Source Intelligence
Follow Discord

DisclaimerUnofficial, and not affiliated with Anthropic. Nearly all of this is read straight out of what ships: npm bundles, captured prompts, published docs. Anthropic's own notes go in verbatim, marked as theirs. The rest is my reading, and every entry carries the strings behind it. If one looks wrong, vote it down and say why.

← All releases Home ←olderv2.1.126 v2.1.129newer→

Claude Code v2.1.128

14 entries read diff v2.1.126 → v2.1.128 Markdown

Claude Code 2.1.128 adds a feature-flagged way to turn team onboarding guides into share links, adds a managed setting for organizations to disable Remote Control, and improves several background-session recovery paths. It also tightens provider-specific messaging for Channels/Fast Mode, expands session-only plugin handling, and makes large/media tool results fail more clearly instead of surfacing confusing output.

Find
Pick an entry · j / k steps through
9 entries

Changesopen

Team Onboarding Guide Sharing [Gradual Rollout]#

Feature flag
tengu_flint_harbor_share Off in both readings

The flag server returned off for the account this site reads and for the anonymous baseline. A reading of off cannot rule out a rollout these two readings sit outside of.

This account: off · anonymous baseline: off · compiled default in v2.1.128: no gate table built for this version

These values were read against a different version of Claude Code, so treat them as the nearest reading available instead of one taken on this release.

Read once, for one account on one subscription tier, against v2.1.128. It isn't a statement about your account. What a flag value here can and cannot tell you

What

The built-in team onboarding flow can now upload ONBOARDING.md and return a Claude Code share link for teammates.

Usage
/team-onboarding
Details
  • The existing onboarding guide flow now has a ShareOnboardingGuide tool that uploads ONBOARDING.md.
  • The tool supports three modes: check, update, and create.
  • If a guide already exists for the org, Claude asks whether to update the existing link or create a new one.
  • The guide must be in the current directory and under 64 KB.
  • This is gated by tengu_flint_harbor_share, so not every user will see it yet.
Evidence

Team onboarding share tool and rollout gate (search for "ShareOnboardingGuide" and "tengu_flint_harbor_share")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Managed Remote Control Disable Setting#

What

Organizations can now disable Remote Control centrally through managed settings.

Usage
{
  "disableRemoteControl": true
}
Details
  • The new setting disables claude.ai/code, claude remote-control, --remote-control, --rc, auto-start, and the in-session Remote Control toggle.
  • When policy blocks the feature, users now see a specific managed-setting error instead of a generic Remote Control denial.
Evidence

Managed setting schema and policy error (search for "disableRemoteControl" and "Remote Control is disabled by your organization's policy (managed setting disableRemoteControl).")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Background Session Recovery and Cleanup#

Claude Code is better at recovering background sessions that stall, crash, or lose their supervisor connection.

Usage
claude attach <id>
claude respawn <id>
claude respawn --all
claude rm <id>
Details
  • attach now attempts to wake sessions that report ENOJOB, reconnects through a transient service when possible, and gives clearer messages when a session moved, exited, or cannot be reattached.
  • Startup stalls can now trigger an automatic worker restart before giving up.
  • respawn and rm now have explicit --help output.
  • rm keeps changed worktrees instead of blindly removing them when uncommitted changes are present.
Evidence

Background recovery messages (search for "Session not responding — restarting it…", "Usage: claude respawn <id>|--all", and "Worktree has uncommitted changes — kept at")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Session-Only Plugin Zip Loading#

--plugin-dir now explicitly supports .zip files and reports when an inline plugin zip is extracted.

Usage
claude --plugin-dir ./my-plugin.zip
Details
  • Help text now documents repeatable directory or zip loading.
  • Zip extraction emits debug output, making plugin load failures easier to diagnose.
  • Existing managed-setting checks still apply to session-only plugins.
Evidence

Plugin flag help and zip extraction path (search for "Load a plugin from a directory or .zip" and "Extracted inline plugin zip to")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Related

Other releases about the same thing. Found by shared names or similar wording; neither means one caused the other.

MCP Tool Search Handles Late-Connecting Servers Better [Gradual Rollout]#

Feature flag
tengu_ashen_kelp On for this account, and not off by default

The flag server returned on for the one account this site reads, and nothing in this release compiles it off by default. The compiled default is shown below, and says which it is when we cannot read one: a fifth of gates compile in a string or a number rather than on or off, and most published releases have no gate table behind them at all. No client can see what the server returns for your account.

This account: on · anonymous baseline: on · compiled default in v2.1.128: no gate table built for this version

These values were read against a different version of Claude Code, so treat them as the nearest reading available instead of one taken on this release.

Read once, for one account on one subscription tier, against v2.1.128. It isn't a statement about your account. What a flag value here can and cannot tell you

Tool search can now refresh MCP clients and retry selection/search when matching tools may still be arriving from connecting servers.

Details
  • When enabled, direct select:<tool_name> can retry after an MCP refresh.
  • Keyword searches can also refresh before declaring that no tools match.
  • Claude receives clearer guidance that connecting MCP servers may expose tools shortly.
  • This behavior is gated by tengu_ashen_kelp.
Evidence

Deferred tool refresh behavior (search for "ToolSearchTool: partial select after MCP refresh" and "tengu_ashen_kelp")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Clearer Provider Restrictions for Channels and Fast Mode#

Provider-specific features now explain why they are unavailable instead of showing generic auth or availability errors.

Details
  • Channels now say they are unavailable on Bedrock, Vertex, or Foundry.
  • Fast Mode now says it is only available when using the Anthropic API directly.
  • Bedrock setup text consistently says “Amazon Bedrock.”
Evidence

Provider-specific messages (search for "Channels are not available on Bedrock, Vertex, or Foundry" and "Fast mode is only available when using the Anthropic API directly")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Related

Other releases about the same thing. Found by shared names or similar wording; neither means one caused the other.

Effort Selection Defaults Back to Auto#

Clearing effort now reports and applies auto instead of saying it set effort to max.

Usage
/effort auto
Details
  • The cache-warning dialog now uses “Change effort level?” when only effort changes.
  • Remote sessions get clearer messaging when an effort setting cannot reach the remote process.
Evidence

Effort reset and cache warning text (search for "Effort level set to auto" and "Change effort level?")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Better Large Input and Media Result Messages#

Claude Code now explains size and media limits more directly.

Details
  • Piped stdin that exceeds the size cap tells users to pass large content as a file path.
  • Oversized remote requests tell users to shorten the prompt or send it in parts.
  • Media rejected by the API is replaced with an explicit placeholder instead of failing opaquely.
  • PDF content returned by inner Read calls can be surfaced as document content blocks.
Evidence

User-facing limit messages (search for "piped stdin input exceeds", "request exceeds", "(media removed — rejected by API)", and "PDFs returned by inner Read calls")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Background Fleet Labels Are More Action-Oriented#

Blocked background work is now labeled as “Needs input,” making it clearer that user action is required.

Evidence

Fleet status label rename (search for "Needs input")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Related

Other releases about the same thing. Found by shared names or similar wording; neither means one caused the other.

1 entry

Bug Fixesopen

#

  • Auto mode fallback now suggests /compact when classifier context is too large, and stage-2 classifier errors say retrying often succeeds. Evidence: Auto mode fallback messages (search for "Auto mode classifier transcript exceeded context window — falling back to manual approval (try /compact to reduce conversation size)" and "Stage 2 classifier error - blocking based on stage 1 assessment (usually transient — retrying often succeeds)")
  • Voice input now pauses after repeated early failures instead of repeatedly starting failing sessions. Evidence: Voice circuit breaker (search for "Voice input is failing repeatedly and has been paused")
  • Remote egress gateway environments now propagate Google Cloud auth and CA certificate variables, improving GCP tooling inside proxied remote sessions. Evidence: GCP proxy environment propagation (search for "CLOUDSDK_AUTH_ACCESS_TOKEN" and "CLOUDSDK_CORE_CUSTOM_CA_CERTS_FILE")
  • MCP configuration now rejects reserved internal server names with clearer guidance. Evidence: Reserved MCP name validation (search for "is a reserved MCP name" and "Rename this server in your MCP config")
  • GitHub PR status fetching drops mergeStateStatus from the GraphQL fragment and adds REST status-list fallback strings, reducing reliance on a fragile PR field. Evidence: PR status query change (search for "fragment pr on PullRequest" and "[ghPrStatus] REST list")
Verbatim
Official · Anthropic

Anthropic’s official release notes

Published verbatim by Anthropic for v2.1.128. Text is unmodified from the upstream changelog. Everything else on this page came out of the bundle instead, which is why the two lists don't match.

  • Bare /color (no args) now picks a random session color
  • /mcp now shows the tool count for connected servers and flags servers that connected with 0 tools
  • --plugin-dir now accepts .zip plugin archives in addition to directories
  • --channels now works with console (API key) authentication — console orgs with managed settings must set channelsEnabled: true to enable
  • Updated /model picker: collapsed duplicate Opus 4.7 entries, and current Opus now shows as "Opus" instead of "Opus 4.7"
  • Subprocesses (Bash, hooks, MCP, LSP) no longer inherit OTEL_* environment variables, so OTEL-instrumented apps run via the Bash tool no longer pick up the CLI's own OTLP endpoint
  • MCP: workspace is now a reserved server name — existing servers with that name will be skipped with a warning
  • Reconnecting MCP servers no longer flood the conversation with full tool-name lists on every reconnect — re-announced tools are summarized by server prefix
  • SDK hosts now receive a persistent localSettings suggestion for Bash permission prompts, so "Always allow" writes to .claude/settings.local.json
  • EnterWorktree now creates the new branch from local HEAD as documented, instead of origin/<default-branch> — unpushed commits are no longer dropped
  • Auto mode: when the classifier can't evaluate an action, the error now includes a hint (retry, /compact, or run with --debug)
  • Fixed focus mode briefly dimming the previous response when submitting a new prompt
  • Fixed stray "4;0;" desktop notification on every /exit in Kitty and other terminals that interpret OSC 9 as a notification
  • Fixed Remote Control showing an empty "Opening your options…" message on rate limit instead of actionable upsell options
  • Fixed drag-and-drop image upload hanging on "Pasting text…" when the image read fails
  • Fixed crash loop when piping very large input (>10 MB) to claude -p via stdin
  • Fixed long URLs not being individually clickable on every wrapped row in fullscreen mode
  • Fixed /plugin Components panel showing "Marketplace 'inline' not found" for plugins loaded via --plugin-dir
  • Fixed MCP tool results dropping images when the server returns both structured content and content blocks
  • Fixed fenced code blocks inside list items carrying leading whitespace into the clipboard on copy-paste
  • Fixed tab navigation in /config stranding focus — the tab header now stays focused so arrows and Esc keep working
  • Fixed markdown link labels being lost on terminals without OSC 8 hyperlink support — links now render as label (url) instead of just the URL
  • Fixed sessions on 1M-context models with a smaller autocompact window being falsely blocked with "Prompt is too long" before reaching the actual API limit
  • Fixed parallel shell tool calls: a failing read-only command (grep, git diff, ls) no longer cancels sibling calls
  • Fixed banner showing "with X effort" on models that don't support effort
  • Fixed /fast on 3P providers fuzzy-matching to an unrelated skill instead of showing "not available"
  • Fixed Bedrock default model resolving to global.* instead of the region-appropriate prefix
  • Fixed vim mode: Space in NORMAL mode now moves the cursor right, matching standard vi/vim behavior
  • Fixed terminal progress indicator (OSC 9;4) flickering off between tool calls — stays visible across the full turn
  • Fixed /rename without args failing on resumed sessions whose last entry is a compact boundary
  • Fixed stale "remote-control is active" status lines from prior sessions appearing after --resume/--continue
  • Fixed stale installed_plugins.json entries pointing at deleted cache directories polluting PATH
  • Fixed MCP stdio servers receiving corrupted arguments when CLAUDE_CODE_SHELL_PREFIX is set and an argument contains spaces or shell metacharacters
  • Fixed sub-agent progress summaries missing the prompt cache (~3× cache_creation reduction)
  • Fixed /plugin update never detecting new versions of npm-sourced plugins
  • Fixed sub-agent summaries firing repeatedly while a sub-agent's transcript is static, capping worst-case token cost on idle sub-agents
  • Headless --output-format stream-json: init.plugin_errors now includes --plugin-dir load failures in addition to dependency demotions
Read it on GitHub →
System prompt

1 of 24 tool descriptions changed.

Claude Code, interactive mode

Read the prompts for this release →