Repository navigation
fix(deps): update dependency claude ( 2.1.322 ➔ 2.1.323 ) - #439
Merged
Merged
Conversation
There was a problem hiding this comment.
claude: v2.1.322 → v2.1.323
Verdict: Changes required before merge
Breaking changes:
- The context window assumed for a model ID Claude Code does not recognize changed from 200K to 1M — introduced in v2.1.323. This repo pins an unrecognized third-party model ID (
minimax-m3) as the session model, so it falls directly into the changed bucket: sessions now auto-compact near 967K instead of at 200K. If the route behindANTHROPIC_BASE_URLreally serves the full 1M window of the model, this is a straight win, and the fix below then just makes the assumption explicit. The risk is that the real ceiling of the route is lower — the 1M figure is provider-dependent, and some providers expose roughly 512K for the same model — and Claude Code cannot see the difference: the docs are explicit that Claude Code cannot detect a lower limit that the gateway or the server behind it enforces. When the conversation outgrows the real limit of the route, the gateway rewrites the overflow error into wording Claude Code does not recognize as a too-long error, so it does not compact and retry automatically and the session fails every turn until/compactor/clear— which is exactly what the changelog for this release warns about for sessions on other vendors models with a smaller window. The same topology already has a filed upstream reproduction, where a third-party Anthropic-compatible gateway serving a 1M-context model drove a session to 1.1M/1M andPrompt is too long.- Affects
home/private_dot_claude/private_settings.json.tmpl(themodelkey and theenvblock) and the inlinesettings:block in.github/workflows/renovate-review.yaml. Note that the mise pin inhome/dot_config/mise/private_config.tomlgoverns the local CLI only — the CI reviewer runs the CLI bundled withanthropics/claude-code-action@v1.0.271(which pulls@anthropic-ai/claude-agent-sdk@^0.3.322), a separate version stream that will pick up the same change independently. - Fix: declare the real window of the route instead of inheriting the new default. Set
CLAUDE_CODE_MAX_CONTEXT_TOKENSto the window the route actually serves — per the Claude Code docs this applies directly to an unrecognized ID that does not start withclaude-and carries no[1m]suffix, whichminimax-m3satisfies, and proactive compaction then continues at the declared window. SetCLAUDE_CODE_AUTO_COMPACT_WINDOWsomewhat below the real limit for headroom; it takes precedence over theautoCompactWindowsetting and is clamped to 100K–1M. Also setCLAUDE_CODE_MAX_OUTPUT_TOKENSbelow the output limit of the gateway model, as the gateway troubleshooting guide recommends.CLAUDE_CODE_DISABLE_1M_CONTEXT=1is the blunt alternative: it holds the window at 200K and auto-compacts there, at the cost of discarding most of the window even when the route really serves 1M — and background agents ignore it when computing the auto-compact threshold, so pair it withCLAUDE_CODE_AUTO_COMPACT_WINDOWas well.
- Affects
Deprecations:
- None affecting this repo. This release removes the
subagentStatusLinesetting and deprecates a pluginagentsetting; the settings template defines neither, and no Claude Code status line or plugins are configured here.
New features worth adopting:
CLAUDE_CODE_DISABLE_AGENT_PROGRESS_SUMMARIES=1— turns off subagent progress summaries and the requests that write them, cutting token spend on every run that spawns subagents. Would change theenvblock inhome/private_dot_claude/private_settings.json.tmpland thesettings:block in.github/workflows/renovate-review.yaml.- Overload retries now back off up to 300s instead of 64s (
CLAUDE_CODE_OVERLOADED_RETRY_MAX_DELAY_MSceiling raised) — no configuration needed, but a metered gateway that returns overload errors is now retried for up to five minutes instead of about one, which is the usual failure mode on a third-party provider. - The ripgrep and internal git commands that Claude Code runs itself now run without the sign-in tokens and API keys of the session, so the gateway credential kept in the settings
envblock is no longer handed to git subprocesses (or to any credential helper or hook they invoke). No repo change needed; worth knowing given this repo keeps the key inenv. /mcpnow reports how many of the tools of a server were left out because the API rejected their input schema — useful for diagnosing the LiteLLM MCP endpoint inhome/.chezmoitemplates/claude-mcp-servers.json, where a rejected schema previously failed silently.
Sources consulted:
- https://togithub.com/anthropics/claude-code/blob/HEAD/CHANGELOG.md#21323
- https://togithub.com/anthropics/claude-code/compare/v2.1.322...v2.1.323
- https://code.claude.com/docs/en/model-config.md
- https://code.claude.com/docs/en/env-vars.md
- https://code.claude.com/docs/en/llm-gateway-connect.md
- https://code.claude.com/docs/en/llm-gateway-protocol.md
- https://togithub.com/anthropics/claude-code/issues/102788
- https://togithub.com/anthropics/claude-code/issues/74373
robot-rosey
Bot
force-pushed
the
renovate/claude-2.x
branch
2 times, most recently
from
November 13, 2026 17:11
a6c53ad to
0500119
Compare
robot-rosey
Bot
force-pushed
the
renovate/claude-2.x
branch
from
November 13, 2026 17:27
0500119 to
9eb2642
Compare
mrdynamo
approved these changes
Nov 13, 2026
mrdynamo
dismissed
clanker-claude[bot]’s stale review
November 13, 2026 18:31
Addressed: context window pinned on main
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This PR contains the following updates:
2.1.322→2.1.323Release Notes
anthropics/claude-code (claude)
v2.1.323Compare Source
engines["claude-code"]toplugin.json: plugins outside their stated Claude Code version range are not installed, updated or loaded, and install and update pick the newest version that fitsclaude plugin validate; each finding is a warning, and--strictfails on itclaude plugin validatethe$nouns a mod'sengine.createhook adds, withholds or replacesCLAUDE_CODE_AUTO_MODE_MODELin managed settings for Claude Desktop sessions on third-party providers (takes effect once Claude Desktop includes this version)claude -pand SDK sessions, so a host such as Claude Desktop won't end the turn as hung (once it includes this version).docxsupport to Artifact tool asset uploads/diff to closehint to the diff panel, and the keys that focus and close a panel to the?shortcut guide*in a variable's value, such asBash(PORT=* npm run dev)CLAUDE_CODE_PERFORCE_MODEwhen Perforce settings are found and the mode is offCLAUDE_CODE_DISABLE_AGENT_PROGRESS_SUMMARIES(available since 2.1.318): set it to1to turn off subagent progress summaries and the requests that write them (#86879)CLAUDE_CODE_RESUME_HINT_COMMAND, which a program that starts Claude Code can set to have the exit hint name its own resume commandunsetOptionCountto theclaude plugin install --jsonresult: how manyuserConfigoptions are not yet setisPromptCoveredto the props a mod's pane is drawn with, telling it when a dialog or a field above the prompt covers the prompt/compactagain, or after a command such as/context, summarizing away a message that Claude had not answered yetsettings.jsonmoments earlier; a settings file you delete during a session now stays deletedclaude plugincommands andclaude mcp listsigning you out of a Console profile when they ended while the token was being renewed/procfiles, and Edit hanging or running out of memory on devices and named pipesExitWorktree, the worktree exit dialog and/resumeleaving the session in the wrong folder when the folder the shell was in before entering the worktree had been deletedpsprocesses piling up when the process listing Claude Code runs before stopping a command takes too longWorktreeCreatehook or belonging to a nested repository/fork, the Left arrow or--bgcontinued it in the background; reopen any session that was open during the updateTOKEN_EXPIRED, instead of asking you to sign in againant auth loginprofile without credit was used instead/loginin another terminal, they carry on at the next messageant auth loginorganization when a customX-Api-Keyheader is set/loginor a token refresh when the keychain was slow to answerawsAuthRefreshorgcpAuthRefreshcommand sometimes running twice for one request when three or more sessions needed a fresh sign-in at onceclaude -pwarning about an ended claude.ai login to say thatclaude auth logoutalso removes saved MCP server logins and plugin secrets.gitor.claude, where the saved rule did not stop the next promptlsofcommands ran without a permission promptcat > file <<EOFbeing shown as "Read 1 file"rmfrom the wrong folder and skipped the dangerous-removal prompt under bypass permissions or a Bash allow rule--permission-mode autoor a settingsdefaultMode: autostarting in a mode that could allow more when auto mode was unavailable. It now starts in Manual mode, which asks firstplansDirectoryis set; a write that can't be shown to stay inside the project is now refusedBash(ls*), skipping the permission prompt in some cases; a rule that names a variable, such asBash(NODE_ENV*), now asksANTHROPIC_BASE_URLon a model whose gateway name contains "haiku"; such a model, other thanclaude-haiku-5-5, is now refused, as Haiku 4.5 issudo,envand similar wrapperssudoflags skipping the permission prompt in some casessandbox.excludedCommandsletting a command that it does not name run outside the sandbox in some casesReaddeny ruletool.checkhook naming every plugin hooked on the tool instead of the one that askedCLAUDE_CODE_SUBPROCESS_ENV_SCRUBis set andsandbox.enabledis offclaude -pand the Agent SDK running Bash unsandboxed when--settingsor SDK settings that turn the sandbox on fail validation; Claude Code now stops at start-up and names the invalid setting/exportsilently stopping early after a long run of calls to an MCP tool that is no longer loadedclaude mcp add --helpsplitting long example commands across lines in narrow terminals and pipes, so a pasted example ran without its headerdefaultEnabled: falsestaying off in a claude.ai chat's cloud session after you turn them on in claude.ai/fast onand/fast offon Vertex not asking a plugin'sconfig.sethook, as the/fastmenu does-prunsnpm,pip,uv,cargo,githuborghbeing refused as reserved namesinit.plugin_errorsnow lists themrenamesmap making Claude Code ignore the marketplace's other renamesclaude --bg --safe-modesessions stopping at startup to ask about project MCP servers that safe mode never starts/mcpand/pluginsopening an MCP server's page with the cursor on "Disable" when the server needs authentication but still holds its sign-in; the page now offers "Reconnect" and opens on itenabledPluginsstaying off with no message:/pluginandclaude plugin listnow say whyUserPromptSubmitor plugin prompt hook was still runningclaude -pcaller removed and re-added it, when a tool call had tried to reach it in betweenclaude -pruns printing nothing to stderr when a--plugin-dirmod's hook hangs or crashes the hooks worker--allowedToolsor anmcp_toolhook less than 2s before the first turn when settings loweredCLAUDE_CODE_MCP_STARTUP_WAIT_MSafter start-upcleanupPeriodDaysmanaged-settings.json,managed-settings.d) sometimes being missed at start-up while another program was rewriting the file: start-up now reads such a file a second time.claude/settings.local.jsonnot being ignored globally when the folder became a git repository later, or on a new machine/diffwaiting for Claude to finish before closing the diff panel in fullscreen/memoryshowing wrong or roundabout paths for folders whose names start with the working directory's or home folder's name, and losing slashes in the file path on WindowsCLAUDE_CODE_DISABLE_WEB_FETCHat each callhttps://claude --teleportopening an empty session when the server could not find the session's history, andclaude -p --teleportexiting without an error when the resume failedctl+k, binding the bare key; the binding is now ignored/btwanswers and recaps sometimes including earlier replies from the conversationCLAUDE_CODE_RESUME_INTERRUPTED_TURNre-running a turn that a hook'scontinue: falsehad stoppedgit add -Nbefore the first commitbrowser_batch, not showing the tool's name in the fullscreen transcript (ctrl+o) or in a subagent's progressclaude doctorand/statussaying a linked managed settings file is still read when it could not be read--chromesession giving no reason for starting without Claude in Chrome when your organization's policy turns it off: a line now says so/rewindlist on network paths and hung folders: it named every tracked file as changed, was slow to show labels, and showed none at all when a file read never returned--debugand--debug-filereplacing a file namedlatestin the debug log's folderclaude updatefinished removed theclaudecommand/modelshowing a picked effort level thatCLAUDE_CODE_EFFORT_LEVELoverridessparsePathslist without.claude, which left worktrees without your project settings and rules/modelpicker/upgradeand Ctrl+clicked links in fullscreen failing silently when no browser can open: the address is now shown/chromefinishes the connection; it now tells you to restart withclaude --chrome--no-session-persistence; they are now removed on normal exit, Ctrl+C or SIGTERMPreCompact,PostCompactand post-compactionSessionStarthooks not receivingagent_idandagent_typewhen a subagent compactsCLAUDE_CODE_TRANSCRIPT_LOCAL_GCdropping messages from the saved transcript when another process rewrote the file while it was being trimmed/usagehiding limit bars for setup tokens andCLAUDE_CODE_OAUTH_TOKEN: it keeps the session and weekly bars and says why the rest are missing/usageshowing a whole-session percentage beside a subagent type's name, and saying sessions ran 8 hours when it only counts hours with requestsclaude remote-controlreconnecting as a new computer when restarted in the same folder more than 4 hours after it stopped; it now keeps its identity for up to 30 daysDISABLE_TELEMETRYorDO_NOT_TRACKis setcommandwas a number, boolean or objectmanagedSettingsignoring the Remote Control, agent view, Workflows and Artifact off switchesotlpexporter had an endpoint but noOTEL_EXPORTER_OTLP_PROTOCOL: it now defaults tohttp/protobuflocalhost:4318endpoint going through a configured proxyOTEL_RESOURCE_ATTRIBUTESentry dropping all of that variable's attributes from the OpenTelemetry resource; the others are now kept, and/statusnames the one left offuser_promptevents and interaction spans recording a background agent's reply whenOTEL_LOG_ASSISTANT_RESPONSES=0OTEL_EXPORTER_OTLP_TIMEOUTcan allow (previously 30)agent.nameshowing a user-defined subagent's name instead ofcustomwhen an official plugin's skill launched itclaude agentsshowing a status from an earlier turn after a session resumesclaude -p, whenCLAUDE_CODE_SUBPROCESS_ENV_SCRUBis set on Linuxclaude agentslosing a typed prompt: Esc now needs a second press to clear it, and ctrl+y restores itclaude attachandclaude agentssessions; to detach with Ctrl+B d, press d within 800 ms of Ctrl+BDISABLE_ERROR_REPORTINGset only in your shell when the background service started without itclaude agentsafter replying to a blocked session, and repeated ones after a background turn ended with no text and only basic tools such as Bash or Readclaude -psilently using another model in text output when settings block the requested one; it now warns on stderrclaude -pprinting notes aboutdeniedModelsandavailableModelsentries on every run;/statusandclaude doctorstill list them.cmdor.batwrapper found beside a program's.exein the same PATH folder (agit.cmdnext togit.exe, say); it now runs the.exe, ascmd.exewould/clear,/resumeor/branch/loginandclaude setup-tokenfor a rejected, unreachable or timed-out request: they now say what happened and what to check/mcpto show how many of a server's tools were left out because the API would reject their input schema, and whycleanupPeriodDayscleanup to count folders it can't read as errors/v1/modelsto statemax_input_tokensfor models whose context window is certain on every upstreamToolSearchto say when a tool Claude searched for by its short name is already available, instead of only reporting no match/homeis an autofs map: Claude Code no longer looks up files in/homeitself. SetCLAUDE_CODE_AUTOMOUNT_ROOTSfor a map root deeper than one folder below/-p, SDK, IDE and cloud sessions: it now says when the limit resets, along with how to continue/security-reviewto ask before Claude runs it, and skills' inline git commands to stop in bare-repository worktrees and similar folders until you allow them, outside auto mode and bypassALL_PROXY; setHTTPS_PROXYif your network needs a proxy for them/autocompactto compact earlier, orCLAUDE_CODE_DISABLE_1M_CONTEXT=1to keep 200K/clear, so setCLAUDE_CODE_DISABLE_1M_CONTEXT=1before they fillMCP_SDK_GENERATION=v1to run the v2 MCP client in a compatibility mode (earlier handshake, relaxed OAuth sign-in checks), because the v1 MCP client is removeddontAskmode, use/mcpglab config setwithout--globalor--host, andglab config edit --local, to fail read-only inside the Bash sandbox; run them outside itsettings.local.jsonthat git tracks, lasts for the session only and isn't savedgit worktree add,moveandremovenow fail read-only, as do commits in a worktree the session wasn't started in or given, even one inside the project--settingsfile isn't valid JSON and has asandboxkey, even one that turns the sandbox offPreToolUsehook returned "allow": Claude Code now decides the call as if no hook had answeredPermissionRequesthook rewrote: Claude Code now asks again, or refuses the call where nobody can answerclaude mcp add --helpexample to a local command*, namespace or negated hook onengine.createnow only observes. To add or withhold a$noun, hookengine.createby nameclaude plugin validateto fail a mod whoseengine.createhook isn't written in the one accepted form; installed mods keep loading.mcp.json) MCP server approvals to match the server's exact name; an approval entry written in another spelling now asks again--trust-workspace falseand VS Code sessions in an untrusted folder: the repository's settings can no longer disable your hooks or turn plugins on or offtool.listplugin hooks so the tools they add or remove apply to the model and every agentclaude plugin installand/pluginto show a warning instead of success when a plugin's install command writes to stderrclaude rmand the agents view to say a session is removed from the list, not deleted;claude rmhelp says the conversation is kept andclaude --resumelists it/stickersto say that the sticker pack is all claimed for now, instead of opening a page in the browserpolicyHelperto no longer take environment variables from user or--settingsfiles; set what it needs in managedenvor the launching shell/skillor@filetyped between two files picked in claude.ai or the desktop app, or after a picked file and a link, to no longer take effect/clearor/resume.gitfolders are links or hold hard links, such as git-new-workdir layouts andcp -alcopies; use a plain cloneCLAUDE_CODE_OVERLOADED_RETRY_MAX_DELAY_MSfrom 64000 to 300000/usage-credits, in cloud sessions and under Remote Control: they no longer give a web address/logoutin a background session, which had no effect: it now signs out every session on the computer, deletes stored MCP logins and plugin secrets, and ends the session, as in a terminallsofand similar tools may show less about processes a command did not startPostToolUsehook from your own settings can no longer replace the plan inExitPlanMode's resultCLAUDE_AUTOCOMPACT_PCT_OVERRIDEin Anthropic-hosted cloud sessions to stop applying to auto-memory extraction and, in some sessions, to/btw; the main conversation still follows itsubagentStatusLinesetting: subagent rows always use the default look, and settings files that still hold the key load as beforeagentsetting: it still applies, and now warns at startup, inclaude plugin listand inclaude plugin validateConfiguration
📅 Schedule: (in timezone America/Chicago)
🚦 Automerge: Disabled by config. Please merge this manually once you are satisfied.
♻ Rebasing: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox.
🔕 Ignore: Close this PR and you won't be reminded about this update again.
This PR has been generated by Mend Renovate CLI.