills or slash commands no longer resurface in subsequent turns. Following /cd or permission changes, records for unloaded instruction files like "instruction file not loaded" are no longer outdated or missing; unread nested instruction files also leave conversation logs. Repeating /name in conversation compaction summaries no longer allows Claude to invoke user-reserved skills. The Artifact tool now allows Claude to see the total number of published artifacts and list up to 200 items at once, raised from 50 items.
BUGFIX
Terminal Input, Paste, and Vim Operation Fixes
When typing quickly in the terminal, using Chinese input methods, or editing prompts in Vim mode, you might have experienced dropped characters or misinterpreted keystrokes. Overlapping multiple paste actions in the same prompt no longer treats portions of pasted content as literal keystrokes. When the bottom bar is selected, fast input, IME text, and decomposed accented characters are no longer lost; ! also no longer preserves the selection state. The Vim (a mode for editing text using key commands) cursor no longer moves past the end of the line, and j/k motions retain their column across shorter lines; f, t, F, T, ;, and , no longer jump to matches on other lines or delete up to them. The /add-dir path box no longer allows Shift+Enter or pasted content to insert newlines, nor does it misinterpret quickly typed "tab", "up", or "down" as key events.
BUGFIX
Terminal Display Rendering and Interactive Operation Improvements
When browsing long responses or retrieving drafts, screens and menus should not force you to redo actions. Pressing Ctrl+O or Ctrl+Z while submitting /bug, /share, or /feedback <text> no longer restarts them, nor do they close in a cancelled state after submission. Opening /remote-env selects the saved default environment, and pressing Enter directly no longer changes it; when no default is active, no rows are checked. When iTerm2 is detected, fullscreen mode no longer sends a fullscreen clear on every resize or Ctrl+L, which may have caused old screens to fill the scrollback history. Long bulleted or numbered responses render faster during streaming, resizing, and reopening history with Ctrl+O. Drafts cleared with Ctrl+C can still be retrieved by pressing Up after running a slash command or sending a message. The /focus prompt invites you to try focus view during a turn and explains how to switch back.
FEATURE
MCP Startup Wait Times and Protocol Versions
When connecting through MCP (a protocol that lets AI connect to external tools), startup should not get stuck waiting for servers. The first turn of claude -p and SDK sessions no longer waits for HTTP/SSE servers to respond to resources/list. Local stdio servers that ignore new protocol checks are remembered for 7 days after one slow connection, subsequently connecting via the legacy method without further waiting. Local stdio MCP connections across all installation environments, including Bedrock, Vertex, and Foundry, negotiate protocol version 2026-07-28 by default, with MCP_PROTOCOL_NEGOTIATION=legacy available to opt out. MCP tools with names exceeding 128 characters are excluded and named in error messages, rather than failing every request.
FEATURE
Proxy Configuration, Overload Retries, and Clickable Links
When corporate networks route through a proxy server, certain addresses should connect directly. HTTPS_PROXY specifies the proxy server. When configured, Claude Code's own API requests—including login, policy, feedback, and artifacts—now respect NO_PROXY (the proxy bypass list), which was previously ignored. A new environment variable, CLAUDE_CODE_OVERLOADED_RETRY_BASE_DELAY_MS, extends the base wait time for backoff retries when encountering 529 overloads, where retries wait for longer intervals after failures. Links to claude.ai settings within usage limit messages now include https://, allowing terminals and applications to render them as clickable links.
BUGFIX
Desktop App and Remote Control Collaboration
When viewing sessions hosted on the desktop app or IDE (an environment combining code editing and development tools) remotely via web, the background subagent panel is no longer blank. Starting a new Remote Control session from claude.ai/code now accepts PDFs and other files in the first message just like subsequent messages, rather than only supporting images. Cross-session delivery notifications no longer merge two sessions with similar names into a single recipient, and overdue notifications no longer blame the desktop app when a terminal session allows messages to expire. When another message is already queued in the desktop app, clicking Send now no longer terminates the subagent that the current turn is waiting for.
FEATURE
Cloud Session State and Background Task Resumption
When delegating complex tasks to cloud environments for long-running computation, clarity is needed on which background agents can resume after a restart. Work units stopping exactly at turn completion no longer leave the interface displaying an endless turn, and long conversations no longer occasionally re-prompt for approved permissions. When Claude is reading scheduled tasks or other queued notifications, retrying or editing messages no longer loses them. After containers (the execution environments hosting sessions) restart, they no longer forget the thinking settings selected in your client, and Claude is informed of which stopped background agents can resume by ID. Cowork cloud sessions no longer falsely report that proxy servers blocked artifacts when Anthropic cannot confirm organization settings. When claude.ai cloud sessions cannot connect to the Claude in Chrome browser extension, Claude receives instructions indicating that it may proceed with alternative methods if the user chooses.
FEATURE
Cloud Routines and Attachment Format Support
When routines automatically generate private files, you no longer need to approve every publication. Scheduled runs or clicking Run now to execute immediately no longer ask for approval when generating new artifacts visible only to you; requests involving connectors (tools that link external services) or other access permissions still prompt for approval. Completed jobs no longer occasionally remain shown as running for hours. Editing or duplicating routines without saved notification settings no longer turns off push notifications. Less common image formats such as SVG, HEIC, and TIFF no longer fail to attach, instead attaching as regular files. Connector tools requiring organizational approval no longer offer the ineffective "Always allow" option.
FEATURE
Claude Tag Team Communication and Spend Controls
When teams mention @Claude in Slack, follow-up replies should not wait for initial requests to finish or be missed due to multi-second gaps. Enterprise administrators can click Edit on the Allowed domains card on a channel's Configure page to open the access bundle that configures its domains. Threads awakened solely by GitHub PR activity or routines switch to the administrator-updated channel or workspace default model. Depleted organization quotas no longer falsely report spend limits, and sessions no longer freeze after Slack auto-rejects unanswerable permission prompts. @Claude !status explains when reading unmentioned messages stopped, why, and that mentioning again resumes reading. The initial card of a thread continued with !fork displays the source, request content, author, and link to the original thread. In admin settings under spend limits, organization-wide and default fields save only upon pressing Save or Enter, without saving when clicking away.
FEATURE
Code Review Analytics Charts and Branch Rule Guardrails
When evaluating team PR review performance or refactoring project conventions, review metrics across projects are clear without concern that branch changes will invalidate checks. Review charts add current period review totals, period-over-period changes, and breakdown analytics by repository. A repository stores project files and their change history. When a PR switches target branches and the old branch has been deleted, pending review tasks automatically re-queue. If submitted changes happen to edit the project rule file CLAUDE.md, the review process strictly adheres to existing rules on the target branch.
Original changelog
Original changelogPlugin Marketplace Installation and Source Verification
Added `--marketplace <source>` to `claude plugin install`: adds the marketplace if needed, under the same policy checks as `claude plugin marketplace add`, then installs the plugin from it
Fixed `claude plugin` commands such as `marketplace add` and `install` running before an organization's managed settings had loaded on a first run
Improved the steps shown when a marketplace declared in a settings file has a name that looks like an official Anthropic marketplace
Original changelogSubagent Thinking Depth and Execution Limits
Added an `effort` parameter to the Agent tool, so Claude runs a sub-agent at the effort level you ask for
Fixed subagent definitions with `permissionMode: auto` entering auto mode when auto mode is unavailable (disabled by settings, circuit breaker, or a model that doesn't support it)
Fixed the usage limit alert repeating once per background agent when agents failed on a limit that had already stopped the main conversation
Changed agent names to allow at most 256 characters: a longer one is rejected, and a skill's or a plugin file's `name` longer than that is ignored
Original changelogPlugin Feature Events and Testing Mechanisms
Added `prompt.autocomplete`, an event a mod hooks to add its own rows to the prompt box's autocomplete list
Added prompt caching to `$.model.complete` for mods: `prompt` and `system` take blocks of text, and `cache: true` on a block caches the request up to it
Added workflow agents to the `agent.spawn` mod hook, with their run and index, so a mod can refuse them
Fixed Write, Edit, NotebookEdit and LSP rows, and single Read, Grep and Glob rows, hiding why a mod denied the call: the row now shows the reason
Changed `claude plugin test`: a failed `expect` inside a hook the test registered, or a stub answer the engine refuses, now fails the test instead of passing silently
Original changelogPlugin Interception Hooks and Permission Logic Fixes
Fixed a mod's `config.set`, `state.set`, `env.set` or `agent.spawn` hook that denies after calling `next(e)` being answered as a refusal: the hook is now reported as failed, by name
Fixed `/theme`, the `/config` Theme menu and the first-run theme step saving a theme before a plugin's `config.set` hook was asked
Fixed a plugin's `tool.check` hook answering allow running a tool that requires your answer (a question, a plan approval) without showing its dialog
Fixed a mod's hook that called `next(e)` and then failed while the turn was interrupted letting the call through; the call is now rejected
Fixed a plugin's prompt drop or setting deny being ignored when its reason was longer than 4,096 characters
Fixed plugin `tool.call` hooks seeing some tool calls before misnamed parameters were repaired; a hook now sees the arguments the tool will run with
Fixed a mod's guard hook with a `.catch` being skipped silently for calls another mod's hook makes beneath the guard's own `$` call; its `.catch` is now asked
Improved hook output handling: `<system-reminder>` tags written in a hook's output are escaped before they reach Claude
Original changelogPlugin Module Loading Performance and Worker Process Fixes
Fixed plugins whose hooks module makes many `$.state` calls through one const taking minutes to load or validate
Fixed `claude plugin validate` listing a matcher or state value for a hooks module that the engine reads from elsewhere
Fixed `claude plugin validate` listing a `$.state` value read through a top-level `var` that was declared again or reassigned; such a module is now refused
Fixed a plugin's served `$` method restarting the hook origin, which could run a guard hook with a `.catch` above it again without end
Fixed plugin interface calls made while the plugin hooks worker restarts running without the hooks other plugins put on them
Fixed a mod's start-up prompt, command or subagent being queued a second time when the hooks worker was replaced
Fixed an organization's plugin being unloaded on its own reload, or after another plugin crashed, when it returned a `$` name that a user-installed mod had added; the mod is now unloaded instead
Fixed tool calls made while the plugin hooks worker restarts being answered without the plugins' permission hooks
Original changelogSandbox Isolation Mechanisms and Sensitive Path Protection
Fixed sandboxed commands being able to read the staged file copies of `/ultrareview` uploads under `~/.claude/seed-admin`
Fixed a managed sandbox read-deny path (and user ones beside it) that appears or re-points mid-session not dropping project grants inside it or ending credential injection from files it covers
Fixed 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
Fixed 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
Fixed `rm -rf` on the 8.3 short name or another alternate Windows spelling of the home folder or a drive not being treated as removing it
Security: Fixed PreToolUse hook approvals and auto mode bypassing the permission prompt for file reads from network (UNC) paths
Improved sandbox auto-allow: with strict sandbox mode set in user, managed or --settings settings, an interpreter command with an env var prefix like `FOO=bar python3 app.py` runs unprompted
Original changelogSession Resumption and Background Scheduling Fixes
Fixed one-shot `claude -p` and Agent SDK runs stopping a background command 5 seconds after the final result, and one-shot `claude -p` runs dropping a scheduled wakeup; both are now waited for
Fixed plan mode not being restored when resuming a session from the `claude --resume` session picker or with `/resume`
Fixed saved scheduled tasks created after `/resume`, `/branch` or `/clear` never firing, and saved tasks ignoring later creates and deletes after two writes to the tasks file milliseconds apart
Fixed a background session's `/loop` silently stopping when the session's process restarted (for example after a crash), because its pending wakeup was lost
Original changelogFile Reading, Search Retries, and Large Attachment Handling
Fixed 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
Fixed the Read tool returning only the first entry, with no error, when a PDF's `pages` was a list such as "6,9,15"; it now returns an error saying to read each page or range separately
Fixed @-mentioned text files over 256KB being left out silently: Claude is now told the file's size and to read it in portions
Fixed a spurious "could not be examined" note for @-words that name no file when a Read deny rule is set and the working directory is under a symlink
Improved tool input handling: Grep accepts `file_path` for `path`, and Write, WebFetch and Read ignore a few stray parameters instead of failing the call
Original changelogPrompt Commands and Operation Permission Management
Fixed a skill's or slash command's `allowed-tools` rule coming back in a later turn when you leave auto mode or plan mode partway through that turn
Fixed "instruction file not loaded" lines going stale or missing after `/cd` or a permission change, and added a transcript line when a nested one isn't loaded
Fixed a compaction summary that repeated `/name` letting Claude invoke a skill that is reserved for the user
Improved the Artifact tool's listing: Claude now sees how many published artifacts you have and can list up to 200 at once instead of 50
Original changelogTerminal Input, Paste, and Vim Operation Fixes
Fixed some pasted text reaching Claude as typed text when several pastes overlapped in one prompt
Fixed 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
Fixed `/add-dir` path box letting Shift+Enter or a paste add a line break, and treating fast-typed "tab", "up" or "down" as those keys
Fixed fast typing, input-method text and decomposed accents being dropped while a prompt footer row was selected, and `!` leaving the row selected
Original changelogTerminal Display Rendering and Interactive Operation Improvements
Fixed `/bug`, `/share` and `/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
Fixed `/remote-env` replacing your saved default environment when you pressed Enter right away: the list now opens on your default, and no row has a check mark when no default is in effect
Fixed 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
Improved rendering speed of long bulleted or numbered replies: they stream, resize and re-open in the transcript (ctrl+o) much faster
Improved Ctrl+C draft recovery: a cleared prompt now stays reachable with Up after a slash command or a sent message
Improved the /focus tip: it now invites you to try focus view mid-turn and shows how to switch back
Original changelogMCP Startup Wait Times and Protocol Versions
Fixed 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
Improved startup of `claude -p` and SDK sessions: the first turn no longer waits for HTTP and SSE MCP servers to answer `resources/list`
Improved startup with 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
Changed local (stdio) MCP server connections to negotiate protocol version 2026-07-28 by default on every install, including Bedrock, Vertex and Foundry; `MCP_PROTOCOL_NEGOTIATION=legacy` opts out
Original changelogProxy Configuration, Overload Retries, and Clickable Links
Added `CLAUDE_CODE_OVERLOADED_RETRY_BASE_DELAY_MS` environment variable to set a longer base delay for the backoff when retrying an overloaded (529) request
Fixed `NO_PROXY` being ignored for Claude Code's own API requests (sign-in, policy, feedback, artifacts) when `HTTPS_PROXY` is set
Changed usage limit messages to write claude.ai settings links with https:// so terminals and apps can make them clickable
Original changelogDesktop App and Remote Control Collaboration
Fixed Remote Control viewers seeing an empty subagent pane for background subagents in sessions hosted by the desktop app or an IDE
Fixed cross-session delivery notices showing two sessions with similar names as one recipient, and the expiry notice blaming the desktop app when a terminal session let the message lapse
Fixed Send now in the desktop app ending the subagent a turn was waiting on when another message was already queued
[Remote Control] Fixed the first message of a new Remote Control session started from claude.ai/code accepting only images; it now accepts PDFs and other files like later messages
Original changelogCloud Session State and Background Task Resumption
Fixed a cloud session showing a turn that never ended when its worker was stopped just as the turn finished
Fixed cloud sessions with a large transcript sometimes asking for a permission again after it was approved
Fixed scheduled tasks and other queued notifications being lost in cloud sessions when a message was retried or edited while Claude was reading them
Fixed cloud sessions forgetting the thinking setting chosen in the client when the session's container restarted
Fixed Cowork cloud sessions saying a proxy blocked artifacts when Anthropic couldn't confirm the organization's settings
Improved cloud sessions after a restart: Claude is now told which stopped background agents it can resume by id
Improved the Claude in Chrome message in claude.ai cloud sessions when the browser can't be reached: Claude is now told it may continue with alternatives if the user prefers
Original changelogCloud Routines and Attachment Format Support
Changed scheduled and Run now routine runs to publish a new artifact only you can see without asking for approval; artifacts that request connectors or other access still ask
[Cloud sessions] Fixed routine runs occasionally staying listed as running for hours after they had finished
[Cloud sessions] Fixed editing or duplicating a routine turning off its push notifications when the routine had no saved notification setting
[Cloud sessions] Fixed SVG, HEIC, TIFF and other less common image files failing to attach; they now attach as regular files
[Cloud sessions] Fixed approval prompts offering "Always allow" for connector tools that an organization set to require approval; the choice had no effect
Original changelogClaude Tag Team Communication and Spend Controls
[Claude Tag] Added an Edit button to the Allowed domains card on a channel's Configure page, so Enterprise admins can open the access bundle that sets the channel's domains
[Claude Tag] Fixed 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] Fixed Slack threads woken only by GitHub pull request activity or a routine staying on their original model after an admin changed the channel or workspace default model
[Claude Tag] Fixed Claude sometimes posting a spend limit notice in Slack when the real cause was that your organization had run out of usage credits
[Claude Tag] Fixed a session hanging until interrupted when a permission prompt that can't be answered from Slack was denied automatically
[Claude Tag] Improved `@Claude !status` in a channel to say when Claude has stopped reading its untagged messages, why, and that an @-mention starts it reading again
[Claude Tag] Changed the first message of a Slack thread continued with `!fork` to a card showing where it came from, the request, and who asked, with a link to the original thread
[Claude Tag] Changed the organization-wide and default spend limit boxes on Claude Tag's spend limits page in admin settings to save only when you press Save or Enter, not when you click away
Original changelogCode Review Analytics Charts and Branch Rule Guardrails
[Code Review] Added the period's total with its change from the previous period, and a breakdown by repository, to the PRs reviewed chart in Code Review analytics
[Code Review] Fixed 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
[Code Review] Fixed reviews ignoring a CLAUDE.md's rules when the pull request edits that file; reviews now use its version from the base branch