Release coverage history

2026-10-23 · Claude Code 2.1.308

Compared v2.1.299–v2.1.308 in the official changelog with Commands and the CLI reference. npm latest, next, and stable all resolve to 2.1.308. The fetched changelog has no entries for v2.1.297 or 298, so their changes are not inferred.

• Permission defaults (v2.1.302): without an explicit mode, claude -p and Agent SDK sessions on third-party providers, with telemetry disabled, or on fresh installs start in default. This supersedes the v2.1.285 auto-default guidance; --permission-mode auto remains an explicit option.

• Skills and hooks (v2.1.299–304): UserPromptSubmit runs before skill or custom-command expansion; blocking it also prevents embedded shell execution. A skill's own hook starts with the following prompt. Tool validation checks updatedInput from PreToolUse and PermissionRequest. At most 12 prompt/agent hooks of an event run concurrently; skill hooks from one source are limited to 1,024 hooks and 16,384 matcher characters per session.

• Subagents (v2.1.302): fork subagents are on by default in -p, the SDK, Desktop, and VS Code (CLAUDE_CODE_FORK_SUBAGENT=0 disables them; claude mcp serve excludes them). Main-conversation subagents run in the background, while their own Agent calls default to foreground in those environments.

• Context (v2.1.299, 302, 304): idle compaction compresses a 200K+ token conversation after roughly 54 idle minutes. /config exposes its toggle; idleCompaction: false disables it and CLAUDE_CODE_IDLE_COMPACT_MIN_TOKENS sets the threshold. The changelog states v2.1.290+ for both settings. Compaction resends the skill list; /reload-skills sends only new or changed skills.

• API and Design: /claude-api managed-agents-secure (v2.1.304) compares configuration against the security guide without modifying it; api-version (v2.1.307) helps migrate to API version 2026-10-21. The reference adds /design consent and /design revoke; the user must enter consent themselves. Standalone claude.ai/design, used by /design-login and /design-sync, closes by 2026-12-14.

• MCP and policy (v2.1.299–307): an mcp__a allow rule no longer reaches server a__b; use mcp__<server>__* with its complete normalized name. From v2.1.305, managed-mcp.json stdio servers using ${VAR} only in env are not filtered by allowedMcpServers; block them explicitly with deniedMcpServers. v2.1.307 limits HTTP MCP retries after disconnects to read-only or idempotent tools.

• Operations and diagnostics: claude remote-control --sandbox and --no-sandbox fail from v2.1.300; use sandbox.enabled. Runner checkout hooks default to 30 minutes in v2.1.307 (--checkout-hook-timeout-min, 0 disables the limit). The claude_code.safety_event OTel event records blocked or prompted tool calls. v2.1.308 announces only general fixes and reliability improvements.

• Security and recovery: reviewed fixes for hook-approved writes in plan mode, sandbox symlink boundaries, third-party gateway credential forwarding, duplicate tool execution (v2.1.299), and worktree removal confirmation (v2.1.301, 304). sandbox.allowNestedUserNamespaces (v2.1.305) can bypass filesystem protections; review isolation before enabling it for Chromium or rootless podman on Linux.

• Code Review and mods: @claude review stop (off or pause) ends a PR’s always-review subscription without starting another review (v2.1.307). v2.1.304 adds e.isReadOnly and decider: "person" to mod tool.check; the latter requires a permission dialog or host callback and denies with the mod reason when neither is available.

Sources: Changelog, Commands, CLI reference, Hooks, Subagents, npm dist-tags.

Key takeaways

• This page preserves earlier review records. Use the detail guides and complete list for current behavior. Earlier minimum-version notes do not establish a newly verified introduction version.

Return to the current guide

v2.1.293–v2.1.296 coverage — October 10, 2026

Reviewed the official changelog from v2.1.293 (October 7) through v2.1.296 (October 9). The official slash table still has 119 rows; only the /doctor, /mcp, and /scroll-speed descriptions changed. The CLI table grew from 34 to 36 rows with claude daemon logs and claude daemon run. The flag table still has 80 rows, with new wording for --max-budget-usd, --mcp-config, and --forward-subagent-text.

• Haiku 5.5 (v2.1.293): claude-haiku-5-5 is the default Haiku on the Anthropic API (1M context, 128K output). It costs $0.10/$0.50 for prompts up to 100,000 tokens and $0.50/$2.50 above that. Default effort in Claude Code is medium, and thinking cannot be turned off. haiku still resolves to Haiku 4.5 on Claude Platform on AWS, Bedrock, Google Cloud Agent Platform, and Foundry. v2.1.296 fixed Haiku 5.5 token counts that failed behind some gateways.
• Cost figures (v2.1.296): /cost, the status line, --max-budget-usd, and the SDK's cost figures price Sonnet 5.5 cache reads at $0.10/MTok (was $0.20). The official price list also gives $0.10/MTok, so this corrects the $0.20 in the October 7 entry below. The CLI reference now says of --max-budget-usd that spend can pass the cap, so leave headroom.
• Hooks: Command and HTTP hooks gained onFailure: "block": a hook that can't start, times out, or exits with an unexpected code blocks the action instead of letting it through (v2.1.295). prompt hooks on Stop and SubagentStop that are written as instructions, such as "Carry on if the build is broken", are judged so that Claude is less likely to stop early (v2.1.294). --debug logs the command, plugin, outcome, and duration of command hooks on tool calls, prompts, SessionStart, and Stop when they finish (v2.1.296).
• Subagents: autoCompactWindow was added to subagent frontmatter and --agents definitions, so a subagent can auto-compact earlier than the main conversation's window (v2.1.296). The skills field preloads at most 32 skills (v2.1.295), and the subagentStatusLine payload includes agentType (v2.1.293). --debug names unrecognized frontmatter fields in custom agent files (v2.1.296).
• MCP: The default limit on MCP tool descriptions sent up front and on server instructions rose from 2,048 to 4,096 characters (v2.1.296). Tool descriptions loaded through tool search are cut at 16,384 characters instead of 2,048 (v2.1.295). WebSocket (ws) servers no longer parse a message over 16 MiB and close the connection (v2.1.295). On installs that fetch no feature flags, claude.ai connectors also negotiate protocol revision 2026-07-28 by default, with MCP_PROTOCOL_NEGOTIATION=legacy as the opt-out (v2.1.295).
• Backgrounding and input: A turn or ! command started while ← moves the session to the background is stopped instead of finishing out of sight (v2.1.296). Stopping the turn while ← waits for the current tool shows "Backgrounding cancelled" (v2.1.295). Ctrl+C at the idle prompt of an attached background session leaves a pending /loop wakeup alone; Esc stops the loop (v2.1.295). The /copy picker lists quoted text, so a drafted message copies without its > markers (v2.1.295).
• CLI and plugins: claude plugin install, enable, disable, and marketplace add warn when the settings file they write to does not load. claude plugin validate prints the install line to paste when a plugin's README has none, without changing the exit code even with --strict (all v2.1.295).
• Settings and environment variables: CLAUDE_CODE_RETRY_WATCHDOG_MAX_WAIT_MS (v2.1.295), and CLAUDE_CODE_OVERLOADED_RETRY_MAX_DELAY_MS and CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL (both v2.1.296). On machines with no managed settings, forceLoginMethod: "gateway" and forceLoginGatewayUrl in user settings make /login open on that Claude apps gateway (v2.1.295).
• Tools and terminal: The Read tool gained an allow_large option (v2.1.296), and Grep runs a search sent with the -l, -c, or -r flag instead of failing it (v2.1.295). Terminals that implement the Program Status Protocol (OSC 7501) can show whether Claude Code is working, waiting on you, or done (v2.1.295).
• Reverted changes (v2.1.293): The v2.1.281 change to the auto mode denial message was reverted. claude.ai skill syncing checks for changes about every 40 minutes, instead of every 10, while no session is in use.

Sources: Changelog, Model configuration, Hooks, Subagents, CLI reference, Haiku 5.5 overview, Pricing.

v2.1.284–v2.1.292 coverage — October 7, 2026

Reviewed the official changelog from v2.1.284 (September 28) through v2.1.292 (October 6); v2.1.291 was published with no entries. The official slash table grew from 114 to 119 rows and six rows changed syntax. Three CLI rows changed syntax, and the flag table gained --desktop.

• Sonnet 5.5 (v2.1.284): claude-sonnet-5-5 is the default Sonnet on the Anthropic API (1M context, $2/$10, cache reads $0.20). Default effort in Claude Code is medium, and thinking cannot be turned off. Cybersecurity flags rerun on Sonnet 5; biology flags end with a refusal. The auto mode classifier ignores a Sonnet pin that names Sonnet 5.5 or Opus 5.5 and uses Sonnet 5 (v2.1.288).
• Ultracode and effort: Ultracode is its own toggle in /effort and no longer forces xhigh (v2.1.284). /effort ultracode [on|off] and the slider's Tab toggle were added, and the effortSlider:* keybinding actions rebind them. Automatic model switches after a flag keep the current effort (v2.1.287), and the Agent tool gained an effort parameter so a subagent runs at the level you ask for (v2.1.292).
• Permission modes: Interactive terminal and VS Code sessions with no configured permission mode start in auto mode on every plan and provider. The changelog lists this under v2.1.284, while the Permission modes reference states v2.1.283 or later. claude -p and Python Agent SDK sessions on third-party providers or with telemetry off follow (v2.1.285). Auto mode's prompt before a read outside the working directories gained "Yes, but ask again next time" (v2.1.284), and a conversation too long for the classifier is compacted instead of prompting on every tool call (v2.1.288).
• New commands and syntax: /mcp reconnect all (v2.1.284), --max-findings <n>|all|default on /code-review and /review (v2.1.288), and URL or quickstart-name arguments for /claude-api managed-agents-onboard (v2.1.290). The official table now reflects /claude-api preserved-thinking-migration (v2.1.282+) and /doctor [prompt-audit [path]], and newly lists /artifact-capabilities, /artifact-diagramming (v2.1.221+), /claude-in-chrome [task], /plugin-authoring (v2.1.287+), and /slides [brief] (v2.1.265+). /rate-limit-options appears in /help and the command menu for claude.ai subscribers (v2.1.284).
• Changed command behavior: /hooks opens on one list grouped by event (v2.1.286), and /tasks folds internal work under a "System tasks" row (v2.1.285). /recap declines when relayed from a chat thread, routine, or webhook (v2.1.284). /resume opens a session that is running in the background (v2.1.285). /autocompact saves the window per model (v2.1.288). /memory can no longer turn Auto-memory on from a background session, and /claude-api cannot be run from Remote Control clients (both v2.1.285). /code-review medium reports cleanup and CLAUDE.md-convention findings on Opus 5.5 and Sonnet 5.5, which have no tuned review settings (v2.1.290). /ultrareview output dropped its browser link (v2.1.286).
• CLI: Added claude --desktop, claude plugin configure <plugin>, and claude plugin install --config <server>.<key>=<value> (all v2.1.285), plus claude plugin install --marketplace <source> (v2.1.292). claude project purge became claude purge, and the older name still works (v2.1.288). claude attach and claude logs accept part of a session name in place of the ID (v2.1.290). --bare connects only the MCP servers named on the command line and runs without system reminders or background tasks (v2.1.286).
• MCP and plugins: URL prompts from servers on the 2025-11-25 protocol with bareElicitationCapability, and alwaysLoad: false deferring all of a server's tools (both v2.1.287); default negotiation of protocol revision 2026-07-28 for stdio servers with MCP_PROTOCOL_NEGOTIATION=legacy as the opt-out (v2.1.292). Claude Mods, plugins that modify deeper behavior, and the built-in mod You should know arrived in v2.1.287.
• Settings and environment variables: managed allowedProviders, CLAUDE_CODE_DISABLE_WEB_FETCH, and CLAUDE_CODE_NONSTREAMING_TIMEOUT_RETRIES (all v2.1.285), CLAUDE_CODE_WEB_SEARCH_REFILLS_PER_HOUR (v2.1.290), and CLAUDE_CODE_OVERLOADED_RETRY_BASE_DELAY_MS (v2.1.292). The interactive WebSearch budget refills at 100 calls per hour instead of ending after 200 calls (v2.1.290). Opus 4.7 and later and Fable use a 1M context window by default on Bedrock, Vertex, Foundry, and the Claude apps gateway with no [1m] suffix (v2.1.287; CLAUDE_CODE_DISABLE_1M_CONTEXT=1 keeps 200K).
• Background command time limit: v2.1.285 gave background Bash and PowerShell commands a time limit (default 30 minutes, maximum 2 hours), and v2.1.288 restricted it to unattended sessions such as -p, the Agent SDK, CI, and cloud.

v2.1.280–v2.1.283 coverage — September 26, 2026

Reviewed the official changelog from v2.1.280 (September 22) through v2.1.283 (September 25); no v2.1.279 was published. The official slash, CLI, and flag tables kept their names and order; only descriptions changed.

• Opus 5.5 (v2.1.280): claude-opus-5-5 is the default Opus (1M context, $4/$20). Pro and Team Standard now default to Opus instead of Sonnet. Default effort is medium, and a top-level effortLevel saved before per-model levels no longer applies to new models such as Opus 5.5. The launch-default effort hold on Opus 4.7, Opus 4.8, and Fable 5 no longer overrides /effort in -p/SDK or explicit settings. Fast mode defaults to Opus 5.5 ($8/$40).
• Model restrictions (v2.1.283): managed deniedModels and availableModelsMatch: "exact"; the /model picker drops "(1M context)" from the Opus row.
• Commands: /doctor prompt-audit (also /checkup prompt-audit, v2.1.283). /insights estimates how many permission prompts auto mode could have handled (v2.1.281); /deep-research is more reliable on long briefs (v2.1.281). Pressing x on a running /ultrareview in /tasks asks for confirmation (v2.1.281), and the /ultrareview launch dialog warns that reviewing a local branch may upload uncommitted changes to tracked files (v2.1.283). The leftover "(removed)" /agents entry is gone from the menu and /help (v2.1.281).
• Input and keybindings: send now (ctrl+enter or ctrl+x ctrl+s) moves running tools to the background instead of cancelling the turn (v2.1.281). A stray y or n no longer confirms or closes dialogs; bind confirm:yes/confirm:no in keybindings.json to restore it (v2.1.280). /rewind and /diff lists use the shared select:* actions; messageSelector:*/diff:* rebinds still work (v2.1.283).
• Namespaces and permission mode: v2.1.282 stopped loading skills, commands, and workflows named anthropic-skills or claude-ai; v2.1.283 reverted the claude-ai reservation. Interactive sessions on third-party providers or with telemetry off start in auto mode when no permission mode is configured (v2.1.283; permissions.defaultMode still wins).
• CLI and settings: --system-prompt/--append-system-prompt accept the text and -file forms together (v2.1.283); --agents accepts a JSON file path with -p (v2.1.281); new maxProseWidth (v2.1.282) and "attribution": false (v2.1.281) settings.

Sources: Changelog, Model configuration, Fast mode, Opus 5.5 overview.

Official command audit — September 22, 2026

Reconciled 114 Commands entries, 34 CLI invocation forms, and 79 common CLI flags. The release baseline is v2.1.278 (September 19, 2026) in the official changelog. Older npm channel values below describe their recorded dates, not the current channels.

• Restored /output-style [style] as a current documented command, with a reference row and usage page.

• Added /claude-api build-eval and hillclimb (v2.1.259+), and /import cursor (v2.1.265+).

• Added plugin install/update --accept-command <sha256> (v2.1.271), /plugin install <plugin> --marketplace <source> (v2.1.275), and runner --drain-marker-file <path> (v2.1.271).

• Updated /config Project instructions and AGENTS.md fallback (v2.1.277), /status Auto mode server (v2.1.278), and /fast in Remote sessions (v2.1.271).

• Added /update-config [request], corrected obsolete /agents and /review guidance, the /design minimum (v2.1.265), and CLI stream/budget details.

All slash commands · Terminal CLI reference

Sources: Commands, CLI reference, Changelog.

The sections below retain earlier update records.

September 12, 2026 release refresh

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.

Updated: both locales of the index, setup, verification, and relevant command/operations chapters. Entries below retain their historical verification dates.

2026-09-05

• Raised the baseline to Claude Code v2.1.261 (latest/next) and v2.1.236 (stable) after reviewing v2.1.252~261, the Model configuration and Fast mode docs, and the Claude Fable 5.1 overview, what's-new, migration guide, and announcement.
• Documented Claude Fable 5.1 (claude-fable-5-1, released 2026-09-01) as the default Fable model from v2.1.257 with a v2.1.255 minimum: alias resolution, /model claude-fable-5 for Fable 5, gateway sessions staying on Fable 5, availableModels prefix behavior, $0.25/MTok cache reads, retention, and retirement dates.
• Added the category-based classifier fallback table (biology to Opus 5, cybersecurity to Opus 4.8), switchModelsOnFlag, and the third-party ANTHROPIC_DEFAULT_OPUS_MODEL requirement.
• Added Fable 5.1 Messages API breaking changes (forced tool_choice, model-bound thinking blocks, history-editing invalidation) and additive features (per-message effort, turn-scoped system messages, display: "updates", content provenance), plus the Mythos 5.1 access distinction.
• Updated /effort for the v2.1.257 session-only s key, session-scoped --effort hold release, the absence of a default-effort hold on Opus 5 and Fable 5.1, and v2.1.260 cache-preserving effort changes on Fable 5.1.
• Updated /fast to state that Fable 5.1 is unsupported and to record the Opus 4.7 removal date, model-switch behavior, and -p constraints.
• Clarified /diff: it is a pre-existing built-in already covered by the 110-row reference, and v2.1.260 adds the fullscreen diff panel toggle, so only the descriptions changed.
• Left /skill-doctor (v2.1.261) out because it appears in the changelog but not yet in the official Commands table; other non-model v2.1.252~261 additions are queued for the next full Commands-table reconciliation.

2026-08-30

• Added complete Korean and English reference pages covering all 110 rows in the official Commands table, plus an automated omission check.
• Updated npm channels to latest/next=2.1.251 and stable=2.1.236 after reviewing v2.1.248~251.
• Added /workflow-authoring, model-switch hooks, /usage spend-limit, /cost prompt-cache metrics, and expanded statusline input.
• Added restricted mode, top-level background-session subcommands, symlink/path-traversal hardening, and current agent, Remote Control, and MCP reliability notes.

2026-08-28

• Updated npm channels to latest/next=2.1.247 and stable=2.1.231 after reviewing v2.1.239~247.
• Added per-loop /usage, Auto-mode rules in /permissions, the current Python 0.x→1.x /claude-api upgrade, release-tracked cost-optimize, Admin API coverage, and exact feedback-draft values/scopes.
• Added immediate destination configuration for /cd, partial subagent continuation, and headless auto-recovery guidance.
• Added managed model/cache controls, the Sonnet 5 compaction threshold, and wildcard, marketplace, and gateway credential hardening.

2026-08-20

• Updated npm channels to latest=2.1.237, next=2.1.237, and stable=2.1.228, after reviewing every official changelog entry from v2.1.221 through v2.1.237.
• Documented interactive fork mode and background subagent spawning, cross-session messaging and @ mentions, and removal of the old 200-per-session subagent total.
• Added ANTHROPIC_DEFAULT_MODEL, the Concise output style, claude remote-control --continue, self-hosted runners, GitLab merge-request workflows, and plugin command/archive sources.
• Added the newer-model Task/Todo default change and compatibility flag, plus the worktree, sandbox, MCP OAuth, trust, and Remote Control hardening from this release window.

2026-08-03

• Re-checked the official changelog, local CLI, and npm; v2.1.220 remains latest and next, and v2.1.212 remains stable.
• Found no command addition, removal, or rename after the August 1 update.
• Kept the current /branch, /fork, /subtask, /verify, and /code-review guidance and synchronized the verification date across both Claude handbooks.

2026-08-01

• Updated the release baseline to Claude Code v2.1.220 (latest/next) and v2.1.212 (stable) after checking npm, the local CLI, and the official v2.1.206–220 changelog.
• Added Opus 5 as the Anthropic API opus path with native 1M context, the current provider matrix, and the v2.1.219 minimum CLI requirement.
• Updated /fast to Opus 5 and Opus 4.8, noting the removal of Opus 4.7 support.
• Split /branch, /fork, and /subtask according to the v2.1.212 lifecycle change.
• Marked /verify and /code-review user-invoked only and documented /code-review background-subagent execution.
• Added current subagent limits, forked-skill defaults, sandbox isolation controls, hook, and headless MCP-error guidance.

2026-07-19

• Updated Fable 5 subscription access after the 2026-07-07 temporary allowance ended, separating Pro/Max/Team and seat-based Enterprise usage credits from API consumption billing.
• Added HTTP 200 stop_reason: "refusal" handling, nullable stop_details, Message Batches detection, and refusal-specific monitoring guidance.
• Documented server-side fallbacks, SDK refusal-fallback middleware, manual retry, and the distinction from Claude Code availability fallback settings.
• Added pre-output refusal billing, automatic versus manual fallback credit, and prompt-cache write-cost protection.
• Added Fable prompting guidance to avoid reasoning_extraction refusals caused by internal-reasoning extraction instructions.

2026-07-10

• Updated the English command baseline to Claude Code v2.1.205 after checking the official changelog and npm dist-tags (latest=2.1.205, stable=2.1.197; excluded next=2.1.206 because it has no public changelog entry yet).
• Added the advisory Dynamic workflow size setting and workflow.run_id/workflow.name OTel attributes from v2.1.202.
• Clarified that /review is the fast single-pass review and /code-review <level> <PR> is the multi-agent path.
• Updated /doctor as a full setup checkup with /checkup alias, and added v2.1.203–205 Agent View/background/MCP roots, headless hook streaming, Auto-mode transcript protection, unresolved rm -rf prompting, and --json-schema validation changes.

2026-07-04

• Updated the English command baseline to Claude Code v2.1.201 after checking the official changelog and npm dist-tags (latest=2.1.201, stable=2.1.185).
• Fable 5 redeployment: Fable 5 and Mythos 5 were released GA on 2026-06-09, suspended for all users on 2026-06-12 under US export controls (~19 days), then Fable 5 was restored globally across the Claude Platform, Claude.ai, Claude Code, and Cowork on 2026-07-01 with an improved safety classifier targeting the reported vulnerability-finding behavior. Mythos 5 (no safety classifiers) was redeployed only to restricted US critical-infrastructure organizations via Project Glasswing and is not selectable through /model.
• Added the Fable 5 plan-usage note to the /model guidance: for Pro/Max/Team/select Enterprise, Fable 5 is included up to 50% of weekly usage limits through 2026-07-07, after which it is billed via usage credits.
• Added the new /dataviz skill (v2.1.198) to the command catalog, and reflected v2.1.198 background-by-default subagents, v2.1.199 SSL/streaming/subagent fixes, v2.1.200 default permission mode "Manual", and v2.1.201 removal of the mid-conversation system role on Sonnet 5 sessions.

2026-07-01

• Updated the English command baseline to Claude Code v2.1.197 after checking the official changelog and npm dist-tags (latest=2.1.197, stable=2.1.185).
• Reflected Sonnet 5 as the current Anthropic API sonnet alias and added the v2.1.196 organization-default model note for /model governance.
• Updated verification and command summaries so the English edition matches the Korean handbook baseline.

2026-06-25

• Updated the English command baseline to Claude Code v2.1.190 after checking the official changelog and npm dist-tags (latest=2.1.190, stable=2.1.179).
• Added model-operation notes for model restrictions, Haiku main-model limits with subagent allowance, and stronger deprecated-model warnings.
• Added MCP-operation notes for claude mcp login/logout, CLAUDE_CODE_MCP_TOOL_IDLE_TIMEOUT, and misleading /mcp status fixes.
• Added sandbox and permission notes for sandbox.credentials, destructive-command sandbox auto-approval, Windows PowerShell sandbox fixes, Agent deny rules, and background permission prompts.
• Refreshed background/subagent guidance for the three-level nesting cap, nested-agent stability, and background-agent worktree/comment/commit edge cases.
• Added operating notes for Korean paste handling, non-interactive fallback reliability, output-schema determinism, and app/schema bug fixes.

2026-06-10

• Updated the English command baseline to Claude Code v2.1.170 and Claude Fable 5.
• Added current model guidance for /model fable, best, Fable fallback to Opus, and fallbackModel availability chains.
• Added /advisor Fable option and detailed /cd cwd-switch guidance from the recent changelog.
• Added claude --safe-mode troubleshooting guidance, including how it differs from --bare, plus bundled skill/workflow disable controls.
• Added the v2.1.170 transcript persistence fix and v2.1.169 self-hosted runner post-session hook to the operating notes.
• Kept Opus 4.8 as the default high-capability Opus path and left fast mode scoped to the documented Opus behavior.

2026-05-29

• Updated the English command baseline to Claude Code v2.1.154 and Claude Opus 4.8.
• Added current model guidance for /model, provider-specific aliases, Opus 4.8 high effort, and Opus 4.8 fast mode.
• Added current command coverage for /workflows, /goal, /background//bg, /usage-credits, /reload-skills, /run, /verify, /run-skill-generator, /code-review, and claude --bg --exec.
• Clarified that /simplify is now cleanup-only and correctness review belongs under /code-review.
• Updated Agent View/background-session guidance in automation and operations pages.

2026-05-06

• Added the English locale for Claude Code Command Master.
• Registered claude-code-command-master under the English content tree.
• Updated the Korean source to the v2.1.131-era command baseline at the same time.
• Covered recent command changes: /skills, /theme, /usage consolidation, /fast, /schedule, /autofix-pr, /web-setup, /desktop, and claude project purge.
• Documented the editorial decision to group English command pages by workflow instead of translating every Korean command detail page one-to-one.