Verification Report

Rechecked 2026-10-23 · Claude Code 2.1.308

Detailed changes and primary sources. Earlier checks below retain their original baselines.

October 10, 2026 release refresh

Checked Claude Code v2.1.293–v2.1.296 (October 7–9) and the Claude Haiku 5.5 docs. npm checked on October 10, 2026 KST: latest/next=2.1.296, stable=2.1.287. This compares official text; Haiku 5.5 was not exercised in a local CLI.

• Haiku 5.5 (claude-haiku-5-5, 2026-10-07) requires v2.1.293+; haiku resolves to it on the Anthropic API (Claude Platform on AWS, Bedrock, Agent Platform, Foundry: Haiku 4.5). Source: changelog 2.1.293, model-config, Haiku 5.5 overview.
• 1M context, 128K output, $0.10/$0.50 per MTok for prompts up to 100,000 tokens and $0.50/$2.50 above that, cache reads $0.01 and $0.05, no retirement before 2027-10-07. Default effort in Claude Code is medium, thinking cannot be turned off, and sessions auto-compact at about 967K tokens. Source: Haiku 5.5 overview, pricing, model-config.
• Corrected: Sonnet 5.5 cache reads cost $0.10/MTok (5% of input), not the $0.20 recorded on October 7. Changelog 2.1.296 moves /cost and the other client-side figures from $0.20 to $0.10. Source: pricing, changelog 2.1.296.
• Hooks: onFailure on command and HTTP hooks defaults to "continue", and "block" requires v2.1.295+; Stop and SubagentStop prompt hooks written as instructions are judged differently from v2.1.294. Source: hooks docs, changelog 2.1.294 and 2.1.295.
• Subagents: skills preloads up to the first 32 distinct names. Source: sub-agents docs, changelog 2.1.295.
• Official tables: still 119 slash rows and 80 flags; the CLI table has 36 rows with claude daemon logs and claude daemon run. Source: Commands, CLI reference.
• claude.ai connectors negotiate protocol revision 2026-07-28 by default on installs that fetch no feature flags, with MCP_PROTOCOL_NEGOTIATION=legacy as the opt-out. Source: changelog 2.1.295, MCP docs.
• Changelog-only items, not yet in the docs compared on October 10: the subagent autoCompactWindow field (2.1.296; absent from the sub-agents frontmatter table), the MCP description limits of 4,096 characters up front and 16,384 through tool search (2.1.296 and 2.1.295; the MCP and env-vars docs still say 2,048), and CLAUDE_CODE_OVERLOADED_RETRY_MAX_DELAY_MS (2.1.296; CLAUDE_CODE_RETRY_WATCHDOG_MAX_WAIT_MS is in the env-vars docs with v2.1.295+).
• Version mismatches kept visible: the auto mode starting default (changelog v2.1.284, Permission modes reference v2.1.283 or later) is unchanged, with no new evidence in this review. For stdio MCP negotiation, the changelog says every install from v2.1.292; the MCP and env-vars docs as of October 10 say only that Claude Code asks stdio servers by default and give no starting version, and the staged-rollout wording recorded on October 7 was not found, so the changelog remains the only version evidence.
• Other v2.1.293–v2.1.296 changes checked against the changelog: the 16 MiB WebSocket message limit, turns stopped while ← backgrounds a session, quoted text in /copy, CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL, and agentType in the subagentStatusLine payload.

Sources: official changelog, Model configuration, Hooks, Subagents, MCP, Environment variables, CLI reference, Claude Haiku 5.5 overview, Pricing, npm dist-tags.

October 7, 2026 release refresh (previous)

Checked Claude Code v2.1.284–v2.1.292 (September 28–October 6) and the Claude Sonnet 5.5 docs. npm checked on October 7, 2026 KST: latest/next=2.1.292, stable=2.1.285. This compares official text; Sonnet 5.5 was not exercised in a local CLI.

• Sonnet 5.5 (claude-sonnet-5-5, 2026-09-28) requires v2.1.284+; sonnet resolves to it on the Anthropic API (Claude Platform on AWS: Sonnet 4.6; Bedrock, Agent Platform, Foundry: Sonnet 4.5). Source: changelog 2.1.284, model-config, Sonnet 5.5 overview.
• 1M context, 128K output, $2/$10 per MTok, $0.20 cache reads, no retirement before 2027-09-28. Default effort in Claude Code is medium and thinking cannot be turned off; the platform model table lists high as the Messages API default. Source: Sonnet 5.5 overview, pricing, model-config.
• Classifier fallback: Sonnet 5.5 cybersecurity flags rerun on Sonnet 5; biology flags are refused. Source: model-config.
• Ultracode is a toggle separate from effort with /effort ultracode [on|off] (v2.1.284+). Source: model-config, Commands table.
• Official tables: 119 slash rows, --desktop, claude purge, and session names for claude attach and claude logs. Source: Commands, CLI reference.
• Version mismatch kept visible: the changelog records auto mode as the every-plan starting mode under v2.1.284, while the Permission modes reference states v2.1.283 or later. Likewise the changelog says stdio MCP servers negotiate protocol revision 2026-07-28 on every install from v2.1.292, while the MCP docs describe a staged rollout from v2.1.285.
• Resolved: the CLI reference now agrees with changelog 2.1.283 that --system-prompt can be combined with its -file form.

Sources: official changelog, Model configuration, Permission modes, MCP, Claude Sonnet 5.5 overview, npm dist-tags.

September 26, 2026 release refresh (previous)

Checked Claude Code v2.1.280–v2.1.283 (September 22–25) and the Claude Opus 5.5 docs. npm checked on September 26, 2026 KST: latest/next=2.1.283, stable=2.1.274. This compares official text; Opus 5.5 was not exercised in a local CLI.

• Opus 5.5 (claude-opus-5-5, 2026-09-22) requires v2.1.280+; opus resolves to it on the Anthropic API, Claude Platform on AWS, Bedrock, and Google Cloud Agent Platform (Foundry: Opus 4.6), and default resolves to it on Pro, Max, Team, Enterprise, the API, and those providers (Foundry: Sonnet 4.5). Source: model-config, changelog 2.1.280.
• 1M context, 128K output, $4/$20 per MTok, 5% cache reads, no retirement before 2027-09-22, default effort medium, always-on thinking, and the top-level user effortLevel exception. Source: Opus 5.5 overview, what's-new, model-config.
• Fast mode defaults to Opus 5.5 at $8/$40 (Opus 5/4.8 at $10/$50). Source: fast-mode.
• Classifier fallback: Fable 5.1/5 and Opus 5.5 biology to Opus 5, cybersecurity to Opus 4.8; Opus 5 refuses biology flags. Source: model-config.
• deniedModels and availableModelsMatch (v2.1.283+) and modelPricing fields (v2.1.242+). Source: model-config, settings-reference.
• Known docs mismatch: changelog 2.1.283 allows --system-prompt together with its -file form, while the CLI reference still calls them mutually exclusive.

Sources: official changelog, Model configuration, Fast mode, Settings reference, Claude Opus 5.5 overview, What's new in Claude Opus 5.5.

September 12, 2026 release refresh (previous)

Updated for Claude Code v2.1.263–v2.1.269 (September 6–11). npm checked on September 12, 2026 KST: latest/next=2.1.269, stable=2.1.236. Keep latest-feature evaluation separate from a team's stable-channel rollout.

• Plugins: claude plugin eval produces JSON and HTML evaluation reports; install, uninstall, update, enable, and disable accept --json. --plugin-dir can point to a parent directory containing multiple plugins.
• Commands and settings: /output-style [name] works over Remote Control and in cloud/headless sessions. maxEffortLevel caps effort globally or per model under modelSettings. bashEditDiffEnabled adds file-edit diffs to Bash results.
• Recovery and compatibility: 2.1.266 fixes forced gateway sign-in; 2.1.268 fixes HTTP 400s on Anthropic-compatible endpoints introduced in 2.1.265. /goal retries with backoff or pauses with a reason after API, network, or usage-limit failures.
• Permissions and operations: fixes cover symlink deny/ask rules, Bash tee write boundaries, and secrets in MCP/plugin errors. WebFetch has a default 300-second deadline. /ultrareview --post posts the PR comment directly when findings arrive.

Sources: official changelog, npm dist-tags.

This refresh compares official release text with npm channels. It does not claim local CLI execution or live feature testing. Tables below retain prior dated checks; the channel values above supersede them.

The English locale should behave as a first-class locale for supported content and as unavailable for unsupported books.

Expected Behavior

• /en renders the English home page.
• /en lists only English-localized handbooks.
• /en/books/claude-code-advanced renders English content.
• /en/books/claude-code-command-master renders English content.
• /en/books/codex-advanced does not fall back to Korean content.
• The preview chapter API receives a locale parameter.
• Sitemap generation includes English URLs only for English pages.

Local Checks

Run the focused handbook checks after changing locale routing or content availability:

yarn workspace handbook typecheck
yarn workspace handbook lint
yarn workspace handbook build
yarn workspace handbook test:e2e

If an E2E test fails, inspect whether it is asserting Korean-only copy or a route that is now locale-specific.

Content Checks

• Every English page has English frontmatter.
• Internal links use /en/books/....
• Korean-only books are not registered in content/books/en/meta.json; translated books are.
• The home catalog uses localized availability from the MDX source.
• Language links appear only when the same page exists in more than one locale.
• The current baseline is the 2026-09-26 refresh above (v2.1.283). The earlier 2026-09-05 check baseline was: v2.1.261 (latest and next) and v2.1.236 (stable) from npm dist-tags, plus the official changelog through v2.1.261 and the Claude Fable 5.1 overview, what's-new, migration guide, and announcement.
• The v2.1.252~261 refresh covers Fable 5.1 as the default Fable model (released 2026-09-01, minimum v2.1.255, default from v2.1.257), gateway sessions staying on Fable 5, the category-based classifier fallback table (biology to Opus 5, cybersecurity to Opus 4.8, switchModelsOnFlag), Fable 5.1 Messages API breaking changes and additive features, $0.25/MTok cache reads, the /effort session-only s key, no default-effort hold on Opus 5 and Fable 5.1, CLAUDE_CODE_SUBAGENT_MODEL_FORCE, the v2.1.260 picker/prompt-cache/[1m] fixes, 1M auto-compact timing, and the Mythos 5.1 access distinction. Fast mode stays Opus 5/4.8 only.
• The v2.1.239~247 refresh covers loop usage statistics, managed model/cache controls, Auto-mode rules, partial subagent continuation, headless recovery, feedback drafts, and security hardening.
• The v2.1.221~237 refresh covers interactive fork/background spawn defaults, cross-session messaging and session mentions, self-hosted runners, GitLab MR workflows, ANTHROPIC_DEFAULT_MODEL, the Concise output style, newer-model Task/Todo defaults, and trust/sandbox/OAuth/Remote hardening.
• The v2.1.206~v2.1.220 refresh covers Opus 5, the /fork and /subtask split, session/concurrency/depth limits for subagents, strict network allowlisting, optional filesystem isolation, DirectoryAdded, headless MCP validation errors, and nested subagent stream forwarding.
• The v2.1.181~v2.1.205 refresh covers Sonnet 5, org default model labeling, sandbox credentials, destructive-command sandbox auto-approval, Windows PowerShell sandbox fixes, MCP login/logout, MCP idle timeout, model restrictions, deprecated-model warnings, nested-agent/background stability, Korean paste handling, and non-interactive fallback fixes. v2.1.198–201 add background-by-default subagents, Claude in Chrome GA, the /dataviz skill, the default permission mode "Manual" (v2.1.200), and removal of the mid-conversation system role on Sonnet 5 sessions (v2.1.201).
• v2.1.202–205 add advisory workflow sizing and OTel attributes, the fast single-pass /review split, MCP root updates, background/Agent View reliability, headless hook streaming, the full /doctor checkup and /checkup alias, Auto-mode transcript protection, and JSON Schema fixes.
• The v2.1.170 content remains valid for Fable 5 and Agent View, while the current Opus alias and default high-capability path moved to Opus 5 in v2.1.219 and to Opus 5.5 in v2.1.280. Historical Opus 4.8 guidance is retained only where a version pin or compatibility path is explicit.
• Background sessions, dynamic workflows, and the command surface were rechecked against current docs.
• Fable-specific details include 30-day data retention, ZDR unavailability, always-on adaptive thinking, DISABLE_PROMPT_CACHING_FABLE, classifier fallback, and availability fallback chains.
• Fable 5 redeployment: suspended 2026-06-12 under US export controls, restored globally 2026-07-01 with an improved safety classifier; plan usage is up to 50% of weekly limits through 2026-07-07 then usage credits. Mythos 5 (no classifiers) is Project Glasswing invitation-only and not selectable via /model.
• Recent operating notes include /cd session relocation, --safe-mode, bundled skill/workflow disabling, self-hosted runner post-session hooks, and the v2.1.170 transcript persistence fix.

Source Baseline

• Claude Code Changelog: https://code.claude.com/docs/en/changelog
• Model configuration: https://code.claude.com/docs/en/model-config
• Commands: https://code.claude.com/docs/en/commands
• Subagents: https://code.claude.com/docs/en/sub-agents
• Hooks: https://code.claude.com/docs/en/hooks
• Sandboxing: https://code.claude.com/docs/en/sandboxing
• Agent View: https://code.claude.com/docs/en/agent-view
• Sonnet 5 announcement: https://www.anthropic.com/news/claude-sonnet-5
• Claude Opus 5.5 overview: https://platform.claude.com/docs/en/models/opus-5-5/overview
• What's new in Claude Opus 5.5: https://platform.claude.com/docs/en/models/opus-5-5/whats-new-opus-5-5
• Settings reference: https://code.claude.com/docs/en/settings-reference
• Claude Fable 5.1 overview: https://platform.claude.com/docs/en/models/fable-5-1/overview
• What's new in Claude Fable 5.1: https://platform.claude.com/docs/en/models/fable-5-1/whats-new-fable-5-1
• Migrating to Claude Fable 5.1: https://platform.claude.com/docs/en/models/fable-5-1/migration-guide
• Fable 5.1 and Mythos 5.1 announcement: https://www.anthropic.com/claude-fable-and-mythos-5-1
• Fast mode: https://code.claude.com/docs/en/fast-mode
• Claude Fable 5 and Claude Mythos 5: https://platform.claude.com/docs/en/about-claude/models/introducing-claude-fable-5-and-claude-mythos-5
• Fable 5 redeployment statement: https://www.anthropic.com/news/redeploying-fable-5
• Claude models overview: https://platform.claude.com/docs/en/about-claude/models/overview
• Opus 4.8 announcement: https://www.anthropic.com/news/claude-opus-4-8
• npm package: https://www.npmjs.com/package/@anthropic-ai/claude-code

Residual Risk

The English locale now has multiple localized handbooks. Shared pages such as About and Graph have English UI labels, but deeper editorial polish should be handled as a separate content pass.