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 ←olderv2.1.64 v2.1.67newer→

Claude Code v2.1.66

17 entries read diff v2.1.64 → v2.1.66 Markdown

This release removes the ultraplan mode and the remote-control server subcommand, replaces the /reload-plugins workflow with a restart-based plugin activation model, and enhances the microcompaction system with boundary messages and tool result clearing. It also introduces a new forkContext property for agent definitions, improved parallel tool call error handling, and a redesigned Plugin Status UI.

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

Changesopen

Sub-Agent Context Forking (forkContext)#

What

Agents can now receive the full parent conversation context when spawned as sub-agents, enabling better awareness of what the user has been working on.

Details
  • New forkContext frontmatter field for agent .md definitions (set to true or false)
  • When enabled, the sub-agent receives prior conversation messages as read-only context
  • Agents with forkContext: true must use model: inherit to avoid context length mismatch
  • A new ### FORKING CONVERSATION CONTEXT ### boundary marker is injected to clearly separate inherited context from the sub-agent's own task
  • The sub-agent is instructed to only use tools available in its own system prompt, even if context messages reference other tools
Evidence

Sub-agent entry system (search for "sub_agent_entered" and "FORKING CONVERSATION CONTEXT")

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

Plugin Status and Error UI#

What

New interactive UI views for inspecting plugin installation status and diagnosing loading errors.

Details
  • Plugin Status view shows installation progress for marketplaces and plugins (pending, installing, installed, failed)
  • Plugin Loading Errors view lists all plugin load failures with detailed error messages and actionable remediation hints
  • Error types include: path-not-found, git-auth-failed, network-error, manifest-parse-error, marketplace-blocked-by-policy, and more
  • Each error includes a → arrow hint suggesting how to fix the issue (e.g., "→ Check your internet connection and try again", "→ Configure SSH keys or use HTTPS URL instead")
Evidence

Plugin error UI (search for "Plugin Loading Errors" and "Plugin Status")

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

Background Plugin Installation#

What

Plugins and marketplaces from settings are now automatically installed in the background when they are missing.

Details
  • On startup, Claude Code detects plugins referenced in settings that are not yet installed
  • Missing marketplaces are installed first, then missing plugins from those marketplaces
  • Installation progress is tracked per-plugin with states: pending → installing → installed/failed
  • Plugins from unknown/unconfigured marketplaces are flagged as uninstallable with clear error messages
  • New telemetry event tengu_marketplace_background_install tracks installation success/failure rates
Evidence

Background plugin installer (search for "plugins automatically" and "Setting installation status")

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.

Plugin Enable/Disable with Scope Support#

What

Plugins can now be enabled or disabled at specific scopes (user, project, local), with validation for managed plugins.

Details
  • Enabling/disabling a plugin now persists the setting at the appropriate scope
  • Managed plugins (from enterprise policy) cannot be enabled or disabled — only updated
  • If a --scope argument is provided but doesn't match the plugin's actual scope, a helpful error message is shown
  • Success messages include the plugin name and scope: "Successfully enabled plugin: X (scope: user)"
Evidence

Plugin scope management (search for "Managed plugins cannot be" and "is installed at")

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

Restart-Based Plugin Activation (Replaces /reload-plugins)#

The /reload-plugins command has been removed. After installing, enabling, disabling, or updating plugins, Claude Code now displays messages like:

  • "✓ Installed X. Restart Claude Code to load new plugins."
  • "Configuration saved. Restart Claude Code for changes to take effect."

This replaces the previous "Run /reload-plugins to activate" workflow with a cleaner restart model.

Evidence

Plugin restart messages (search for "Restart Claude Code")

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.

Enhanced Microcompaction with Boundary Messages#

The automatic context microcompaction system now produces visible boundary messages when it operates. When tool results are cleared to save context space, a "Context microcompacted" system message is inserted with metadata including token counts saved, number of tool results compacted, and images cleared. Old tool results are replaced with "[Old tool result content cleared]" markers that point to saved files on disk.

Evidence

Microcompaction boundaries (search for "Context microcompacted" and "[Old tool result content cleared]")

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.

Sibling Tool Call Error Handling#

When multiple tools run in parallel and one fails, the remaining sibling tool calls now receive a clean error message ("Sibling tool call errored") instead of hanging or producing ambiguous results. This improves reliability when using parallel tool execution.

Evidence

Parallel tool error handling (search for "Sibling tool call errored")

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.

Updated Input Guides#

New contextual keyboard shortcut hints appear during various interactions:

  • "ctrl+c to interrupt · Esc again to clear" during active operations
  • "Tab to complete · Enter to add · Esc to cancel" in settings input
  • "↑/↓ to select · Enter to confirm · Esc to cancel" in selection lists
Evidence

Input guides (search for "ctrl+c to interrupt")

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.

Remote Control Simplified#

The claude remote-control command has been simplified. The server subcommand and its associated options (--spawn-worktree-sessions, --spawn-same-dir-sessions, --max-sessions, --idle-timeout, --workspace, --host, --port) have been removed. The --name option has also been removed. Remote control now focuses solely on connecting a single terminal for remote sessions via claude.ai/code.

Evidence

Simplified remote-control help text (search for "Remote Control - Connect your local environment")

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.

Output Efficiency System Upgraded [Gradual Rollout]#

Feature flag
tengu_sotto_voce 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 v2.1.66: 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.

tengu_swann_brevity Not enough to say

Nothing here resolved what this flag was doing on this version, so nothing here should be read as on or off.

This account: no value returned · anonymous baseline: no value returned · compiled default in v2.1.66: 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 v2.1.66. It isn't a statement about your account. What a flag value here can and cannot tell you

The output brevity system has been upgraded from a single on/off toggle (tengu_sotto_voce) to a three-tier system (tengu_swann_brevity) with modes:

  • strict: Extremely concise — "Use the fewest words necessary. Omit preamble, filler, pleasantries."
  • focused: Extra concise — "Lead with the answer or action, not the reasoning."
  • polished: Balanced — "Concise and polished. Never omit important information."
Evidence

Output efficiency modes (search for "tengu_swann_brevity")

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.

Model Update Notification#

The model migration notification has changed from "Model updated to Opus 4.6" to "Model updated to Opus 4.5", reflecting changes in default model availability for certain account types.

Evidence

Model migration notification (search for "Model updated to Opus 4.5")

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.

Ultraplan Mode Removed#

The ultraplan mode has been completely removed. This includes the ultraplan and ultraplan-mode identifiers, the ultraplan completion messages, and the verification specialist agent that was part of the ultraplan workflow.

Evidence

All "ultraplan" references removed from source

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

Scheduled Skills Removed#

The scheduled skills system (cron-based skill execution) has been removed. This includes the schedule display formatting ("Every hour", "Every day at", "Weekdays at"), the scheduled skills UI, and the scheduled skills status line.

Evidence

All "scheduled skill" and "Scheduled skill" references removed

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.

Voice Mode Tip Removed#

The tip "Voice mode is now available · /voice to enable" has been removed.

Evidence

String "Voice mode is now available" removed

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.

Claude Code Desktop Cross-Promotion Removed#

The following promotional elements have been removed:

  • "Try Claude Code Desktop" callout
  • "Open in Claude Code Desktop" button/link
  • "Same Claude Code with visual diffs, live app preview, parallel sessions, and more" description
Evidence

All Desktop cross-promotion strings removed

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.

/btw Side Question Tip Removed#

The tip text "Use /btw to ask a quick side question without interrupting Claude's current work" has been removed. Note: The /btw command itself still exists and functions — only the tip promoting it has been removed.

Evidence

Tip string "Use /btw to ask a quick side question" removed; /btw regex pattern still present (search for "/btw")

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.

Code Intelligence Description Removed#

The description string "code intelligence (definitions, references, symbols, hover)" for LSP features has been removed from the codebase.

Evidence

String "code intelligence (definitions, references, symbols, hover)" removed

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

Verbatim
Official · Anthropic

Anthropic’s official release notes

Published verbatim by Anthropic for v2.1.66. 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.

  • Reduced spurious error logging
Read it on GitHub →
System prompt

The system prompt was not captured for this release, so this page cannot say whether it moved.

Which releases were captured →