diff --git a/.claude/skills/lesson-quiz/SKILL.md b/.claude/skills/lesson-quiz/SKILL.md index 853bd0f..3a8a535 100644 --- a/.claude/skills/lesson-quiz/SKILL.md +++ b/.claude/skills/lesson-quiz/SKILL.md @@ -3,7 +3,7 @@ name: lesson-quiz description: "Test a learner on a single Claude Code tutorial lesson (01-10) with 10 questions, scoring answers and flagging weak spots. Use before, during, or after a lesson. Don't use for whole-tutorial assessment or explaining a topic instead of testing it." effort: high metadata: - version: 1.3.0 + version: 1.4.0 author: Luong NGUYEN --- diff --git a/.claude/skills/lesson-quiz/references/question-bank.md b/.claude/skills/lesson-quiz/references/question-bank.md index ea65abd..ae0c7cd 100644 --- a/.claude/skills/lesson-quiz/references/question-bank.md +++ b/.claude/skills/lesson-quiz/references/question-bank.md @@ -72,10 +72,10 @@ ### Q9 - **Category**: practical -- **Question**: You want to restrict which tools a skill can use. Which frontmatter field do you add? +- **Question**: You want a skill to use certain tools without asking for permission each time. Which frontmatter field do you add? - **Options**: A) `tools: [Read, Grep]` | B) `allowed-tools: [Read, Grep]` | C) `permissions: [Read, Grep]` | D) `restrict-tools: [Read, Grep]` - **Correct**: B -- **Explanation**: The `allowed-tools` field in SKILL.md frontmatter scopes which tools the command can invoke. +- **Explanation**: `allowed-tools` pre-approves the listed tools during the turn that invokes the skill, so Claude can use them without a permission prompt; the grant clears when you send your next message. It does not restrict which tools are available — to remove tools while a skill is active, use `disallowed-tools`. - **Review**: Frontmatter Reference ### Q10 @@ -152,7 +152,7 @@ - **Options**: A) Delete the ~/.claude/projects directory | B) Set `CLAUDE_CODE_DISABLE_AUTO_MEMORY=1` | C) Add `auto-memory: false` to CLAUDE.md | D) Use `/memory disable auto` - **Correct**: B - **Explanation**: Setting `CLAUDE_CODE_DISABLE_AUTO_MEMORY=1` disables auto memory. Value `0` forces it on. Unset = default on. -- **Review**: Auto Memory configuration section +- **Review**: Turning Auto Memory On or Off ### Q9 - **Category**: conceptual @@ -235,7 +235,7 @@ - **Question**: What characters are allowed in the `name` field of a skill's frontmatter? - **Options**: A) Any characters | B) Lowercase letters, numbers, and hyphens only (max 64 chars) | C) Letters and underscores | D) Alphanumeric only - **Correct**: B -- **Explanation**: The name must be kebab-case (lowercase, hyphens), max 64 characters, and cannot contain "anthropic" or "claude". `name` is itself optional — if omitted it defaults to the skill's directory name — but when supplied it must follow these rules. +- **Explanation**: The name must be kebab-case (lowercase, hyphens), max 64 characters. `name` is itself optional — if omitted it defaults to the skill's directory name — but when supplied it must follow these rules. In Claude Code, the `anthropic-skills` name and namespace are reserved for skills synced from claude.ai, so a skill with that name doesn't load; the broader ban on "anthropic" or "claude" applies to claude.ai skill uploads and the Skills API. - **Review**: SKILL.md format section ### Q9 @@ -327,15 +327,15 @@ - **Question**: What are the valid `permissionMode` values for a subagent? - **Options**: A) read, write, admin | B) manual, acceptEdits, bypassPermissions, plan, dontAsk, auto | C) safe, normal, dangerous | D) restricted, standard, elevated - **Correct**: B -- **Explanation**: Subagents support six permission modes: manual (prompts for everything — renamed from `default` in v2.1.200, which is still accepted as an alias), acceptEdits (auto-accepts file edits), bypassPermissions (skips all), plan (read-only), dontAsk (auto-denies unless pre-approved), auto (background classifier decides). +- **Explanation**: Subagents support six permission modes: manual (prompts for everything — shown as Manual; its config value is `default`, and manual is accepted as an alias since v2.1.200), acceptEdits (auto-accepts file edits), bypassPermissions (skips all), plan (read-only), dontAsk (auto-denies unless pre-approved), auto (background classifier decides). - **Review**: Configuration fields section ### Q10 - **Category**: practical - **Question**: How do you resume a subagent that returned an agentId from a previous run? -- **Options**: A) Use `/resume agent-id` | B) Pass the `resume` parameter with the agentId when calling Task tool | C) Use `claude -r agent-id` | D) Subagents cannot be resumed +- **Options**: A) Use `/resume agent-id` | B) Ask Claude to resume it — Claude messages the subagent with the SendMessage tool, using its agent ID or name | C) Use `claude -r agent-id` | D) Subagents cannot be resumed - **Correct**: B -- **Explanation**: Subagents can be resumed by passing the `resume` parameter with the previously returned agentId, continuing with full context preserved. +- **Explanation**: Each invocation starts a new instance. To continue earlier work, ask Claude to resume the subagent: Claude uses the SendMessage tool with the agent's ID or name, and the subagent resumes with its full conversation history. The built-in Explore and Plan agents are one-shot and can't be resumed. - **Review**: Resumable agents section --- @@ -360,10 +360,10 @@ ### Q3 - **Category**: conceptual -- **Question**: What happens when MCP tool descriptions exceed 10% of the context window? -- **Options**: A) They are truncated | B) Tool Search auto-enables to dynamically select relevant tools | C) Claude shows an error | D) Extra tools are disabled +- **Question**: With `ENABLE_TOOL_SEARCH=auto`, what happens when MCP tool definitions reach 10% of the context window? +- **Options**: A) They are truncated | B) Tool search kicks in: MCP tools are deferred and discovered on demand | C) Claude shows an error | D) Extra tools are disabled - **Correct**: B -- **Explanation**: MCP Tool Search auto-enables when tools exceed 10% of context. It requires Sonnet 4 or Opus 4 minimum (Haiku not supported). +- **Explanation**: By default (`ENABLE_TOOL_SEARCH` unset) tool search is already on and all MCP tools are deferred. `auto` is threshold mode: tools load upfront until their definitions reach 10% of the context window, then all are deferred (`auto:N` sets a custom percentage). Tool search needs a model that supports `tool_reference` blocks: Sonnet 4.5, Haiku 4.5, Opus 4.5 and later. - **Review**: MCP Tool Search section ### Q4 @@ -471,7 +471,7 @@ - **Question**: You want a hook that only runs once when a skill is first loaded, not on every tool call. What field do you add? - **Options**: A) `run-once: true` | B) `once: true` in the component hook definition | C) `single: true` | D) `max-runs: 1` - **Correct**: B -- **Explanation**: Component-scoped hooks (defined in SKILL.md or agent frontmatter) support `once: true` to run only on first activation. +- **Explanation**: Hooks declared in a skill's frontmatter support `once: true` — Claude Code removes the hook after its first successful run (a run that fails, blocks, or times out leaves it in place). `once` is honored only in skill frontmatter. - **Review**: Component-scoped hooks section ### Q7 @@ -683,7 +683,7 @@ - **Question**: What are the six permission modes in Claude Code? - **Options**: A) read, write, execute, admin, root, sudo | B) manual, acceptEdits, plan, auto, dontAsk, bypassPermissions | C) safe, normal, elevated, admin, unrestricted, god | D) view, edit, run, deploy, full, bypass - **Correct**: B -- **Explanation**: The six modes are: manual (prompts for everything — renamed from `default` in v2.1.200, which is still accepted as an alias), acceptEdits (auto-accepts file edits), plan (read-only analysis), auto (background classifier decides), dontAsk (auto-denies unless pre-approved), bypassPermissions (skips all checks). +- **Explanation**: The six modes are: manual (prompts for everything — shown as Manual; its config value is `default`, and manual is accepted as an alias since v2.1.200), acceptEdits (auto-accepts file edits), plan (read-only analysis), auto (background classifier decides), dontAsk (auto-denies unless pre-approved), bypassPermissions (skips all checks). - **Review**: Permission Modes section ### Q2 @@ -704,10 +704,10 @@ ### Q4 - **Category**: practical -- **Question**: How do you toggle extended thinking on or off during a session? +- **Question**: On a model where thinking can be turned off (for example Sonnet 5), how do you toggle extended thinking on or off during a session? - **Options**: A) Type `/effort max` | B) Press `Option+T` (macOS) or `Alt+T` | C) Include "ultrathink" in prompt | D) It's always enabled and cannot be toggled - **Correct**: B -- **Explanation**: Option+T (macOS) or Alt+T toggles extended thinking on/off for the session. (`Ctrl+O` toggles verbose mode to show/hide the reasoning text.) For one-off deep reasoning, include "ultrathink" in your prompt; for session-level control, use `/effort` command. +- **Explanation**: Option+T (macOS) or Alt+T toggles extended thinking on/off for the session. (`Ctrl+O` toggles verbose mode to show/hide the reasoning text.) On Opus 5.5 (the default model since v2.1.280) and the Fable models, thinking can't be turned off: the toggle, `alwaysThinkingEnabled` and `MAX_THINKING_TOKENS=0` have no effect there. For one-off deep reasoning, include "ultrathink" in your prompt; for session-level control, use the `/effort` command. - **Review**: Extended Thinking section ### Q5 @@ -798,9 +798,9 @@ - **Category**: conceptual - **Question**: Which flag only works in print mode (-p) and has no effect in interactive mode? - **Options**: A) `--model` | B) `--system-prompt-file` | C) `--verbose` | D) `--max-turns` -- **Correct**: B -- **Explanation**: `--system-prompt-file` loads a system prompt from a file but only works in print mode. Use `--system-prompt` (inline string) for interactive sessions. -- **Review**: System prompt flags comparison table +- **Correct**: D +- **Explanation**: `--max-turns` limits the number of agentic turns and applies only in print mode (`-p`); it exits with an error when the limit is reached. `--system-prompt-file`, like the other system-prompt flags, works in both interactive and print modes. +- **Review**: Advanced Features ### Q6 - **Category**: practical @@ -844,8 +844,8 @@ --- -**Last Updated**: September 2, 2026 -**Claude Code Version**: 2.1.257 +**Last Updated**: September 26, 2026 +**Claude Code Version**: 2.1.283 **Sources**: - https://code.claude.com/docs/en/hooks - https://code.claude.com/docs/en/skills @@ -856,3 +856,4 @@ - https://code.claude.com/docs/en/checkpointing - https://code.claude.com/docs/en/plugins-reference - https://code.claude.com/docs/en/cli-reference +- https://code.claude.com/docs/en/model-config#extended-thinking diff --git a/.claude/skills/self-assessment/README.md b/.claude/skills/self-assessment/README.md index 00d6a35..92e6b73 100644 --- a/.claude/skills/self-assessment/README.md +++ b/.claude/skills/self-assessment/README.md @@ -42,7 +42,7 @@ graph TD ### Deep Assessment (~5 min) - 5 rounds of questions covering 10 feature areas (2 topics per round) -- Per-topic scoring (0-2 points each, 19 points maximum — Checkpoints contributes 0-1) +- Per-topic scoring (0-2 points each, 20 points maximum — Checkpoints contributes 0-1, Advanced Features 0-3) - Mastery table with strength areas, priority gaps, and review items - Dependency-aware learning path with phases and time estimates - Recommended practice projects combining gap topics @@ -50,7 +50,7 @@ graph TD ## Usage -``` +```text /self-assessment ``` diff --git a/.claude/skills/self-assessment/SKILL.md b/.claude/skills/self-assessment/SKILL.md index 3e9c651..679cda6 100644 --- a/.claude/skills/self-assessment/SKILL.md +++ b/.claude/skills/self-assessment/SKILL.md @@ -1,6 +1,6 @@ --- name: self-assessment -version: 2.5.0 +version: 2.5.1 description: Comprehensive Claude Code self-assessment and learning path advisor. Runs a multi-category quiz covering 10 feature areas, produces a detailed skill profile with per-topic scores, identifies specific gaps, and generates a personalized learning path with prioritized next steps. Use when asked to "assess my level", "take the quiz", "find my level", "where should I start", "what should I learn next", "check my skills", "skill check", or "level up". --- @@ -292,7 +292,7 @@ Use these specific recommendations when a topic is a gap: **Memory (score 0)**: - Tutorial: [02-memory/](../../../02-memory/) -- Focus on: CLAUDE.md creation, `/init` and `/memory` commands, `#` prefix for quick updates +- Focus on: CLAUDE.md creation, `/init` and `/memory` commands, asking Claude to remember something (the old `#` shortcut was removed) - Key exercise: Create a project CLAUDE.md with your coding standards - Done when: Claude remembers your preferences across sessions @@ -337,7 +337,7 @@ Use these specific recommendations when a topic is a gap: - Done when: Claude delegates code review to your custom agent **Subagents (score 1 — review)**: -- Focus on: Worktree isolation (`isolation: worktree`), persistent agent memory (`memory` field with scopes), background agents (Ctrl+B/Ctrl+F), agent allowlists with `Agent(agent_type)` (`Task(...)` remains a back-compat alias), agent teams (`--teammate-mode`) +- Focus on: Worktree isolation (`isolation: worktree`), persistent agent memory (`memory` field with scopes), background agents (Ctrl+B to background; Ctrl+X Ctrl+K twice to stop them all), agent allowlists with `Agent(agent_type)` (`Task(...)` remains a back-compat alias), agent teams (`--teammate-mode`) - Done when: You have a subagent with persistent memory running in worktree isolation **Checkpoints (score 0)**: @@ -348,7 +348,7 @@ Use these specific recommendations when a topic is a gap: **Advanced Features (score 0)**: - Tutorial: [09-advanced-features/](../../../09-advanced-features/) -- Focus on: Planning mode (/plan or Shift+Tab), permission modes (6 types: manual — renamed from default in v2.1.200 — acceptEdits, plan, auto, dontAsk, bypassPermissions), extended thinking (Alt+T toggle) +- Focus on: Planning mode (/plan or Shift+Tab), permission modes (6 types: manual — renamed from default in v2.1.200 — acceptEdits, plan, auto, dontAsk, bypassPermissions), extended thinking (Alt+T/Option+T toggle — no effect on Opus 5.5, the default model, where thinking stays on) - Key exercise: Use planning mode to design a feature, then implement it - Done when: You can switch between planning and implementation modes fluently diff --git a/.claude/skills/self-assessment/references/topic-recommendations.md b/.claude/skills/self-assessment/references/topic-recommendations.md index 1855e8b..31af852 100644 --- a/.claude/skills/self-assessment/references/topic-recommendations.md +++ b/.claude/skills/self-assessment/references/topic-recommendations.md @@ -14,7 +14,7 @@ Use these specific recommendations when a topic is a gap. Paths are relative to **Memory (score 0)**: - Tutorial: [02-memory/](../../../../02-memory/) -- Focus on: CLAUDE.md creation, `/init` and `/memory` commands, `#` prefix for quick updates +- Focus on: CLAUDE.md creation, `/init` and `/memory` commands, asking Claude to remember something (the old `#` shortcut was removed) - Key exercise: Create a project CLAUDE.md with your coding standards - Done when: Claude remembers your preferences across sessions @@ -59,7 +59,7 @@ Use these specific recommendations when a topic is a gap. Paths are relative to - Done when: Claude delegates code review to your custom agent **Subagents (score 1 — review)**: -- Focus on: Worktree isolation (`isolation: worktree`), persistent agent memory (`memory` field with scopes), background agents (Ctrl+B/Ctrl+F), agent allowlists with `Agent(agent_type)` (`Task(...)` remains a back-compat alias), agent teams (`--teammate-mode`) +- Focus on: Worktree isolation (`isolation: worktree`), persistent agent memory (`memory` field with scopes), background agents (Ctrl+B to background; Ctrl+X Ctrl+K twice to stop them all), agent allowlists with `Agent(agent_type)` (`Task(...)` remains a back-compat alias), agent teams (`--teammate-mode`) - Done when: You have a subagent with persistent memory running in worktree isolation **Checkpoints (score 0)**: @@ -70,7 +70,7 @@ Use these specific recommendations when a topic is a gap. Paths are relative to **Advanced Features (score 0)**: - Tutorial: [09-advanced-features/](../../../../09-advanced-features/) -- Focus on: Planning mode (/plan or Shift+Tab), permission modes (6 types: manual — renamed from default in v2.1.200 — acceptEdits, plan, auto, dontAsk, bypassPermissions), extended thinking (Alt+T toggle) +- Focus on: Planning mode (/plan or Shift+Tab), permission modes (6 types: manual — renamed from default in v2.1.200 — acceptEdits, plan, auto, dontAsk, bypassPermissions), extended thinking (Alt+T/Option+T toggle — no effect on Opus 5.5, the default model, where thinking stays on) - Key exercise: Use planning mode to design a feature, then implement it - Done when: You can switch between planning and implementation modes fluently @@ -100,8 +100,8 @@ Use these specific recommendations when a topic is a gap. Paths are relative to --- -**Last Updated**: September 2, 2026 -**Claude Code Version**: 2.1.257 +**Last Updated**: September 26, 2026 +**Claude Code Version**: 2.1.283 **Sources**: - https://code.claude.com/docs/en/hooks - https://code.claude.com/docs/en/memory @@ -111,3 +111,5 @@ Use these specific recommendations when a topic is a gap. Paths are relative to - https://code.claude.com/docs/en/checkpointing - https://code.claude.com/docs/en/permission-modes - https://code.claude.com/docs/en/plugins-reference +- https://code.claude.com/docs/en/interactive-mode +- https://code.claude.com/docs/en/model-config#extended-thinking diff --git a/01-slash-commands/README.md b/01-slash-commands/README.md index 7cec7e9..03797ce 100644 --- a/01-slash-commands/README.md +++ b/01-slash-commands/README.md @@ -26,7 +26,7 @@ Built-in commands are shortcuts for common actions. There are **60+ built-in com |---------|---------| | `/add-dir ` | Add working directory | | `/advisor [model\|off]` | Configure the advisor. Opens as an interactive dialog; in the desktop app, Remote Control, and headless (`-p` / Agent SDK) sessions it takes a text form instead — bare `/advisor`, `/advisor `, or `/advisor off` (v2.1.260+) | -| `/agents` | Manage agent configurations | +| `/agents` | No longer a working command: the interactive wizard was removed in v2.1.198, and since v2.1.281 `/agents` no longer appears in the `/` menu or `/help` — typing it only explains where the wizard went. Ask Claude to create or manage subagents, or edit `.claude/agents/` files directly | | `/branch [name]` | Switch into a copy of the conversation at this point, preserving the original (return to it with `/resume`) | | `/fork [prompt]` | Copy the current conversation into a new **background session** and keep working here; the two are independent from that point on and the copy gets its own row in `claude agents` (v2.1.212+). Except when the copy edits in place, Claude Code instructs it to create a worktree of its own before making code changes (isolation instruction requires v2.1.221+) | | `/subtask ` | Spawn a **forked subagent** that inherits the full conversation and works on the task while you keep going; its result returns to this conversation when it finishes (v2.1.212+) | @@ -42,12 +42,12 @@ Built-in commands are shortcuts for common actions. There are **60+ built-in com | `/cost` | Typing-shortcut alias for `/usage` — opens the cost tab (v2.1.118+) | | `/desktop` | Continue in Desktop app (alias: `/app`) | | `/diff` | Interactive diff viewer for uncommitted changes. In fullscreen rendering it instead opens a diff panel beside the conversation that stays open while you keep working — it lists changed files with added/removed line counts and refreshes every time Claude edits a file or runs a shell command; run `/diff` again or click `✕` to close it (v2.1.260+). The classic renderer opens the viewer in place of the prompt | -| `/doctor` | Diagnose installation health — openable while Claude is responding; shows status icons; press `f` to auto-fix issues (enhanced in v2.1.116; layout refreshed to a flat tree with clearer icons in v2.1.178) | -| `/effort [low\|medium\|high\|xhigh\|max\|auto]` | Set effort level via interactive arrow-key slider. Levels: `low` → `medium` → `high` → `xhigh` (new in v2.1.111) → `max`. Default is `high` on Opus 5, Sonnet 5, and Opus 4.8 (`xhigh` on Opus 4.7); `xhigh` needs Opus 5, Sonnet 5, Opus 4.8, or Opus 4.7; `max` works on Opus 5, Sonnet 5, Opus 4.8/4.7/4.6 and Sonnet 4.6. The menu also offers `ultracode` (not a model effort level — it sends `xhigh` and has Claude orchestrate dynamic workflows; session-only) | +| `/doctor` | Diagnose installation health — openable while Claude is responding; shows status icons; press `f` to auto-fix issues (enhanced in v2.1.116; layout refreshed to a flat tree with clearer icons in v2.1.178); `/doctor prompt-audit` (also `/checkup prompt-audit`) audits CLAUDE.md files, skills, agents, and commands for prompting patterns written for older models (v2.1.283) | +| `/effort [low\|medium\|high\|xhigh\|max\|auto]` | Set effort level via interactive arrow-key slider. Levels: `low` → `medium` → `high` → `xhigh` (new in v2.1.111) → `max`. Default is `medium` on Opus 5.5 (the default model since v2.1.280), `high` on Opus 5, Sonnet 5, and Opus 4.8, and `xhigh` on Opus 4.7; `xhigh` needs Opus 5.5, Opus 5, Sonnet 5, Opus 4.8, or Opus 4.7; `max` works on Opus 5.5, Opus 5, Sonnet 5, Opus 4.8/4.7/4.6 and Sonnet 4.6. A level saved before `/effort` became per-model doesn't apply to newly released models such as Opus 5.5 (v2.1.280). The menu also offers `ultracode` (not a model effort level — it sends `xhigh` and has Claude orchestrate dynamic workflows; session-only) | | `/exit` | Exit the REPL (alias: `/quit`) | | `/export [filename]` | Export the current conversation to a file or clipboard | | `/usage-credits` | Configure extra usage for rate limits (renamed from `/extra-usage` in v2.1.144; `/extra-usage` still works as an alias) | -| `/fast [on\|off]` | Toggle fast mode. Applies to Opus 5 and Opus 4.8 (v2.1.219) | +| `/fast [on\|off]` | Toggle fast mode. Applies to Opus 5.5 (the fast-mode default since v2.1.280), Opus 5, and Opus 4.8 | | `/feedback` | Submit feedback (alias: `/bug`). Since v2.1.141, can attach recent sessions (last 24h or 7d) so reports spanning more than one session include context. As of v2.1.178, `/bug` requires a description before it can be submitted. | | `/focus` | Toggle focus view (added v2.1.110; replaces `Ctrl+O` for focus toggle) | | `/goal ` | Register a session-level completion condition; Claude keeps working until the goal is met. `/goal clear` removes it. Active goal appears in the status line, with a live overlay panel showing elapsed time, turn count, and token usage (added v2.1.139). | @@ -65,7 +65,7 @@ Built-in commands are shortcuts for common actions. There are **60+ built-in com | `/mcp` | Manage MCP servers and OAuth | | `/memory` | Edit `CLAUDE.md`, toggle auto-memory | | `/mobile` | QR code for mobile app (aliases: `/ios`, `/android`) | -| `/model [model]` | Select model with left/right arrows for effort. Since v2.1.153, the choice is **saved as the default** for new sessions (matching the IDE); press `s` after selecting to apply it to the current session only. (The keybinding `modelPicker:setAsDefault` was renamed to `modelPicker:thisSessionOnly`; the old `d` action is now `s`.) As of v2.1.219, the picker shows the merged Opus row as "Opus (1M context)". | +| `/model [model]` | Select model with left/right arrows for effort. Since v2.1.153, the choice is **saved as the default** for new sessions (matching the IDE); press `s` after selecting to apply it to the current session only. (The keybinding `modelPicker:setAsDefault` was renamed to `modelPicker:thisSessionOnly`; the old `d` action is now `s`.) From v2.1.219 the picker showed the merged Opus row as "Opus (1M context)"; v2.1.283 drops that label from the Opus and Default rows where Opus already has a 1M window (the window is unchanged). | | `/output-style [name]` | List and switch output styles (re-added in v2.1.269 after being removed in v2.1.91). With no argument it lists the styles and marks the current one. Works in headless and Remote Control sessions; the selection is saved to `.claude/settings.local.json` | | `/passes` | Share free week of Claude Code | | `/permissions` | View/update permissions (alias: `/allowed-tools`) | @@ -650,8 +650,8 @@ If both exist with the same name, the **skill takes precedence**. Remove one or --- -**Last Updated**: September 19, 2026 -**Claude Code Version**: 2.1.278 +**Last Updated**: September 26, 2026 +**Claude Code Version**: 2.1.283 **Sources**: - https://code.claude.com/docs/en/output-styles - https://code.claude.com/docs/en/skills @@ -668,6 +668,9 @@ If both exist with the same name, the **skill takes precedence**. Remove one or - https://github.com/anthropics/claude-code/releases/tag/v2.1.152 - https://github.com/anthropics/claude-code/releases/tag/v2.1.153 - https://github.com/anthropics/claude-code/releases/tag/v2.1.154 +- https://github.com/anthropics/claude-code/releases/tag/v2.1.281 +- https://code.claude.com/docs/en/fast-mode +- https://github.com/anthropics/claude-code/releases/tag/v2.1.283 **Compatible Models**: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 *Part of the [Claude How To](../) guide series* diff --git a/02-memory/README.md b/02-memory/README.md index 74d16c3..040e1c8 100644 --- a/02-memory/README.md +++ b/02-memory/README.md @@ -298,7 +298,7 @@ Set it through `/config`, or in settings: This key is honored in user and managed settings only — setting it in project or local settings has no effect. -**Where direct reading is unavailable.** On versions before v2.1.277, on Bedrock/Vertex/Foundry, with telemetry disabled, during the first session after upgrading, under `disableAllHooks` or `allowManagedHooksOnly`, or with the built-in `agents-md` plugin disabled, Claude Code will not pick up `AGENTS.md` on its own. In those cases, import it from CLAUDE.md: +**Where direct reading is unavailable.** On versions before v2.1.277 (or before v2.1.281 on Bedrock/Vertex/Foundry, behind LLM gateways, or with telemetry disabled), during the first session after upgrading, or with the built-in `agents-md` plugin disabled, Claude Code will not pick up `AGENTS.md` on its own. In those cases, import it from CLAUDE.md: ```markdown @AGENTS.md @@ -1244,9 +1244,10 @@ Auto Memory is a separate mechanism (`~/.claude/projects//memory/`), no --- -**Last Updated**: September 19, 2026 -**Claude Code Version**: 2.1.278 +**Last Updated**: September 26, 2026 +**Claude Code Version**: 2.1.283 **Sources**: - https://code.claude.com/docs/en/memory - https://code.claude.com/docs/en/memory#agents-md +- https://code.claude.com/docs/en/memory#when-agents-md-support-is-unavailable **Compatible Models**: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 diff --git a/03-skills/README.md b/03-skills/README.md index e536c7d..635a112 100644 --- a/03-skills/README.md +++ b/03-skills/README.md @@ -141,7 +141,7 @@ Show concrete examples of using this Skill. ### Recommended Fields - **description** (recommended): what the Skill does AND when to use it. If omitted, Claude Code uses the first paragraph of markdown content. The combined `description` + `when_to_use` text is truncated at **1,536 characters** in the skill listing (configurable via `skillListingMaxDescChars`). This is what Claude matches on to decide when to activate the skill. -- **name** (optional): defaults to the skill's **directory name**. When supplied, it sets the display name — lowercase letters, numbers, hyphens only (max 64 characters), and cannot contain "anthropic" or "claude". For plugin skills, `name` also sets the last segment of the command. +- **name** (optional): defaults to the skill's **directory name**. When supplied, it sets the display name — lowercase letters, numbers, hyphens only (max 64 characters). In Claude Code, `anthropic-skills` and `synced` are reserved and don't load outside a plugin; the stricter ban on "anthropic" or "claude" applies to skills uploaded to claude.ai or the Skills API. For plugin skills, `name` also sets the last segment of the command. All SKILL.md frontmatter fields are optional; `description` is the only one that is recommended. @@ -174,7 +174,7 @@ paths: "src/api/**/*.ts" # Glob patterns limiting when skill activ | Field | Description | |-------|-------------| -| `name` | Lowercase letters, numbers, hyphens only (max 64 chars). Cannot contain "anthropic" or "claude". | +| `name` | Lowercase letters, numbers, hyphens only (max 64 chars). `anthropic-skills` and `synced` are reserved in Claude Code. | | `description` | What the Skill does AND when to use it. The combined `description` + `when_to_use` text is truncated at 1,536 chars in the skill listing (configurable via `skillListingMaxDescChars`). Critical for auto-invocation matching. | | `when_to_use` | Additional context for when Claude should invoke the skill. Appended to `description` in the skill listing and counts toward the 1,536-character cap. | | `argument-hint` | Hint shown in the `/` autocomplete menu (e.g., `"[filename] [format]"`). | @@ -183,7 +183,7 @@ paths: "src/api/**/*.ts" # Glob patterns limiting when skill activ | `allowed-tools` | Comma-separated list of tools the skill may use without permission prompts. | | `disallowed-tools` | Comma-separated list of tools to remove while the skill is active (complements `allowed-tools`). Added v2.1.152. | | `model` | Model override while the skill is active (e.g., `opus`, `sonnet`). | -| `effort` | Effort level override while the skill is active: `low`, `medium`, `high`, `xhigh`, or `max` — all five are supported on Opus 5, Sonnet 5, Opus 4.8, and Opus 4.7. The default effort is `high` on every model that supports effort, except Opus 4.7 which defaults to `xhigh`. | +| `effort` | Effort level override while the skill is active: `low`, `medium`, `high`, `xhigh`, or `max` — all five are supported on Opus 5.5, Opus 5, Sonnet 5, Opus 4.8, and Opus 4.7. The default effort is `high` on every model that supports effort, except Opus 5.5, which defaults to `medium`, and Opus 4.7, which defaults to `xhigh`. | | `context` | `fork` to run the skill in a forked subagent context with its own context window. | | `agent` | Subagent type when `context: fork` (e.g., `Explore`, `Plan`, `general-purpose`). | | `background` | Only meaningful with `context: fork`. Defaults to `true` for `context: fork` skills, so they run in the background; set `false` to run them in the foreground. Added v2.1.218. | @@ -897,11 +897,12 @@ Once you start building skills seriously, two things become essential: a library --- -**Last Updated**: September 6, 2026 -**Claude Code Version**: 2.1.263 +**Last Updated**: September 26, 2026 +**Claude Code Version**: 2.1.283 **Sources**: - https://code.claude.com/docs/en/skills - https://code.claude.com/docs/en/slash-commands - https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md - https://code.claude.com/docs/en/model-config +- https://code.claude.com/docs/en/model-config#adjust-effort-level **Compatible Models**: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 diff --git a/03-skills/claude-md/SKILL.md b/03-skills/claude-md/SKILL.md index b44c052..caf17bd 100644 --- a/03-skills/claude-md/SKILL.md +++ b/03-skills/claude-md/SKILL.md @@ -193,7 +193,7 @@ Before finalizing, verify: If the user requests AGENTS.md creation/update: -**Since v2.1.277, Claude Code reads `AGENTS.md` directly as project instructions** — but only when the working directory and every directory above it contain no `CLAUDE.md`, `.claude/CLAUDE.md`, or `CLAUDE.local.md`. `~/.claude/CLAUDE.md`, managed CLAUDE.md, and `.claude/rules/` do not count for that check and keep loading alongside. Which files are read is controlled by **Project instructions** in `/config`: `claude-md-or-agents-md` (default), `claude-md-and-agents-md`, `claude-md`, or `managed-only`. Where direct reading is unavailable — Bedrock/Vertex/Foundry, telemetry disabled, the first session after upgrading, or `disableAllHooks`/`allowManagedHooksOnly` — fall back to importing it from CLAUDE.md with `@AGENTS.md`, or symlinking `CLAUDE.md` to it. +**Since v2.1.277, Claude Code reads `AGENTS.md` directly as project instructions** — but only when the working directory and every directory above it contain no `CLAUDE.md`, `.claude/CLAUDE.md`, or `CLAUDE.local.md`. `~/.claude/CLAUDE.md`, managed CLAUDE.md, and `.claude/rules/` do not count for that check and keep loading alongside. Which files are read is controlled by **Project instructions** in `/config`: `claude-md-or-agents-md` (default), `claude-md-and-agents-md`, `claude-md`, or `managed-only`. Where direct reading is unavailable — versions before v2.1.277 (before v2.1.281 on Bedrock/Vertex/Foundry, LLM gateways, or with telemetry disabled), or the first session after upgrading — fall back to importing it from CLAUDE.md with `@AGENTS.md`, or symlinking `CLAUDE.md` to it. AGENTS.md is a cross-tool project-context file — the same *category* of document as CLAUDE.md, not an agent-definition format. It exists so several coding agents can share one set of project conventions: - Build, test, and lint commands @@ -217,9 +217,10 @@ Apply similar principles: --- -**Last Updated**: September 19, 2026 -**Claude Code Version**: 2.1.278 +**Last Updated**: September 26, 2026 +**Claude Code Version**: 2.1.283 **Sources**: - https://code.claude.com/docs/en/skills - https://code.claude.com/docs/en/memory#agents-md +- https://code.claude.com/docs/en/memory#when-agents-md-support-is-unavailable **Compatible Models**: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 diff --git a/04-subagents/README.md b/04-subagents/README.md index 4389a23..09529ea 100644 --- a/04-subagents/README.md +++ b/04-subagents/README.md @@ -547,7 +547,7 @@ description: Performs long-running analysis tasks in the background | Shortcut | Action | |----------|--------| | `Ctrl+B` | Background a currently running subagent task | -| `Ctrl+F` | Kill all background agents (press twice to confirm) | +| `Ctrl+X` `Ctrl+K` | Stop all running background subagents (press twice within 3 seconds to confirm; this replaced `Ctrl+F`) | ### Disabling Background Tasks @@ -1332,8 +1332,8 @@ See the OpenTelemetry section in [Advanced Features → Telemetry](../09-advance --- -**Last Updated**: September 19, 2026 -**Claude Code Version**: 2.1.278 +**Last Updated**: September 26, 2026 +**Claude Code Version**: 2.1.283 **Sources**: - https://code.claude.com/docs/en/sub-agents - https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md @@ -1347,4 +1347,5 @@ See the OpenTelemetry section in [Advanced Features → Telemetry](../09-advance - https://github.com/anthropics/claude-code/releases/tag/v2.1.139 - https://github.com/anthropics/claude-code/releases/tag/v2.1.140 - https://code.claude.com/docs/en/model-config +- https://code.claude.com/docs/en/interactive-mode **Compatible Models**: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 diff --git a/05-mcp/README.md b/05-mcp/README.md index 9f5b967..8f2bed2 100644 --- a/05-mcp/README.md +++ b/05-mcp/README.md @@ -259,16 +259,17 @@ Type `/mcp` inside a session to list connected servers, trigger OAuth flows, and ## MCP Tool Search -When MCP tool descriptions exceed 10% of the context window, Claude Code automatically enables tool search to efficiently select the right tools without overwhelming the model context. +Tool search is on by default: MCP tools are deferred and discovered on demand, so their definitions don't fill the model context. Claude Code turns it off when `ANTHROPIC_BASE_URL` points to a non-first-party host, since most proxies don't forward the blocks it needs. | Setting | Value | Description | |---------|-------|-------------| -| `ENABLE_TOOL_SEARCH` | `auto` (default) | Automatically enables when tool descriptions exceed 10% of context | -| `ENABLE_TOOL_SEARCH` | `auto:` | Automatically enables at a custom threshold of `N` tools | -| `ENABLE_TOOL_SEARCH` | `true` | Always enabled regardless of tool count | +| `ENABLE_TOOL_SEARCH` | (unset — default) | All MCP tools deferred and loaded on demand | +| `ENABLE_TOOL_SEARCH` | `auto` | Threshold mode: tools load upfront until their definitions reach 10% of the context window, then all are deferred | +| `ENABLE_TOOL_SEARCH` | `auto:` | Threshold mode with a custom percentage (0-100) — for example, `auto:5` for 5% | +| `ENABLE_TOOL_SEARCH` | `true` | All MCP tools deferred | | `ENABLE_TOOL_SEARCH` | `false` | Disabled; all tool descriptions sent in full | -> **Note:** Tool search requires Sonnet 4 or later, or Opus 4 or later. Haiku models are not supported for tool search. +> **Note:** Tool search requires a model that supports `tool_reference` blocks: Claude Sonnet 4.5, Claude Haiku 4.5, Claude Opus 4.5, and later models. ### Bypassing Tool Search per Server (v2.1.121+) @@ -305,7 +306,7 @@ MCP servers can request structured input from the user via interactive dialogs ( ## Tool Description and Instruction Cap -As of v2.1.84, Claude Code enforces a **2 KB cap** on tool descriptions and instructions per MCP server. This prevents individual servers from consuming excessive context with overly verbose tool definitions, reducing context bloat and keeping interactions efficient. +As of v2.1.84, Claude Code enforces a **2 KB cap** on tool descriptions and instructions per MCP server. This prevents individual servers from consuming excessive context with overly verbose tool definitions, reducing context bloat and keeping interactions efficient. Since v2.1.280, you can change this 2,048-character cap with the `CLAUDE_CODE_MAX_MCP_DESCRIPTION_LENGTH` environment variable. ## MCP Prompts as Slash Commands @@ -1299,8 +1300,8 @@ export GITHUB_TOKEN="your_token" --- -**Last Updated**: September 6, 2026 -**Claude Code Version**: 2.1.263 +**Last Updated**: September 26, 2026 +**Claude Code Version**: 2.1.283 **Sources**: - https://code.claude.com/docs/en/mcp - https://code.claude.com/docs/en/managed-mcp @@ -1309,4 +1310,6 @@ export GITHUB_TOKEN="your_token" - https://github.com/anthropics/claude-code/releases/tag/v2.1.139 - https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md - https://code.claude.com/docs/en/model-config +- https://code.claude.com/docs/en/mcp#configure-tool-search +- https://code.claude.com/docs/en/env-vars **Compatible Models**: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 diff --git a/06-hooks/README.md b/06-hooks/README.md index 787143f..904e2f3 100644 --- a/06-hooks/README.md +++ b/06-hooks/README.md @@ -227,6 +227,8 @@ Subagent-based verification hooks that spawn a dedicated agent to evaluate condi > **Note**: Agent hooks are experimental and may change. +> **Note**: Since v2.1.280, agent hooks don't run on `PermissionRequest` — their answer could never allow or deny the request. Use a command, http, mcp_tool, or prompt hook there. + ```json { "type": "agent", @@ -1404,7 +1406,7 @@ Plugins can include hooks in their `hooks/hooks.json` file: "hooks": [ { "type": "command", - "command": "${CLAUDE_PLUGIN_ROOT}/scripts/validate.sh" + "command": "\"${CLAUDE_PLUGIN_ROOT}\"/scripts/validate.sh" } ] } @@ -1627,12 +1629,13 @@ Edit `~/.claude/settings.json` or `.claude/settings.json` with the hook configur --- -**Last Updated**: September 19, 2026 -**Claude Code Version**: 2.1.278 +**Last Updated**: September 26, 2026 +**Claude Code Version**: 2.1.283 **Sources**: - https://code.claude.com/docs/en/hooks - https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md - https://code.claude.com/docs/en/sub-agents - https://code.claude.com/docs/en/permissions - https://code.claude.com/docs/en/tools-reference#task-tool-availability +- https://code.claude.com/docs/en/plugins/manifest-reference#quoting-and-path-separators **Compatible Models**: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 diff --git a/07-plugins/README.md b/07-plugins/README.md index ac5a859..7fda0d0 100644 --- a/07-plugins/README.md +++ b/07-plugins/README.md @@ -823,6 +823,8 @@ claude plugin init # Scaffold a new plugin (alias: cla | `plugin validate` | `--strict` | Treat warnings as errors | | `plugin validate` | `--json` | Emit a machine-readable validation report (v2.1.259+) | +`claude plugin validate` also checks `.mcp.json` entries (v2.1.281+) and fails when `outputStyles`, `themes`, `monitors`, or `lspServers` paths are missing or point outside the plugin directory (v2.1.283+). + Example: `claude plugin tag ./my-plugin` takes a **path** to the plugin (not a version string). It creates a `{name}--v{version}` git tag derived from `plugin.json`, validating that `plugin.json` and any enclosing marketplace entry agree, and is the recommended way to cut plugin releases for distribution. `claude plugin prune` is useful after installing or uninstalling marketplace plugins that pulled in their own dependencies — it removes any auto-installed plugins whose parent plugin has since been removed. `plugin uninstall --prune` does the same cascade in a single step. @@ -846,7 +848,7 @@ LSP servers were added to the details pane in v2.1.142. See also the marketplace ### `claude plugin eval` (v2.1.269+) -`claude plugin eval` runs a plugin's eval suite and reports scored, reproducible results, so you can tell whether a change to a skill or agent actually improved it. Results come back as JSON plus an HTML report. +`claude plugin eval` runs a plugin's eval suite and reports scored, reproducible results, so you can tell whether a change to a skill or agent actually improved it. Results come back as JSON plus an HTML report. Since v2.1.283, it requires git 2.31 or later when git is installed. ```bash claude plugin eval # Run the plugin's eval suite @@ -1303,8 +1305,8 @@ The following Claude Code features work together with plugins: --- -**Last Updated**: September 19, 2026 -**Claude Code Version**: 2.1.278 +**Last Updated**: September 26, 2026 +**Claude Code Version**: 2.1.283 **Sources**: - https://code.claude.com/docs/en/plugins - https://code.claude.com/docs/en/plugins-reference @@ -1324,4 +1326,8 @@ The following Claude Code features work together with plugins: - https://code.claude.com/docs/en/cli-reference - https://code.claude.com/docs/en/model-config - https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md +- https://code.claude.com/docs/en/plugins/cli-reference +- https://code.claude.com/docs/en/plugin-evals +- https://github.com/anthropics/claude-code/releases/tag/v2.1.281 +- https://github.com/anthropics/claude-code/releases/tag/v2.1.283 **Compatible Models**: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 diff --git a/09-advanced-features/README.md b/09-advanced-features/README.md index 9012865..fb47d43 100644 --- a/09-advanced-features/README.md +++ b/09-advanced-features/README.md @@ -225,11 +225,11 @@ Extended thinking is a deliberate, step-by-step reasoning process where Claude: ### Activating Extended Thinking **Keyboard shortcut**: -- `Option + T` (macOS) / `Alt + T` (Windows/Linux) - Toggle extended thinking +- `Option + T` (macOS) / `Alt + T` (Windows/Linux) - Toggle extended thinking. On Opus 5.5 (the default model since v2.1.280) and the Fable models, thinking can't be turned off — the toggle, `alwaysThinkingEnabled`, and `MAX_THINKING_TOKENS=0` have no effect there. **Automatic activation**: - Enabled by default for all models (Opus 5, Opus 4.8, Opus 4.7, Sonnet 4.6, Haiku 4.5) -- Opus 5 / Opus 4.8: Adaptive reasoning with effort levels: `low` (○), `medium` (◐), `high` (●), `xhigh`, `max`. The default is `high` on Opus 5 (v2.1.219), Opus 4.8 (v2.1.154), Opus 4.6, and Sonnet 4.6, and `xhigh` on Opus 4.7. `xhigh` is available on Opus 5, Opus 4.8, and Opus 4.7 (it falls back to `high` on Opus 4.6 / Sonnet 4.6). `max` works on Opus 5 and Opus 4.8/4.7/4.6 and Sonnet 4.6 (session-only). Haiku 4.5 has no effort levels. Opus 5, Opus 4.8, and Opus 4.7 have a 1M-token native context window (1M context fix landed in v2.1.117 — before that, `/context` miscounted Opus 4.7 against a 200K window and triggered premature autocompact). Since v2.1.129, `/context` shows its visualization in-UI only; the ASCII viz no longer leaks into the conversation context (~1.6k tokens saved per call), so `/context` is safe to invoke freely. +- Opus 5 / Opus 4.8: Adaptive reasoning with effort levels: `low` (○), `medium` (◐), `high` (●), `xhigh`, `max`. The default is `medium` on Opus 5.5 (the default model since v2.1.280), `high` on Opus 5 (v2.1.219), Opus 4.8 (v2.1.154), Opus 4.6, and Sonnet 4.6, and `xhigh` on Opus 4.7. `xhigh` is available on Opus 5.5, Opus 5, Opus 4.8, and Opus 4.7 (it falls back to `high` on Opus 4.6 / Sonnet 4.6). `max` works on Opus 5.5, Opus 5, and Opus 4.8/4.7/4.6 and Sonnet 4.6 (session-only). Haiku 4.5 has no effort levels. Opus 5, Opus 4.8, and Opus 4.7 have a 1M-token native context window (1M context fix landed in v2.1.117 — before that, `/context` miscounted Opus 4.7 against a 200K window and triggered premature autocompact). Since v2.1.129, `/context` shows its visualization in-UI only; the ASCII viz no longer leaks into the conversation context (~1.6k tokens saved per call), so `/context` is safe to invoke freely. - Pro/Max subscribers on Opus 4.6 / Sonnet 4.6: default effort was raised from `medium` to `high` in v2.1.117. - Other models: Fixed budget up to 31,999 tokens @@ -851,19 +851,19 @@ Permission modes control what actions Claude can take without explicit approval. | `manual` | Read files only; prompts for all other actions. Renamed from `default` in v2.1.200 — `default` is still accepted as an alias | | `acceptEdits` | Read and edit files; prompts for commands | | `plan` | Read files only (research mode, no edits) | -| `auto` | All actions with background safety classifier checks. Requires an eligible model (Opus 5, Sonnet 5, Opus 4.7/4.8, or Fable 5 on most providers) and provider — available on all plans, see [Auto Mode](#auto-mode) | +| `auto` | All actions with background safety classifier checks. Requires an eligible model (Opus 5.5, Opus 5, Sonnet 5, Opus 4.7/4.8, or Fable 5 on most providers) and provider — available on all plans, see [Auto Mode](#auto-mode) | | `bypassPermissions` | All actions, no permission checks (dangerous) | | `dontAsk` | Only pre-approved tools execute; all others denied | > **Note**: The interactive default mode was renamed from `default` to **Manual** in v2.1.200 (across the CLI, `--help`, VS Code, and JetBrains), and a grey ⏸ badge appears in the footer while it is active (v2.1.203). Both `--permission-mode manual` and `--permission-mode default` work, as do `"defaultMode": "manual"` and `"defaultMode": "default"` in settings. Note that the settings key is `permissions.defaultMode` — there is no `permissions.mode` key, so the examples below use the canonical spelling. -Cycle through modes with `Shift+Tab` in the CLI. Set a default with the `--permission-mode` flag or the `permissions.defaultMode` setting. +Cycle through modes with `Shift+Tab` in the CLI. Set a default with the `--permission-mode` flag or the `permissions.defaultMode` setting. Since v2.1.283, an interactive session on a third-party provider (Bedrock, Google Cloud's Agent Platform, Foundry) or with telemetry off starts in auto mode when no permission mode is configured; `permissions.defaultMode` still overrides it. > **Plan mode defers shell commands to the classifier (v2.1.218)**: When [auto mode](#auto-mode) is available and the `useAutoModeDuringPlan` setting is on — which it is by default — the classifier reviews shell commands during planning instead of prompting you. Approved commands run, and rejected ones are blocked. Plan mode still blocks file writes unconditionally. As of v2.1.160, even `acceptEdits` prompts before writing shell-startup files (`.zshenv`, `.zlogin`, `.bash_login`, `~/.config/git/`) and code-executing build configs (`.npmrc`, `.yarnrc*`, `bunfig.toml`, `.bazelrc`, `.pre-commit-config.yaml`, `.devcontainer/`, …), which could otherwise lead to unintended command execution. -> **`--dangerously-skip-permissions` extended path coverage (v2.1.121, v2.1.126)**: The `--dangerously-skip-permissions` CLI flag (and equivalent `bypassPermissions` mode) now bypasses prompts for writes to a much broader allowlist — `.claude/skills/`, `.claude/agents/`, `.claude/commands/`, `.claude/`, `.git/`, `.vscode/`, and shell config files. Catastrophic removal commands (`rm -rf /`, etc.) still prompt in this mode (in auto mode they are judged by the classifier instead — see [Auto Mode](#auto-mode)). Treat the flag as a sharper tool than before; use it only in throwaway sandboxes. +> **`--dangerously-skip-permissions` extended path coverage (v2.1.121, v2.1.126)**: The `--dangerously-skip-permissions` CLI flag (and equivalent `bypassPermissions` mode) now bypasses prompts for writes to a much broader allowlist — `.claude/skills/`, `.claude/agents/`, `.claude/commands/`, `.claude/`, `.git/`, `.vscode/`, and shell config files. Catastrophic removal commands (`rm -rf /`, etc.) still prompt in this mode — since v2.1.281 that prompt waits 2 minutes, then denies the command with a rewrite hint so unattended sessions keep going (`CLAUDE_CODE_DISABLE_DANGEROUS_RM_TIMEOUT=1` turns this off) (in auto mode, critical-path removals go to the classifier instead, while other dangerous `rm` forms such as `rm -rf "$(pwd)"` still prompt — see [Auto Mode](#auto-mode)). Treat the flag as a sharper tool than before; use it only in throwaway sandboxes. > **Windows shell detection (v2.1.120, v2.1.126)**: Git for Windows / Git Bash is no longer required. When Git Bash is absent, Claude Code uses PowerShell as the shell tool. From v2.1.126 PowerShell is the *primary* shell when the PowerShell tool is enabled, and detection covers PowerShell 7 installed via the Microsoft Store, MSI without PATH, or as a `.NET global tool`. @@ -1293,7 +1293,7 @@ Claude Code supports keyboard shortcuts for efficiency. Here's the complete refe | `Option+T` / `Alt+T` | Toggle extended thinking | | `Option+O` / `Alt+O` | Toggle fast mode (`/fast`) | | `Ctrl+X` `Ctrl+K` | Stop all background subagents | -| `Ctrl+Enter` (or `Ctrl+X` `Ctrl+S`) | Send now — interrupt the current turn and send every queued message at once. Sent and queued messages stay gray until the model receives them (v2.1.275) | +| `Ctrl+Enter` (or `Ctrl+X` `Ctrl+S`) | Send now — send every queued message at once. Since v2.1.281, running tools and subagents move to the background instead of the turn being cancelled. Sent and queued messages stay gray until the model receives them (v2.1.275) | | `Ctrl+S` | Stash the current prompt; press again to restore it | | `Ctrl+_` | Undo the last edit to the prompt input | | `:` | Type `:` at the start of a word to open emoji shortcode completion, e.g. `:heart:` (v2.1.217+) | @@ -1386,7 +1386,7 @@ Keybindings support chord sequences (multi-key combinations): ``` **Keystroke syntax**: -- **Modifiers**: `ctrl`, `alt` (or `opt`), `shift`, `meta` (or `cmd`) +- **Modifiers**: `ctrl` (or `control`), `alt` (or `opt`, `option`, `meta`), `shift`, and `cmd` (or `command`, `super`, `win`) — `cmd` is a separate modifier, not an alias of `meta`, and is only detected in terminals that report the Super key - **Uppercase implies Shift**: `K` is equivalent to `shift+k` - **Special keys**: `escape`, `enter`, `return`, `tab`, `space`, `backspace`, `delete`, arrow keys @@ -2535,7 +2535,7 @@ export CLAUDE_CODE_GOAL_CHECKIN_MINUTES=30 # (v2.1.234) Minutes export CLAUDE_CODE_MCP_STARTUP_WAIT_MS=5000 # (v2.1.274) How long the first non-interactive turn waits for MCP servers that are still connecting. 0 = do not wait. export CLAUDE_CODE_WEBFETCH_DEADLINE_MS=300000 # (v2.1.268) Overrides WebFetch's new 300-second deadline. 0 disables the deadline. export CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS=32 # (v2.1.269) Raises the Workflow tool's per-run concurrent agent limit. Accepts 1-256. -export CLAUDE_CODE_AUTO_MODE_SERVER=0 # (v2.1.278) Opt out of server-side auto-mode checks on Bedrock, Vertex, Foundry, and gateways +export CLAUDE_CODE_AUTO_MODE_SERVER=0 # (v2.1.278; direct Anthropic API since v2.1.281) Opt out of server-side auto-mode classifier review ``` > **v2.1.223 — `CLAUDE_CODE_DISABLE_1M_CONTEXT` widened**: the variable now holds **every** @@ -2738,8 +2738,8 @@ For more information about Claude Code and related features: --- -**Last Updated**: September 19, 2026 -**Claude Code Version**: 2.1.278 +**Last Updated**: September 26, 2026 +**Claude Code Version**: 2.1.283 **Sources**: - https://code.claude.com/docs/en/output-styles - https://raw.githubusercontent.com/anthropics/claude-code/main/CHANGELOG.md @@ -2758,4 +2758,11 @@ For more information about Claude Code and related features: - https://code.claude.com/docs/en/settings.md - https://code.claude.com/docs/en/settings-reference - https://code.claude.com/docs/en/whats-new/2026-w34 +- https://code.claude.com/docs/en/model-config#adjust-effort-level +- https://code.claude.com/docs/en/interactive-mode +- https://code.claude.com/docs/en/keybindings#modifiers +- https://code.claude.com/docs/en/model-config#extended-thinking +- https://github.com/anthropics/claude-code/releases/tag/v2.1.281 +- https://github.com/anthropics/claude-code/releases/tag/v2.1.283 +- https://code.claude.com/docs/en/permission-modes#critical-paths **Compatible Models**: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 diff --git a/10-cli/README.md b/10-cli/README.md index 6c9d971..6e2febb 100644 --- a/10-cli/README.md +++ b/10-cli/README.md @@ -186,7 +186,7 @@ claude --model opusplan "design and implement the caching layer" | Flag | Description | Example | |------|-------------|---------| | `--system-prompt` | Replace entire default prompt | `claude --system-prompt "You are a Python expert"` | -| `--system-prompt-file` | Load prompt from file (print mode) | `claude -p --system-prompt-file ./prompt.txt "query"` | +| `--system-prompt-file` | Load prompt from file, replacing the default prompt | `claude -p --system-prompt-file ./prompt.txt "query"` | | `--append-system-prompt` | Append to default prompt | `claude --append-system-prompt "Always use TypeScript"` | | `--append-subagent-system-prompt` | Append text to every subagent's system prompt (non-interactive) | `claude -p --append-subagent-system-prompt "Cite sources" "query"` | | `--append-subagent-system-prompt-file` | (v2.1.261) Load that appended text from a file instead, for prompts too long to pass on the command line. Non-interactive only, and **cannot be combined** with `--append-subagent-system-prompt` | `claude -p --append-subagent-system-prompt-file ./subagent-rules.txt "query"` | @@ -209,10 +209,10 @@ claude -p --system-prompt-file ./prompts/code-reviewer.txt "review main.py" | Flag | Behavior | Interactive | Print | |------|----------|-------------|-------| | `--system-prompt` | Replaces entire default system prompt | ✅ | ✅ | -| `--system-prompt-file` | Replaces with prompt from file | ❌ | ✅ | +| `--system-prompt-file` | Replaces with prompt from file | ✅ | ✅ | | `--append-system-prompt` | Appends to default system prompt | ✅ | ✅ | -**Use `--system-prompt-file` only in print mode. For interactive mode, use `--system-prompt` or `--append-system-prompt`.** +**All system prompt flags work in both interactive and print mode (the file forms since v2.1.69). `--system-prompt` and `--system-prompt-file` replace the default prompt; the append flags add to it.** ## Tool & Permission Management @@ -789,8 +789,9 @@ Claude Code supports multiple models with different capabilities: | Model | ID | Context Window | Notes | |-------|-----|----------------|-------| -| Sonnet 5 | `claude-sonnet-5` | 1M tokens | Default on Pro / Team Standard / Enterprise seats (v2.1.197); native 1M-token context window. As of v2.1.219, **Opus 5** is the default Opus model on Max, Team Premium, Enterprise pay-as-you-go, and the Anthropic API; Microsoft Foundry still resolves the `opus` alias to Opus 4.6 | -| Opus 5 | `claude-opus-5` | 1M tokens | Default Opus model on Max, Team Premium, Enterprise pay-as-you-go, Anthropic API, Claude Platform on AWS, Amazon Bedrock, and Google Cloud's Agent Platform (v2.1.219); adaptive effort levels `low → max`, default effort `high` | +| Opus 5.5 | `claude-opus-5-5` | 1M tokens | Default model and default Opus model since v2.1.280 on Pro, Max, Team, Enterprise, the Anthropic API, Claude Platform on AWS, Amazon Bedrock, and Google Cloud's Agent Platform (Pro and Team Standard previously defaulted to Sonnet 5); Microsoft Foundry still resolves `opus` to Opus 4.6 and `default` to Sonnet 4.5. Adaptive effort levels `low → max`, default effort `medium`; thinking can't be turned off. Requires v2.1.280+ | +| Sonnet 5 | `claude-sonnet-5` | 1M tokens | Default on Pro / Team Standard seats from v2.1.197 until v2.1.280, when those plans moved to Opus 5.5; native 1M-token context window | +| Opus 5 | `claude-opus-5` | 1M tokens | Default Opus model from v2.1.219 until v2.1.280 (now Opus 5.5); still selectable; adaptive effort levels `low → max`, default effort `high` | | Opus 4.8 | `claude-opus-4-8` | 1M tokens | Previous flagship Opus, still selectable; adaptive effort levels `low → max`; default effort `high` (v2.1.154) | | Sonnet 4.6 | `claude-sonnet-4-6` | 1M tokens | Balanced speed and capability; default effort for Pro/Max subscribers raised from `medium` to `high` in v2.1.117 | | Haiku 4.5 | `claude-haiku-4-5` | 200K tokens | Fastest, best for quick tasks; no effort levels | @@ -814,11 +815,11 @@ claude --model opusplan "design and implement the API" > **Fable 5.1 and the `fable` alias (v2.1.257)**: Fable 5.1 (`claude-fable-5-1`) ships in **v2.1.257**, and the `fable` alias now resolves to it rather than to Fable 5. The official model-config page says Fable 5.1 "requires Claude Code v2.1.255 or later", but 2.1.255 was never published — v2.1.257 is the first release users can actually install it with. On a Claude apps gateway, `fable` and `best` still resolve to **Fable 5**; pick 5.1 explicitly in `/model` there. -> **Fast Mode runs on Opus 5 and Opus 4.8 (v2.1.219)**: As of v2.1.219, `/fast` applies to **Opus 5 and Opus 4.8** — Opus 4.7 was removed from fast mode. Opus 5's fast mode is billed at $10/$50 per Mtok. Fast mode first moved to Opus 4.8 in v2.1.154 (about 2× the standard rate for ~2.5× the output speed), having flipped from Opus 4.6 to Opus 4.7 in v2.1.142. The `CLAUDE_CODE_OPUS_4_6_FAST_MODE_OVERRIDE` env var was **deprecated in v2.1.154 and removed on 2026-06-01**; fast mode is no longer available on Opus 4.6 — select Opus 5 or Opus 4.8 instead. +> **Fast Mode runs on Opus 5 and Opus 4.8 (v2.1.219)**: As of v2.1.219, `/fast` applies to **Opus 5 and Opus 4.8** — Opus 4.7 was removed from fast mode. Opus 5's fast mode is billed at $10/$50 per Mtok. Fast mode first moved to Opus 4.8 in v2.1.154 (about 2× the standard rate for ~2.5× the output speed), having flipped from Opus 4.6 to Opus 4.7 in v2.1.142. The `CLAUDE_CODE_OPUS_4_6_FAST_MODE_OVERRIDE` env var was **deprecated in v2.1.154 and removed on 2026-06-01**; fast mode is no longer available on Opus 4.6 — select Opus 5 or Opus 4.8 instead. **Since v2.1.280, fast mode defaults to Opus 5.5** ($8/$40 per Mtok); Opus 5 and Opus 4.8 remain supported at $10/$50. ### Effort Levels (Opus 5 / Sonnet 5 / Opus 4.8 / Opus 4.7) -Opus 5, Sonnet 5, Opus 4.8, and Opus 4.7 support adaptive reasoning with effort levels, ordered from lightest to heaviest: `low` (○), `medium` (◐), `high` (●), `xhigh`, and `max`. The **default** is `high` on Opus 5, Sonnet 5, Opus 4.8 (since v2.1.154), Opus 4.6, and Sonnet 4.6, and `xhigh` on Opus 4.7. `xhigh` is available on Opus 5, Sonnet 5, Opus 4.8, and Opus 4.7; `max` works on Opus 5, Sonnet 5, Opus 4.8/4.7/4.6 and Sonnet 4.6 (session-only). Haiku 4.5 has no effort levels. On Opus 4.6 / Sonnet 4.6, the default effort for Pro/Max subscribers was raised from `medium` to `high` in v2.1.117. +Opus 5, Sonnet 5, Opus 4.8, and Opus 4.7 support adaptive reasoning with effort levels, ordered from lightest to heaviest: `low` (○), `medium` (◐), `high` (●), `xhigh`, and `max`. Opus 5.5 supports all five levels too. The **default** is `medium` on Opus 5.5 (the default model since v2.1.280) and `high` on Opus 5, Sonnet 5, Opus 4.8 (since v2.1.154), Opus 4.6, and Sonnet 4.6, and `xhigh` on Opus 4.7. `xhigh` is available on Opus 5.5, Opus 5, Sonnet 5, Opus 4.8, and Opus 4.7; `max` works on Opus 5.5, Opus 5, Sonnet 5, Opus 4.8/4.7/4.6 and Sonnet 4.6 (session-only). Haiku 4.5 has no effort levels. On Opus 4.6 / Sonnet 4.6, the default effort for Pro/Max subscribers was raised from `medium` to `high` in v2.1.117. ```bash # Set effort level via CLI flag @@ -847,7 +848,7 @@ The "ultrathink" keyword in prompts activates deep reasoning. The `/effort` menu | `ANTHROPIC_DEFAULT_SONNET_MODEL` | Override default Sonnet model ID | | `ANTHROPIC_DEFAULT_HAIKU_MODEL` | Override default Haiku model ID | | `MAX_THINKING_TOKENS` | Set extended thinking token budget | -| `CLAUDE_CODE_EFFORT_LEVEL` | Set effort level (`low`/`medium`/`high`/`xhigh`/`max`) — default is `high` on Opus 5, Sonnet 5, and Opus 4.8 (`xhigh` on Opus 4.7); `xhigh` needs Opus 5, Sonnet 5, or Opus 4.8/4.7; `max` works on Opus 5, Sonnet 5, Opus 4.8/4.7/4.6 and Sonnet 4.6 | +| `CLAUDE_CODE_EFFORT_LEVEL` | Set effort level (`low`/`medium`/`high`/`xhigh`/`max`) — default is `medium` on Opus 5.5, `high` on Opus 5, Sonnet 5, and Opus 4.8 (`xhigh` on Opus 4.7); `xhigh` needs Opus 5.5, Opus 5, Sonnet 5, or Opus 4.8/4.7; `max` works on Opus 5.5, Opus 5, Sonnet 5, Opus 4.8/4.7/4.6 and Sonnet 4.6 | | `CLAUDE_CODE_SIMPLE` | Minimal mode, set by `--bare` flag | | `CLAUDE_CODE_SAFE_MODE` | Set to `1` to start with all customizations disabled (CLAUDE.md, plugins, skills, hooks, MCP) — env-var form of `--safe-mode`, for isolating config problems (v2.1.169) | | `CLAUDE_CODE_DISABLE_BUNDLED_SKILLS` | Set to `1` to hide the bundled skills, workflows, and commands from the model (v2.1.169) | @@ -902,7 +903,8 @@ The "ultrathink" keyword in prompts activates deep reasoning. The `/effort` menu | `CLAUDE_CODE_MCP_STARTUP_WAIT_MS` | Bounds how long the first non-interactive turn waits for MCP servers that are still connecting. Set `0` to not wait at all (v2.1.274) | | `CLAUDE_CODE_WEBFETCH_DEADLINE_MS` | Overrides WebFetch's 300-second deadline, introduced in v2.1.268. Set `0` to disable the deadline (v2.1.268) | | `CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS` | Raises the Workflow tool's per-run concurrent agent limit. Accepts 1–256 (v2.1.269) | -| `CLAUDE_CODE_AUTO_MODE_SERVER` | Set to `0` to opt out of server-side auto-mode checks on Bedrock, Vertex, Foundry, and gateways (v2.1.278) | +| `CLAUDE_CODE_AUTO_MODE_SERVER` | Set to `0` to use Claude Code's own auto-mode classifier requests instead of server-side review — on Bedrock, Vertex, Foundry, and gateways since v2.1.278, and on a direct Anthropic API connection since v2.1.281 (there the local classifier then counts toward usage, and `1` opts in) | +| `CLAUDE_CODE_DISABLE_DANGEROUS_RM_TIMEOUT` | Set to `1` to turn off the 2-minute timeout on the dangerous `rm` prompt in `--dangerously-skip-permissions` and auto mode, which otherwise denies the command with a rewrite hint so unattended sessions keep going (v2.1.281) | | `CLAUDE_CODE_ENABLE_TODO_TOOLS` | Set to `1` to restore the todo/task-tracking tools (`TaskCreate`/`Get`/`Update`/`List`, `TodoWrite`), which are available by default only on Claude 3.x models, Opus 4 through 4.7, Sonnet 4 through 4.6, and Haiku 4.5 (v2.1.268) | | `CLAUDE_CODE_WEBFETCH_CACHE_TTL_MS` | How long WebFetch caches a fetched URL. Default 15 minutes (v2.1.233) | | `CLAUDE_CODE_TOOL_MEMORY_LIMIT` | Linux only: opt in to a memory cgroup applied to Bash commands (v2.1.233) | @@ -920,6 +922,8 @@ The "ultrathink" keyword in prompts activates deep reasoning. The `/effort` menu > **`ENABLE_TOOL_SEARCH` on Vertex AI (v2.1.119+)**: Tool search is **disabled by default on Google Cloud Vertex AI** deployments. Users who want the tool-search capability on Vertex must explicitly opt in with `export ENABLE_TOOL_SEARCH=true`. On direct Anthropic API it remains enabled by default. +> **OpenTelemetry variables in project settings (v2.1.282)**: project and local settings ignore OpenTelemetry variables that turn on export, set its endpoint, or capture content, such as `CLAUDE_CODE_ENABLE_TELEMETRY` and `OTEL_LOG_*`. + --- ## Settings.json Keys @@ -940,12 +944,16 @@ These keys live in a `settings.json` file (`~/.claude/settings.json` for user sc | `promptCacheTtl` | (v2.1.243) Choose the prompt cache lifetime for the main conversation. | | `subagentPromptCacheTtl` | (v2.1.243) The same choice for subagents and other requests outside the main conversation. | | `modelPricing` | (v2.1.243) **Managed setting.** Supplies your organization's contracted rates so `/cost`, the status line, and telemetry report those instead of list price. | +| `availableModelsMatch` | (v2.1.283) **Managed setting.** With `"exact"`, an `availableModels` entry allows only the model version it names, so new releases stay blocked until listed. | +| `deniedModels` | (v2.1.283) **Managed setting.** Blocks specific models, even when `availableModels` allows them. | | `keybindingFlavor` | **Deprecated since v2.1.261 and has no effect.** The prompt's word-editing keys always follow readline conventions, as Bash does: `Ctrl+W` deletes back to whitespace, `Alt+F` and `Alt+D` stop at word end, and punctuation separates words. Claude Code still accepts the key, so a settings file that sets it stays valid. (In v2.1.238–v2.1.260 it chose between `"classic"` and `"readline"`.) | | `bashOutputMaxChars` | (v2.1.261) How many characters of a **successful** Bash or PowerShell command's output Claude receives inline, up to 128K. Past the limit Claude Code saves the output to a file and Claude gets a short preview plus the path. Setting it makes Claude Code ignore `BASH_MAX_OUTPUT_LENGTH`. | | `taskOutputMaxChars` | (v2.1.261) **Superseded in v2.1.278 and has no effect.** It used to cap how many characters of a **background task's** output Claude received inline when reading it with the `TaskOutput` tool. v2.1.278 removed the `TaskOutput` tool — Claude now reads a background task's output file with `Read` — so this key and `TASK_MAX_OUTPUT_LENGTH` are both inert. Claude Code still accepts the key, so a settings file that sets it stays valid. | | `maxEffortLevel` | (v2.1.267) Caps the effort level Claude Code may use, on every provider including Bedrock, Vertex, and Foundry. Set it top-level or per-model under `modelSettings`. Users can still pick a lower level. | | `bashEditDiffEnabled` | (v2.1.269) Bash tool results include a diff of the files the command changed. | | `syncClaudeAiSkills` / `syncClaudeAiPlugins` | (v2.1.275) Set either to `false` to opt out of syncing the skills or plugins you have enabled on your claude.ai account into terminal sessions. | +| `attribution` | (v2.1.281) Set `"attribution": false` to hide all commit and PR attribution. Older CLI versions skip a settings file that holds it, so keep the object form in files shared across versions. | +| `maxProseWidth` | (v2.1.282) Caps the width of Claude's prose in wide terminals; tables and code blocks keep the full width. | ```json { @@ -1059,8 +1067,8 @@ claude -p --output-format json "query" --- -**Last Updated**: September 19, 2026 -**Claude Code Version**: 2.1.278 +**Last Updated**: September 26, 2026 +**Claude Code Version**: 2.1.283 **Sources**: - https://raw.githubusercontent.com/anthropics/claude-code/main/CHANGELOG.md - https://code.claude.com/docs/en/workflows#set-a-size-guideline @@ -1088,4 +1096,8 @@ claude -p --output-format json "query" - https://code.claude.com/docs/en/cli-reference.md - https://code.claude.com/docs/en/settings.md - https://code.claude.com/docs/en/settings-reference +- https://code.claude.com/docs/en/fast-mode +- https://github.com/anthropics/claude-code/releases/tag/v2.1.281 +- https://github.com/anthropics/claude-code/releases/tag/v2.1.282 +- https://github.com/anthropics/claude-code/releases/tag/v2.1.283 **Compatible Models**: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 diff --git a/CATALOG.md b/CATALOG.md index c4500fd..2a66f41 100644 --- a/CATALOG.md +++ b/CATALOG.md @@ -41,7 +41,7 @@ Commands are user-invoked shortcuts that execute specific actions. | `/diff` | Interactive diff viewer. In the fullscreen TUI (v2.1.260+) it opens a diff panel beside the conversation that stays open and refreshes each time Claude edits a file or runs a command; the classic renderer opens the viewer in place of the prompt instead | Review changes | | `/config` | View/edit configuration | Customize behavior | | `/status` | Show session status | Check current state | -| `/agents` | List available agents | See delegation options | +| `/agents` | Removed: the wizard went in v2.1.198 and the command left the `/` menu and `/help` in v2.1.281 — ask Claude to create or manage subagents, or edit `.claude/agents/` | Manage subagents by asking Claude | | `/skills` | List available skills | See auto-invoke capabilities | | `/hooks` | List configured hooks | Debug automation | | `/insights` | Analyze session patterns | Session optimization | @@ -89,7 +89,7 @@ Commands are user-invoked shortcuts that execute specific actions. | `/stats` | Shortcut alias that opens the stats tab of `/usage` (v2.1.118+) | Review session metrics | | `/statusline` | Configure status line | Customize status display | | `/stickers` | View session stickers | Fun rewards | -| `/fast` | Toggle fast output mode; applies to **Opus 5 and Opus 4.8** (v2.1.219) | Speed up responses | +| `/fast` | Toggle fast output mode; applies to **Opus 5.5** (the default since v2.1.280), **Opus 5, and Opus 4.8** | Speed up responses | | `/terminal-setup` | Configure terminal integration | Setup terminal features | | `/undo` | **No longer documented** — added as an alias for `/rewind` in v2.1.108, but it appears nowhere in the official commands reference | Use `/rewind` (or `Esc Esc`) instead | | `/upgrade` | Check for updates | Version management | @@ -573,8 +573,8 @@ chmod +x ~/.claude/hooks/*.sh --- -**Last Updated**: September 19, 2026 -**Claude Code Version**: 2.1.278 +**Last Updated**: September 26, 2026 +**Claude Code Version**: 2.1.283 **Sources**: - https://code.claude.com/docs/en/sub-agents - https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md @@ -595,4 +595,5 @@ chmod +x ~/.claude/hooks/*.sh - https://code.claude.com/docs/en/settings - https://code.claude.com/docs/en/plugins-reference - https://code.claude.com/docs/en/slash-commands +- https://code.claude.com/docs/en/fast-mode **Compatible Models**: Claude Fable 5.1, Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 diff --git a/INDEX.md b/INDEX.md index 8da4af6..49a6438 100644 --- a/INDEX.md +++ b/INDEX.md @@ -482,7 +482,7 @@ Command-line interface usage patterns and reference. - `claude` - Start interactive session - `claude -p "prompt"` - Headless/non-interactive mode - `claude web` - Launch web session -- `claude --model` - Select model (Opus 5, Sonnet 5, Sonnet 4.6, Opus 4.8, Haiku 4.5) +- `claude --model` - Select model (Opus 5.5, Opus 5, Sonnet 5, Sonnet 4.6, Opus 4.8, Haiku 4.5) - `claude --permission-mode` - Set permission mode - `claude --remote` - Enable remote control via WebSocket @@ -897,8 +897,8 @@ Want to add more examples? Follow the structure: --- -**Last Updated**: September 19, 2026 -**Claude Code Version**: 2.1.278 +**Last Updated**: September 26, 2026 +**Claude Code Version**: 2.1.283 **Sources**: - https://code.claude.com/docs/en/tools-reference#task-tool-availability - https://code.claude.com/docs/en/workflows#set-a-size-guideline diff --git a/QUICK_REFERENCE.md b/QUICK_REFERENCE.md index 928a69d..34f5b6a 100644 --- a/QUICK_REFERENCE.md +++ b/QUICK_REFERENCE.md @@ -402,7 +402,7 @@ cp -r 03-skills/code-review-specialist ~/.claude/skills/ | **Auto Memory** | Automatic memory saving from conversations | Claude auto-saves key context to CLAUDE.md | | **AGENTS.md** *(v2.1.277)* | Cross-tool project-context file, read as project instructions when no `CLAUDE.md` exists at or above the working directory | Switch modes with **Project instructions** in `/config`. See [AGENTS.md](02-memory/README.md#agentsmd) | | **Git Worktrees** | Isolated workspaces for parallel development | `/worktree` to create isolated workspace | -| **Model Selection** | Switch between Fable 5.1, Fable 5, Opus 5, Sonnet 5, Sonnet 4.6, Opus 4.8, and Haiku 4.5 | `/model` — since v2.1.153 the choice is saved as the default for new sessions; press `s` for session-only | +| **Model Selection** | Switch between Fable 5.1, Fable 5, Opus 5.5, Opus 5, Sonnet 5, Sonnet 4.6, Opus 4.8, and Haiku 4.5 | `/model` — since v2.1.153 the choice is saved as the default for new sessions; press `s` for session-only | | **Agent Teams** | Coordinate multiple agents on tasks | Enable with `CLAUDE_AGENT_TEAMS=1` env var | | **Dynamic Workflows** *(v2.1.154)* | Deterministic multi-agent orchestration; since v2.1.219 the default size guideline is medium (aim for fewer than 10 agents) | `/workflows` to view runs; ask Claude to create one; change the size via **Dynamic workflow size** in `/config` | | **Scheduled Tasks** | Recurring tasks with `/loop` | `/loop 5m /command` or CronCreate tool | @@ -520,8 +520,8 @@ Getting started checklist: --- -**Last Updated**: September 19, 2026 -**Claude Code Version**: 2.1.278 +**Last Updated**: September 26, 2026 +**Claude Code Version**: 2.1.283 **Sources**: - https://code.claude.com/docs/en/workflows#set-a-size-guideline - https://code.claude.com/docs/en/memory#agents-md diff --git a/README.md b/README.md index 4cc5e8f..ccb4c97 100644 --- a/README.md +++ b/README.md @@ -12,7 +12,7 @@ [![GitHub Stars](https://img.shields.io/github/stars/luongnv89/claude-howto?style=flat&color=gold)](https://github.com/luongnv89/claude-howto/stargazers) [![GitHub Forks](https://img.shields.io/github/forks/luongnv89/claude-howto?style=flat)](https://github.com/luongnv89/claude-howto/network/members) [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE) -[![Version](https://img.shields.io/badge/version-2.1.278-brightgreen)](CHANGELOG.md) +[![Version](https://img.shields.io/badge/version-2.1.283-brightgreen)](CHANGELOG.md) [![Claude Code](https://img.shields.io/badge/Claude_Code-2.1+-purple)](https://code.claude.com) 🌐 **Language / Ngôn ngữ / 语言 / Мова:** [English](README.md) | [Tiếng Việt](vi/README.md) | [中文](zh/README.md) | [Українська](uk/README.md) | [日本語](ja/README.md) @@ -102,7 +102,7 @@ Run `/lesson-quiz [topic]` after each module. The quiz pinpoints what you missed - **GitHub stars** from developers who use Claude Code daily - **Forks** from teams adapting this guide for their own workflows -- **Actively maintained** — synced with every Claude Code release (latest: v2.1.278, September 2026) +- **Actively maintained** — synced with every Claude Code release (latest: v2.1.283, September 2026) - **Community-driven** — contributions from developers who share their real-world configurations [![Star History Chart](https://api.star-history.com/svg?repos=luongnv89/claude-howto&type=Date)](https://star-history.com/#luongnv89/claude-howto&Date) @@ -201,7 +201,7 @@ cp -r 03-skills/code-review-specialist ~/.claude/skills/ Yes. MIT licensed, free forever. Use it in personal projects, at work, in your team — no restrictions beyond including the license notice. **Is this maintained?** -Actively. The guide is synced with every Claude Code release. Current version: v2.1.278 (September 2026), compatible with Claude Code 2.1+. +Actively. The guide is synced with every Claude Code release. Current version: v2.1.283 (September 2026), compatible with Claude Code 2.1+. **How is this different from the official docs?** The official docs are a feature reference. This guide is a tutorial with diagrams, production-ready templates, and a progressive learning path. They complement each other — start here to learn, reference the docs when you need specifics. @@ -210,7 +210,7 @@ The official docs are a feature reference. This guide is a tutorial with diagram 11-13 hours for the full path. But you'll get immediate value in 15 minutes — just copy a slash command template and try it. **Can I use this with Claude Sonnet / Haiku / Opus?** -Yes. All templates work with Claude Fable 5.1, Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, and Claude Haiku 4.5. +Yes. All templates work with Claude Fable 5.1, Claude Fable 5, Claude Opus 5.5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, and Claude Haiku 4.5. **Can I contribute?** Absolutely. See [CONTRIBUTING.md](CONTRIBUTING.md) for guidelines. We welcome new examples, bug fixes, documentation improvements, and community templates. @@ -873,8 +873,8 @@ MIT License - see [LICENSE](LICENSE). Free to use, modify, and distribute. The o --- -**Last Updated**: September 19, 2026 -**Claude Code Version**: 2.1.278 +**Last Updated**: September 26, 2026 +**Claude Code Version**: 2.1.283 **Sources**: - https://code.claude.com/docs/en/tools-reference#task-tool-availability - https://code.claude.com/docs/en/hooks diff --git a/claude_concepts_guide.md b/claude_concepts_guide.md index fcb8cf5..a82a8d3 100644 --- a/claude_concepts_guide.md +++ b/claude_concepts_guide.md @@ -1108,7 +1108,7 @@ graph TD ### Week 3: Distribute Work - Create first Subagent for specific role -- Use `/agents` command +- Ask Claude to create it, or add a `.claude/agents/.md` file (the `/agents` wizard was removed) - Test delegation with simple task ### Week 4: Automate Everything @@ -1453,14 +1453,15 @@ Claude Code supports the following models with adaptive reasoning effort: | Model | Context Window | Effort Levels | Default Effort (Claude Code) | |-------|----------------|---------------|------------------------------| -| Claude Opus 5 | 1M tokens (native) | `low`, `medium`, `high`, `xhigh`, `max` | `high` — default Opus model since v2.1.219 (requires Claude Code v2.1.219+) | -| Claude Sonnet 5 | 1M tokens (native) | `low`, `medium`, `high`, `xhigh`, `max` | `high` — default model for Pro/Team Standard/Enterprise since v2.1.197 | +| Claude Opus 5.5 | 1M tokens (native) | `low`, `medium`, `high`, `xhigh`, `max` | `medium` — default model and default Opus model since v2.1.280 (requires Claude Code v2.1.280+) | +| Claude Opus 5 | 1M tokens (native) | `low`, `medium`, `high`, `xhigh`, `max` | `high` — default Opus model from v2.1.219 until v2.1.280 (requires Claude Code v2.1.219+) | +| Claude Sonnet 5 | 1M tokens (native) | `low`, `medium`, `high`, `xhigh`, `max` | `high` — default model for Pro/Team Standard from v2.1.197 until v2.1.280 (now Opus 5.5) | | Claude Opus 4.8 | 1M tokens (native) | `low`, `medium`, `high`, `xhigh`, `max` | `high` (since v2.1.154) | | Claude Opus 4.7 (legacy) | 1M tokens (native) | `low`, `medium`, `high`, `xhigh`, `max` | `xhigh` (since Opus 4.7 launch, 2026-04-16) | | Claude Sonnet 4.6 | 1M tokens | `low`, `medium`, `high`, `max` | `high` for Pro/Max subscribers (raised from `medium` in v2.1.117) | | Claude Haiku 4.5 | 200K tokens | — (no effort support) | — | -> **Note**: `xhigh` is available on Opus 5, Sonnet 5, Opus 4.8, and Opus 4.7; `max` works on Opus 5, Sonnet 5, Opus 4.8/4.7/4.6 and Sonnet 4.6 (session-only). Haiku 4.5 does not support effort levels. +> **Note**: `xhigh` is available on Opus 5.5, Opus 5, Sonnet 5, Opus 4.8, and Opus 4.7; `max` works on Opus 5.5, Opus 5, Sonnet 5, Opus 4.8/4.7/4.6 and Sonnet 4.6 (session-only). Haiku 4.5 does not support effort levels. > **Note**: v2.1.117 fixed a bug where Opus 4.7 sessions computed `/context` against 200K instead of the native 1M window — upgrade to v2.1.117 or later to actually get the 1M context on Opus 4.7. Opus 5 and Opus 4.8 also have a native 1M-token window. @@ -1474,8 +1475,8 @@ Claude Code supports the following models with adaptive reasoning effort: - [Anthropic Cookbook](https://github.com/anthropics/anthropic-cookbook) --- -**Last Updated**: September 19, 2026 -**Claude Code Version**: 2.1.278 +**Last Updated**: September 26, 2026 +**Claude Code Version**: 2.1.283 **Sources**: - https://code.claude.com/docs/en/tools-reference#task-tool-availability - https://www.anthropic.com/news/claude-sonnet-5 @@ -1483,4 +1484,5 @@ Claude Code supports the following models with adaptive reasoning effort: - https://code.claude.com/docs/en/model-config - https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md - https://code.claude.com/docs/en/hooks +- https://code.claude.com/docs/en/sub-agents **Compatible Models**: Claude Fable 5.1, Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 diff --git a/ja/01-slash-commands/README.md b/ja/01-slash-commands/README.md index 028c6fe..7f2fc96 100644 --- a/ja/01-slash-commands/README.md +++ b/ja/01-slash-commands/README.md @@ -27,7 +27,7 @@ | コマンド | 用途 | |---------|------| | `/add-dir ` | 作業ディレクトリを追加 | -| `/agents` | エージェント設定を管理 | +| `/agents` | 現在は動作するコマンドではない。対話型ウィザードは v2.1.198 で削除され、v2.1.281 以降 `/agents` は `/` メニューにも `/help` にも表示されない — 入力してもウィザードの移動先が説明されるだけ。サブエージェントの作成や管理は Claude に依頼するか、`.claude/agents/` のファイルを直接編集する | | `/branch [name]` | 会話を分岐させて新しいセッションにする（エイリアス: `/fork`）。注: `/fork` は v2.1.77 で `/branch` に改名 | | `/btw ` | Claude がメインタスクを進めている間に、メイン会話のコンテキストを汚さず一時的なサイド質問をする | | `/chrome` | Chrome ブラウザ統合を設定 | @@ -603,8 +603,8 @@ Deploy the application to production: --- -**最終更新**: 2026 年 9 月 19 日 -**Claude Code バージョン**: 2.1.278 +**最終更新**: 2026 年 9 月 26 日 +**Claude Code バージョン**: 2.1.283 **Sources**: - https://code.claude.com/docs/en/commands - https://code.claude.com/docs/en/output-styles @@ -612,6 +612,7 @@ Deploy the application to production: - https://code.claude.com/docs/en/changelog - https://github.com/anthropics/claude-code/releases/tag/v2.1.118 - https://github.com/anthropics/claude-code/releases/tag/v2.1.116 +- https://github.com/anthropics/claude-code/releases/tag/v2.1.281 **Compatible Models**: Claude Sonnet 4.6, Claude Opus 4.7, Claude Haiku 4.5 *[Claude How To](../) ガイドシリーズの一部* diff --git a/ja/03-skills/README.md b/ja/03-skills/README.md index 09f6bc1..f746b89 100644 --- a/ja/03-skills/README.md +++ b/ja/03-skills/README.md @@ -140,7 +140,7 @@ Show concrete examples of using this Skill. ### 推奨フィールド - **description**（推奨）: スキルが何をするかと、いつ使うか。省略した場合、Claude Code は markdown 本文の最初の段落を使用する。`description` と `when_to_use` を合わせたテキストはスキル一覧で **1,536 文字**で切り詰められる（`skillListingMaxDescChars` で変更可能）。Claude がスキルを発火させるべきタイミングを判断する際に照合するのはこのテキストである。 -- **name**（任意）: 既定ではスキルの**ディレクトリ名**が使われる。指定した場合は表示名となり、小文字・数字・ハイフンのみ（最大 64 文字）で、"anthropic" や "claude" を含めることはできない。プラグインのスキルでは、`name` はコマンドの最後のセグメントも決定する。 +- **name**（任意）: 既定ではスキルの**ディレクトリ名**が使われる。指定した場合は表示名となり、小文字・数字・ハイフンのみ（最大 64 文字）。Claude Code では `anthropic-skills` と `synced` が予約されており、プラグイン外では読み込まれない。"anthropic" や "claude" を含められないというより厳しい制限は、claude.ai や Skills API にアップロードするスキルに適用される。プラグインのスキルでは、`name` はコマンドの最後のセグメントも決定する。 SKILL.md のフロントマターのフィールドはすべて任意であり、推奨されるのは `description` だけである。 @@ -171,7 +171,7 @@ paths: "src/api/**/*.ts" # スキルの発火を制限する glob | フィールド | 説明 | |-------|-------------| -| `name` | 小文字、数字、ハイフンのみ（最大 64 文字）。"anthropic" や "claude" を含められない。 | +| `name` | 小文字、数字、ハイフンのみ（最大 64 文字）。Claude Code では `anthropic-skills` と `synced` が予約されている。 | | `description` | スキルが何をするかと、いつ使うか。`description` と `when_to_use` を合わせたテキストはスキル一覧で 1,536 文字で切り詰められる（`skillListingMaxDescChars` で変更可能）。自動呼び出しのマッチングに重要。 | | `when_to_use` | Claude がスキルを呼び出すべきタイミングに関する補足情報。スキル一覧では `description` の後ろに連結され、1,536 文字の上限に算入される。 | | `argument-hint` | `/` の自動補完メニューに表示されるヒント（例: `"[filename] [format]"`）。 | @@ -826,8 +826,8 @@ chmod +x ~/.claude/skills/my-skill/scripts/*.py - [フックガイド](../06-hooks/) - イベント駆動の自動化 --- -**最終更新**: 2026 年 9 月 2 日 -**Claude Code バージョン**: 2.1.257 +**最終更新**: 2026 年 9 月 26 日 +**Claude Code バージョン**: 2.1.283 **情報源**: - https://code.claude.com/docs/en/skills - https://code.claude.com/docs/en/settings diff --git a/ja/03-skills/claude-md/SKILL.md b/ja/03-skills/claude-md/SKILL.md index bcac448..9e733ab 100644 --- a/ja/03-skills/claude-md/SKILL.md +++ b/ja/03-skills/claude-md/SKILL.md @@ -196,7 +196,7 @@ Brief one-line description. ユーザーが AGENTS.md の作成・更新を求めた場合: -**v2.1.277 以降、Claude Code は `AGENTS.md` をプロジェクト指示として直接読み込む。** ただし、作業ディレクトリとその上位すべてのディレクトリに `CLAUDE.md`、`.claude/CLAUDE.md`、`CLAUDE.local.md` のいずれも存在しない場合に限る。`~/.claude/CLAUDE.md`、管理 CLAUDE.md、`.claude/rules/` はこの判定の対象外であり、引き続き併せて読み込まれる。この挙動は `/config` の **Project instructions** で制御する: `claude-md-or-agents-md`（デフォルト）、`claude-md-and-agents-md`、`claude-md`、`managed-only`。直接読み込みが利用できない環境 — Bedrock/Vertex/Foundry、テレメトリ無効、アップグレード後の最初のセッション、`disableAllHooks`/`allowManagedHooksOnly` — では、CLAUDE.md から `@AGENTS.md` でインポートするか、`CLAUDE.md` をこのファイルへのシンボリックリンクにするフォールバックを使う。 +**v2.1.277 以降、Claude Code は `AGENTS.md` をプロジェクト指示として直接読み込む。** ただし、作業ディレクトリとその上位すべてのディレクトリに `CLAUDE.md`、`.claude/CLAUDE.md`、`CLAUDE.local.md` のいずれも存在しない場合に限る。`~/.claude/CLAUDE.md`、管理 CLAUDE.md、`.claude/rules/` はこの判定の対象外であり、引き続き併せて読み込まれる。この挙動は `/config` の **Project instructions** で制御する: `claude-md-or-agents-md`（デフォルト）、`claude-md-and-agents-md`、`claude-md`、`managed-only`。直接読み込みが利用できない環境 — v2.1.277 より前のバージョン（Bedrock/Vertex/Foundry、LLM ゲートウェイ、テレメトリ無効の場合は v2.1.281 より前）、またはアップグレード後の最初のセッション — では、CLAUDE.md から `@AGENTS.md` でインポートするか、`CLAUDE.md` をこのファイルへのシンボリックリンクにするフォールバックを使う。 AGENTS.md はツール横断のプロジェクトコンテキストファイルであり、CLAUDE.md と同じ *種類* のドキュメントである。エージェント定義フォーマットではない。複数のコーディングエージェントが 1 つのプロジェクト規約を共有できるようにするために存在する: - ビルド、テスト、リントのコマンド diff --git a/ja/04-subagents/README.md b/ja/04-subagents/README.md index 990f901..35e0245 100644 --- a/ja/04-subagents/README.md +++ b/ja/04-subagents/README.md @@ -51,7 +51,7 @@ 各サブエージェントは独立して動作し、クリーンスレートの状態でそのタスクに必要な特定のコンテキストのみを受け取り、結果をメインエージェントに返して統合する。 -**クイックスタート**：`/agents` コマンドを使うと、サブエージェントを対話的に作成・表示・編集・管理できる。 +**クイックスタート**：Claude にサブエージェントの作成を頼む（「セキュリティをレビューするサブエージェントを作成して」）か、`.claude/agents/.md` ファイルを直接追加する。詳しくは後述の[サブエージェントの管理](#サブエージェントの管理)を参照。 --- @@ -334,17 +334,13 @@ Claude Code には常に利用可能な組み込みサブエージェントが ### `/agents` コマンドを使う（推奨） -```bash -/agents +> **注**：`/agents` の対話型ウィザードは v2.1.198 で削除され、v2.1.281 以降は `/agents` が `/` メニューにも `/help` にも表示されない。サブエージェントの作成や管理は Claude に頼む（例：「セキュリティをレビューするサブエージェントを作成して」）か、`.claude/agents/.md` ファイルを直接追加すること。 + +```text +セキュリティ脆弱性をレビューするサブエージェントを作成して。 ``` -これは以下を行う対話メニューを提供する： - -- 利用可能なすべてのサブエージェント（組み込み、ユーザー、プロジェクト）の表示 -- ガイド付きセットアップで新しいサブエージェントを作成 -- 既存のカスタムサブエージェントとツールアクセスの編集 -- カスタムサブエージェントの削除 -- 重複がある場合にどのサブエージェントが有効かを確認 +Claude が `.claude/agents/.md` ファイルを書き、適切なフロントマター（ツール、モデル、説明）を選んでくれる。その後、ファイルを手で調整するか、Claude に修正を頼めばよい。 ### 直接ファイル管理 @@ -531,7 +527,7 @@ description: Performs long-running analysis tasks in the background | ショートカット | 動作 | |-------------|------| | `Ctrl+B` | 現在実行中のサブエージェントタスクをバックグラウンド化 | -| `Ctrl+F` | バックグラウンドエージェントをすべて終了（確認のため 2 回押下） | +| `Ctrl+X` `Ctrl+K` | 実行中のバックグラウンドサブエージェントをすべて停止（確認のため 3 秒以内に 2 回押下。`Ctrl+F` から置き換わった） | ### バックグラウンドタスクの無効化 @@ -1150,17 +1146,14 @@ graph TB ### 方法 1：/agents コマンドを使う（推奨） -```bash -/agents +> **注**：`/agents` の対話型ウィザードは v2.1.198 で削除され、v2.1.281 以降は `/agents` が `/` メニューにも `/help` にも表示されない。サブエージェントの作成や管理は Claude に頼む（例：「セキュリティをレビューするサブエージェントを作成して」）か、`.claude/agents/.md` ファイルを直接追加すること。 + +```text +テストを実行して失敗を修正する、プロジェクトレベルのサブエージェントを作成して。 +Bash、Read、Edit、Grep へのアクセスを与えて。 ``` -その後： - -1. 'Create New Agent' を選択 -2. プロジェクトレベルかユーザーレベルを選択 -3. サブエージェントを詳しく説明する -4. アクセスを許可するツールを選択（または空欄ですべてを継承） -5. 保存して使用 +Claude が適切なフロントマター付きで `.claude/agents/.md` を書く。生成されたファイルを確認してから使用する。 ### 方法 2：プロジェクトにコピー @@ -1196,13 +1189,13 @@ cp /path/to/04-subagents/debugger.md ~/.claude/agents/ ### 検証 -インストール後、エージェントが認識されているか確認する： +インストール後、ディレクトリを一覧表示してエージェントが認識されているか確認する： ```bash -/agents +ls .claude/agents/ ``` -組み込みエージェントとともにインストールしたエージェントが一覧表示されるはずである。 +現在のセッションで利用可能なサブエージェントを Claude に尋ねることもできる。Claude は委任できる組み込みエージェントとカスタムエージェントを報告する。 --- @@ -1275,12 +1268,14 @@ graph TD --- -**最終更新**：2026 年 8 月 25 日 -**Claude Code バージョン**：2.1.245 +**最終更新**：2026 年 9 月 26 日 +**Claude Code バージョン**：2.1.283 **ソース**： - https://code.claude.com/docs/en/sub-agents - https://code.claude.com/docs/en/agent-teams +- https://code.claude.com/docs/en/commands +- https://code.claude.com/docs/en/interactive-mode - https://github.com/anthropics/claude-code/releases/tag/v2.1.117 - https://github.com/anthropics/claude-code/releases/tag/v2.1.119 diff --git a/ja/05-mcp/README.md b/ja/05-mcp/README.md index 5fc985e..b2ed449 100644 --- a/ja/05-mcp/README.md +++ b/ja/05-mcp/README.md @@ -222,16 +222,17 @@ sequenceDiagram ## MCP ツール検索 -MCP ツールの説明文がコンテキストウィンドウの 10% を超える場合、Claude Code は自動的にツール検索を有効にし、モデルコンテキストを圧迫せずに適切なツールを効率的に選択する。 +ツール検索はデフォルトで有効。MCP ツールは遅延され、必要に応じて検出されるため、その定義がモデルコンテキストを埋めることはない。`ANTHROPIC_BASE_URL` がファーストパーティ以外のホストを指している場合、ほとんどのプロキシは必要なブロックを転送しないため、Claude Code はツール検索を無効にする。 | 設定 | 値 | 説明 | |------|-----|------| -| `ENABLE_TOOL_SEARCH` | `auto`（デフォルト） | ツールの説明がコンテキストの 10% を超えた時点で自動的に有効化 | -| `ENABLE_TOOL_SEARCH` | `auto:` | カスタムしきい値 `N` 個のツール数で自動的に有効化 | -| `ENABLE_TOOL_SEARCH` | `true` | ツール数に関係なく常に有効 | +| `ENABLE_TOOL_SEARCH` | （未設定 — デフォルト） | すべての MCP ツールを遅延し、必要に応じて読み込む | +| `ENABLE_TOOL_SEARCH` | `auto` | しきい値モード：ツール定義がコンテキストウィンドウの 10% に達するまではツールを事前に読み込み、達した時点ですべてを遅延する | +| `ENABLE_TOOL_SEARCH` | `auto:` | カスタムのパーセンテージ（0-100）によるしきい値モード。例：`auto:5` で 5% | +| `ENABLE_TOOL_SEARCH` | `true` | すべての MCP ツールを遅延 | | `ENABLE_TOOL_SEARCH` | `false` | 無効。すべてのツールの説明が完全な形で送信される | -> **注意：** ツール検索には Sonnet 4 以降、もしくは Opus 4 以降が必要。Haiku モデルはツール検索をサポートしない。 +> **注意：** ツール検索には `tool_reference` ブロックをサポートするモデルが必要：Claude Sonnet 4.5、Claude Haiku 4.5、Claude Opus 4.5、およびそれ以降のモデル。 ## 動的ツール更新 @@ -1167,10 +1168,11 @@ export GITHUB_TOKEN="your_token" --- -**最終更新：** 2026 年 9 月 6 日 -**Claude Code バージョン：** 2.1.263 +**最終更新：** 2026 年 9 月 26 日 +**Claude Code バージョン：** 2.1.283 **情報源：** - https://code.claude.com/docs/en/mcp +- https://code.claude.com/docs/en/mcp#configure-tool-search - https://code.claude.com/docs/en/managed-mcp - https://code.claude.com/docs/en/changelog - https://github.com/anthropics/claude-code/releases/tag/v2.1.117 diff --git a/ja/06-hooks/README.md b/ja/06-hooks/README.md index 2980dbd..42e7364 100644 --- a/ja/06-hooks/README.md +++ b/ja/06-hooks/README.md @@ -179,6 +179,8 @@ LLM はプロンプトを評価し、構造化された判定を返す（詳細 > **注意**: Agent フックは実験的機能であり、変更される可能性がある。 +> **注意**: v2.1.280 以降、agent フックは `PermissionRequest` では実行されない。その応答ではリクエストを許可も拒否もできないためである。`PermissionRequest` には command、http、mcp_tool、または prompt フックを使うこと。 + **主要プロパティ：** - `"type": "agent"` -- agent フックであることを示す - `"prompt"` -- サブエージェントへのタスク説明 @@ -1161,7 +1163,7 @@ open local-progress/index.html "hooks": [ { "type": "command", - "command": "${CLAUDE_PLUGIN_ROOT}/scripts/validate.sh" + "command": "\"${CLAUDE_PLUGIN_ROOT}\"/scripts/validate.sh" } ] } @@ -1382,10 +1384,11 @@ chmod +x ~/.claude/hooks/*.sh --- -**最終更新：** 2026 年 9 月 19 日 -**Claude Code バージョン：** 2.1.278 +**最終更新：** 2026 年 9 月 26 日 +**Claude Code バージョン：** 2.1.283 **情報源：** - https://code.claude.com/docs/en/hooks +- https://code.claude.com/docs/en/plugins/manifest-reference#quoting-and-path-separators - https://code.claude.com/docs/en/permissions - https://code.claude.com/docs/en/tools-reference#task-tool-availability - https://code.claude.com/docs/en/changelog diff --git a/ja/09-advanced-features/README.md b/ja/09-advanced-features/README.md index 409281b..1a3c27c 100644 --- a/ja/09-advanced-features/README.md +++ b/ja/09-advanced-features/README.md @@ -750,7 +750,7 @@ export CLAUDE_CODE_DISABLE_CRON=1 | `bypassPermissions` | 全操作、権限チェックなし（危険） | | `dontAsk` | 事前承認済みツールのみ実行、その他はすべて拒否 | -CLI では `Shift+Tab` でモードを切り替える。デフォルトは `--permission-mode` フラグまたは `permissions.defaultMode` 設定で指定する。 +CLI では `Shift+Tab` でモードを切り替える。デフォルトは `--permission-mode` フラグまたは `permissions.defaultMode` 設定で指定する。v2.1.283 以降、サードパーティプロバイダー（Bedrock、Google Cloud の Agent Platform、Foundry）上またはテレメトリをオフにした対話セッションは、パーミッションモードが未設定の場合 auto モードで開始する。`permissions.defaultMode` は引き続きこれより優先される。 ### 起動方法 @@ -1119,7 +1119,7 @@ Claude Code は効率向上のためキーボードショートカットをサ ``` **キー記法：** -- **修飾キー：** `ctrl`、`alt`（または `opt`）、`shift`、`meta`（または `cmd`） +- **修飾キー：** `ctrl`（または `control`）、`alt`（または `opt`、`option`、`meta`）、`shift`、`cmd`（または `command`、`super`、`win`）。`cmd` は `meta` の別名ではなく独立した修飾キーで、Super キーを報告するターミナルでのみ検出される - **大文字は Shift を含意：** `K` は `shift+k` と等価 - **特殊キー：** `escape`、`enter`、`return`、`tab`、`space`、`backspace`、`delete`、矢印キー @@ -2107,8 +2107,8 @@ Claude Code および関連機能の詳細： --- -**最終更新：** 2026 年 9 月 19 日 -**Claude Code バージョン：** 2.1.278 +**最終更新：** 2026 年 9 月 26 日 +**Claude Code バージョン：** 2.1.283 **情報源：** - https://code.claude.com/docs/en/permissions - https://code.claude.com/docs/en/auto-mode-classifier-billing @@ -2119,5 +2119,8 @@ Claude Code および関連機能の詳細： - https://www.anthropic.com/news/claude-opus-4-7 - https://github.com/anthropics/claude-code/releases/tag/v2.1.117 - https://github.com/anthropics/claude-code/releases/tag/v2.1.118 +- https://code.claude.com/docs/en/keybindings#modifiers +- https://code.claude.com/docs/en/permission-modes +- https://github.com/anthropics/claude-code/releases/tag/v2.1.283 **互換モデル：** Claude Sonnet 4.6、Claude Opus 4.7、Claude Haiku 4.5 diff --git a/ja/10-cli/README.md b/ja/10-cli/README.md index d6c2c3b..789b9cd 100644 --- a/ja/10-cli/README.md +++ b/ja/10-cli/README.md @@ -157,7 +157,7 @@ claude --model opusplan "design and implement the caching layer" | フラグ | 説明 | 例 | |------|------|-----| | `--system-prompt` | デフォルトプロンプト全体を置き換える | `claude --system-prompt "You are a Python expert"` | -| `--system-prompt-file` | ファイルからプロンプトを読み込む（プリントモード） | `claude -p --system-prompt-file ./prompt.txt "query"` | +| `--system-prompt-file` | ファイルからプロンプトを読み込み、デフォルトのプロンプトを置き換える | `claude -p --system-prompt-file ./prompt.txt "query"` | | `--append-system-prompt` | デフォルトプロンプトに追記 | `claude --append-system-prompt "Always use TypeScript"` | ### システムプロンプトの例 @@ -178,10 +178,10 @@ claude -p --system-prompt-file ./prompts/code-reviewer.txt "review main.py" | フラグ | 挙動 | 対話 | プリント | |------|-----|------|---------| | `--system-prompt` | デフォルトのシステムプロンプト全体を置き換える | ✅ | ✅ | -| `--system-prompt-file` | ファイルのプロンプトに置き換える | ❌ | ✅ | +| `--system-prompt-file` | ファイルのプロンプトに置き換える | ✅ | ✅ | | `--append-system-prompt` | デフォルトに追記する | ✅ | ✅ | -**`--system-prompt-file` はプリントモードでのみ使用する。対話モードでは `--system-prompt` または `--append-system-prompt` を使うこと。** +**すべてのシステムプロンプトフラグは対話モードとプリントモードの両方で動作する（ファイル形式は v2.1.69 以降）。`--system-prompt` と `--system-prompt-file` はデフォルトのプロンプトを置き換え、append 系のフラグはそれに追加する。** ## ツールと権限の管理 @@ -873,8 +873,8 @@ claude -p --output-format json "query" --- -**最終更新**: 2026 年 9 月 2 日 -**Claude Code バージョン**: 2.1.257 +**最終更新**: 2026 年 9 月 26 日 +**Claude Code バージョン**: 2.1.283 **出典**: - https://code.claude.com/docs/en/cli-reference - https://code.claude.com/docs/en/settings diff --git a/ja/CATALOG.md b/ja/CATALOG.md index cde75a4..5920000 100644 --- a/ja/CATALOG.md +++ b/ja/CATALOG.md @@ -45,7 +45,7 @@ | `/diff` | 対話式 diff ビューア | 変更レビュー | | `/config` | 設定を表示・編集 | 動作カスタマイズ | | `/status` | セッション状態を表示 | 現状確認 | -| `/agents` | 利用可能エージェントを表示 | 委譲オプション確認 | +| `/agents` | 削除済み：ウィザードは v2.1.198 で削除され、コマンドは v2.1.281 で `/` メニューと `/help` から外れた — サブエージェントの作成や管理は Claude に依頼するか、`.claude/agents/` を編集する | Claude に依頼してサブエージェントを管理 | | `/skills` | 利用可能スキルを表示 | 自動起動機能の確認 | | `/hooks` | 設定済みフックを表示 | 自動化のデバッグ | | `/insights` | セッションパターンを分析 | セッション最適化 | @@ -540,8 +540,8 @@ chmod +x ~/.claude/hooks/*.sh --- -**最終更新**：2026 年 9 月 19 日 -**Claude Code バージョン**：2.1.278 +**最終更新**：2026 年 9 月 26 日 +**Claude Code バージョン**：2.1.283 **情報源**： - https://code.claude.com/docs/en/tools-reference#task-tool-availability - https://code.claude.com/docs/en/overview diff --git a/ja/claude_concepts_guide.md b/ja/claude_concepts_guide.md index 0491077..cd4135e 100644 --- a/ja/claude_concepts_guide.md +++ b/ja/claude_concepts_guide.md @@ -2789,7 +2789,7 @@ graph TD ### 第 3 週：作業の分担 - 特定の役割を持つサブエージェントを最初に作る -- `/agents` コマンドを使う +- Claude に作成を依頼するか、`.claude/agents/.md` ファイルを追加する（`/agents` ウィザードは削除済み） - 簡単なタスクで委譲をテストする ### 第 4 週：すべてを自動化する @@ -3156,8 +3156,8 @@ Claude Code は推論努力を調整可能な 3 モデルをサポートする *収録機能：フック、チェックポイント、プラニングモード、拡張思考、バックグラウンドタスク、権限モード（6 種）、ヘッドレスモード、セッション管理、Auto Memory、Agent Teams、定期タスク、Chrome 連携、Channels、音声入力、同梱スキル* --- -**最終更新：** 2026 年 9 月 19 日 -**Claude Code バージョン：** 2.1.278 +**最終更新：** 2026 年 9 月 26 日 +**Claude Code バージョン：** 2.1.283 **出典：** - https://code.claude.com/docs/en/tools-reference#task-tool-availability - https://code.claude.com/docs/en/overview @@ -3165,5 +3165,6 @@ Claude Code は推論努力を調整可能な 3 モデルをサポートする - https://www.anthropic.com/news/claude-opus-4-7 - https://github.com/anthropics/claude-code/releases/tag/v2.1.117 - https://code.claude.com/docs/en/model-config +- https://code.claude.com/docs/en/sub-agents **対応モデル：** Claude Fable 5、Claude Opus 5、Claude Sonnet 5、Claude Sonnet 4.6、Claude Opus 4.8、Claude Haiku 4.5 diff --git a/uk/01-slash-commands/README.md b/uk/01-slash-commands/README.md index aef57f3..48720e7 100644 --- a/uk/01-slash-commands/README.md +++ b/uk/01-slash-commands/README.md @@ -27,7 +27,7 @@ | Команда | Призначення | |---------|------------| | `/add-dir ` | Додати робочий каталог | -| `/agents` | Управління конфігураціями агентів | +| `/agents` | Більше не робоча команда: інтерактивний майстер видалено у v2.1.198, а з v2.1.281 `/agents` не з'являється ні в меню `/`, ні в `/help` — введення команди лише пояснює, куди перемістився майстер. Попросіть Claude створити субагентів або керувати ними, або редагуйте файли в `.claude/agents/` напряму | | `/branch [name]` | Розгалужити розмову в нову сесію (аліас: `/fork`). Примітка: `/fork` перейменовано на `/branch` у v2.1.77 | | `/btw ` | Побічне запитання без додавання в історію | | `/chrome` | Налаштування інтеграції з Chrome | @@ -576,11 +576,12 @@ Deploy the application to production: --- -**Останнє оновлення**: 19 вересня 2026 -**Версія Claude Code**: 2.1.278 +**Останнє оновлення**: 26 вересня 2026 +**Версія Claude Code**: 2.1.283 **Джерела**: - https://code.claude.com/docs/en/commands - https://code.claude.com/docs/en/output-styles +- https://github.com/anthropics/claude-code/releases/tag/v2.1.281 **Сумісні моделі**: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 *Частина серії посібників [Claude How To](../)* diff --git a/uk/03-skills/README.md b/uk/03-skills/README.md index d254246..3881233 100644 --- a/uk/03-skills/README.md +++ b/uk/03-skills/README.md @@ -140,7 +140,7 @@ Show concrete examples of using this Skill. ### Рекомендовані поля - **description** (рекомендовано): що навичка робить І коли її використовувати. Якщо поле пропущено, Claude Code бере перший абзац markdown-вмісту. Об'єднаний текст `description` + `when_to_use` обрізається до **1 536 символів** у переліку навичок (налаштовується через `skillListingMaxDescChars`). Саме за цим текстом Claude визначає, коли активувати навичку. -- **name** (необов'язкове): за замовчуванням дорівнює **імені каталогу** навички. Якщо вказане, задає відображувану назву — тільки малі літери, цифри, дефіси (макс. 64 символи), не може містити "anthropic" або "claude". Для навичок у плагінах `name` також задає останній сегмент команди. +- **name** (необов'язкове): за замовчуванням дорівнює **імені каталогу** навички. Якщо вказане, задає відображувану назву — тільки малі літери, цифри, дефіси (макс. 64 символи). У Claude Code імена `anthropic-skills` і `synced` зарезервовані й не завантажуються поза плагіном; суворіша заборона на "anthropic" або "claude" стосується навичок, завантажених у claude.ai або Skills API. Для навичок у плагінах `name` також задає останній сегмент команди. Усі поля фронтматера SKILL.md є необов'язковими; рекомендованим є лише `description`. @@ -171,7 +171,7 @@ paths: "src/api/**/*.ts" # Glob-патерни, що обмежу | Поле | Опис | |------|------| -| `name` | Тільки малі літери, цифри, дефіси (макс. 64 символи). Не може містити "anthropic" або "claude". | +| `name` | Тільки малі літери, цифри, дефіси (макс. 64 символи). `anthropic-skills` і `synced` зарезервовані в Claude Code. | | `description` | Що навичка робить І коли її використовувати. Об'єднаний текст `description` + `when_to_use` обрізається до 1 536 символів у переліку навичок (налаштовується через `skillListingMaxDescChars`). Критично для автоматичного зіставлення. | | `when_to_use` | Додатковий контекст про те, коли Claude має викликати навичку. Додається до `description` у переліку навичок і враховується в ліміті 1 536 символів. | | `argument-hint` | Підказка в меню автодоповнення `/` (наприклад, `"[filename] [format]"`). | @@ -813,8 +813,8 @@ chmod +x ~/.claude/skills/my-skill/scripts/*.py - [Посібник хуків](../06-hooks/) — автоматизація за подіями --- -**Останнє оновлення**: 2 вересня 2026 -**Версія Claude Code**: 2.1.257 +**Останнє оновлення**: 26 вересня 2026 +**Версія Claude Code**: 2.1.283 **Джерела**: - https://code.claude.com/docs/en/skills **Сумісні моделі**: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 diff --git a/uk/03-skills/claude-md/SKILL.md b/uk/03-skills/claude-md/SKILL.md index 197d040..9e76074 100644 --- a/uk/03-skills/claude-md/SKILL.md +++ b/uk/03-skills/claude-md/SKILL.md @@ -193,7 +193,7 @@ For detailed build instructions, refer to `agent_docs/building_the_project.md` Якщо користувач запитує створення/оновлення AGENTS.md: -**Починаючи з v2.1.277, Claude Code читає `AGENTS.md` напряму як інструкції проєкту** — але лише тоді, коли ні в робочому каталозі, ні в жодному каталозі вище немає `CLAUDE.md`, `.claude/CLAUDE.md` чи `CLAUDE.local.md`. `~/.claude/CLAUDE.md`, керований CLAUDE.md і `.claude/rules/` не враховуються в цій перевірці та завантажуються паралельно. Поведінку визначає **Project instructions** у `/config`: `claude-md-or-agents-md` (за замовчуванням), `claude-md-and-agents-md`, `claude-md` або `managed-only`. Там, де пряме читання недоступне — Bedrock/Vertex/Foundry, вимкнена телеметрія, перша сесія після оновлення або `disableAllHooks`/`allowManagedHooksOnly` — використовуйте запасний варіант: імпортуйте файл з CLAUDE.md через `@AGENTS.md` або зробіть `CLAUDE.md` символічним посиланням на нього. +**Починаючи з v2.1.277, Claude Code читає `AGENTS.md` напряму як інструкції проєкту** — але лише тоді, коли ні в робочому каталозі, ні в жодному каталозі вище немає `CLAUDE.md`, `.claude/CLAUDE.md` чи `CLAUDE.local.md`. `~/.claude/CLAUDE.md`, керований CLAUDE.md і `.claude/rules/` не враховуються в цій перевірці та завантажуються паралельно. Поведінку визначає **Project instructions** у `/config`: `claude-md-or-agents-md` (за замовчуванням), `claude-md-and-agents-md`, `claude-md` або `managed-only`. Там, де пряме читання недоступне — версії до v2.1.277 (до v2.1.281 на Bedrock/Vertex/Foundry, за LLM-шлюзами або з вимкненою телеметрією) або перша сесія після оновлення — використовуйте запасний варіант: імпортуйте файл з CLAUDE.md через `@AGENTS.md` або зробіть `CLAUDE.md` символічним посиланням на нього. AGENTS.md — це міжінструментальний файл контексту проєкту, тобто документ тієї ж *категорії*, що й CLAUDE.md, а не формат визначення агентів. Він існує для того, щоб кілька кодових агентів могли користуватися спільним набором конвенцій проєкту: - Команди збірки, тестування та лінтингу diff --git a/uk/04-subagents/README.md b/uk/04-subagents/README.md index 45c3367..af9ac46 100644 --- a/uk/04-subagents/README.md +++ b/uk/04-subagents/README.md @@ -51,7 +51,7 @@ Кожен субагент працює незалежно з чистого аркуша, отримуючи лише конкретний контекст, необхідний для його завдання, а потім повертає результати головному агенту для синтезу. -**Швидкий старт**: Використовуйте команду `/agents` для створення, перегляду, редагування та управління субагентами в інтерактивному режимі. +**Швидкий старт**: Попросіть Claude створити субагента для вас («створи субагента, який перевіряє безпеку») або додайте файл `.claude/agents/.md` безпосередньо — див. [Управління субагентами](#управління-субагентами) нижче. --- @@ -291,16 +291,13 @@ Claude Code включає кілька вбудованих субагенті ### Команда `/agents` (рекомендовано) -```bash -/agents +> **Примітка:** Інтерактивний майстер `/agents` видалено у v2.1.198, а починаючи з v2.1.281 `/agents` більше не з'являється в меню `/` чи в `/help`. Попросіть Claude створити субагентів або керувати ними (наприклад, «створи субагента, який перевіряє безпеку») або додайте файл `.claude/agents/.md` безпосередньо. + +```text +Створи субагента, який перевіряє код на вразливості безпеки. ``` -Надає інтерактивне меню для: -- Перегляду всіх доступних субагентів (вбудованих, користувача та проєкту) -- Створення нових субагентів з керованим налаштуванням -- Редагування наявних субагентів та доступу до інструментів -- Видалення субагентів -- Перегляду активних субагентів при наявності дублікатів +Claude запише файл `.claude/agents/.md` за вас і підбере доречний frontmatter (інструменти, модель, опис). Потім ви можете доопрацювати файл вручну або попросити Claude його змінити. ### Пряме управління файлами @@ -485,7 +482,7 @@ description: Performs long-running analysis tasks in the background | Скорочення | Дія | |-----------|-----| | `Ctrl+B` | Перемістити поточне завдання субагента у фон | -| `Ctrl+F` | Завершити всіх фонових агентів (натисніть двічі для підтвердження) | +| `Ctrl+X` `Ctrl+K` | Зупинити всіх запущених фонових субагентів (натисніть двічі протягом 3 секунд для підтвердження; це замінило `Ctrl+F`) | ### Вимкнення фонових завдань @@ -1017,16 +1014,14 @@ graph TB ### Метод 1: Команда /agents (рекомендовано) -```bash -/agents +> **Примітка:** Інтерактивний майстер `/agents` видалено у v2.1.198, а починаючи з v2.1.281 `/agents` більше не з'являється в меню `/` чи в `/help`. Попросіть Claude створити субагентів або керувати ними (наприклад, «створи субагента, який перевіряє безпеку») або додайте файл `.claude/agents/.md` безпосередньо. + +```text +Створи субагента рівня проєкту, який запускає тести та виправляє збої. +Надай йому доступ до Bash, Read, Edit і Grep. ``` -Далі: -1. Оберіть 'Create New Agent' -2. Оберіть рівень проєкту або користувача -3. Опишіть субагента детально -4. Оберіть інструменти (або залиште порожнім для успадкування всіх) -5. Збережіть і використовуйте +Claude запише `.claude/agents/.md` з відповідним frontmatter. Перегляньте згенерований файл, а потім використовуйте його. ### Метод 2: Копіювання в проєкт @@ -1062,13 +1057,13 @@ cp /path/to/04-subagents/debugger.md ~/.claude/agents/ ### Перевірка -Після встановлення переконайтесь, що агенти розпізнаються: +Після встановлення переконайтесь, що агенти розпізнаються, переглянувши вміст каталогу: ```bash -/agents +ls .claude/agents/ ``` -Ви повинні побачити встановлених агентів поряд з вбудованими. +Ви також можете запитати Claude, які субагенти доступні в поточній сесії, — він назве вбудованих і власних агентів, яким може делегувати завдання. --- @@ -1140,8 +1135,10 @@ graph TD - [Посібник хуків](../06-hooks/) — для автоматизації за подіями --- -**Останнє оновлення**: 25 серпня 2026 -**Версія Claude Code**: 2.1.245 +**Останнє оновлення**: 26 вересня 2026 +**Версія Claude Code**: 2.1.283 **Джерела**: - https://code.claude.com/docs/en/sub-agents +- https://code.claude.com/docs/en/commands +- https://code.claude.com/docs/en/interactive-mode **Сумісні моделі**: Claude Sonnet 4.6, Claude Opus 4.6, Claude Haiku 4.5 diff --git a/uk/05-mcp/README.md b/uk/05-mcp/README.md index a4d671a..fdbc5cb 100644 --- a/uk/05-mcp/README.md +++ b/uk/05-mcp/README.md @@ -220,16 +220,17 @@ sequenceDiagram ## Пошук інструментів MCP -Коли описи інструментів MCP перевищують 10% контекстного вікна, Claude Code автоматично вмикає пошук інструментів для ефективного вибору правильних інструментів без перевантаження контексту моделі. +Пошук інструментів увімкнено за замовчуванням: інструменти MCP відкладаються й знаходяться за потреби, тож їхні визначення не заповнюють контекст моделі. Claude Code вимикає його, коли `ANTHROPIC_BASE_URL` вказує на хост, що не належить Anthropic (не first-party), оскільки більшість проксі не пересилають потрібні йому блоки. | Налаштування | Значення | Опис | |-------------|----------|------| -| `ENABLE_TOOL_SEARCH` | `auto` (за замовч.) | Автоматично вмикається, коли описи інструментів перевищують 10% контексту | -| `ENABLE_TOOL_SEARCH` | `auto:` | Автоматично вмикається при користувацькому порозі `N` інструментів | -| `ENABLE_TOOL_SEARCH` | `true` | Завжди увімкнено незалежно від кількості інструментів | +| `ENABLE_TOOL_SEARCH` | (не задано — за замовч.) | Усі інструменти MCP відкладаються й завантажуються за потреби | +| `ENABLE_TOOL_SEARCH` | `auto` | Пороговий режим: інструменти завантажуються одразу, доки їхні визначення не досягнуть 10% контекстного вікна, після чого всі відкладаються | +| `ENABLE_TOOL_SEARCH` | `auto:` | Пороговий режим із власним відсотком (0-100) — наприклад, `auto:5` для 5% | +| `ENABLE_TOOL_SEARCH` | `true` | Усі інструменти MCP відкладаються | | `ENABLE_TOOL_SEARCH` | `false` | Вимкнено; всі описи інструментів надсилаються повністю | -> **Примітка:** Пошук інструментів потребує Sonnet 4 або новіший, або Opus 4 або новіший. Моделі Haiku не підтримують пошук інструментів. +> **Примітка:** Пошук інструментів потребує моделі з підтримкою блоків `tool_reference`: Claude Sonnet 4.5, Claude Haiku 4.5, Claude Opus 4.5 та новіших моделей. ## Динамічне оновлення інструментів @@ -1138,9 +1139,10 @@ export GITHUB_TOKEN="your_token" - [Документація Claude API](https://docs.anthropic.com) --- -**Останнє оновлення**: 6 вересня 2026 -**Версія Claude Code**: 2.1.263 +**Останнє оновлення**: 26 вересня 2026 +**Версія Claude Code**: 2.1.283 **Джерела**: - https://code.claude.com/docs/en/mcp +- https://code.claude.com/docs/en/mcp#configure-tool-search - https://code.claude.com/docs/en/managed-mcp **Сумісні моделі**: Claude Sonnet 4.6, Claude Opus 4.6, Claude Haiku 4.5 diff --git a/uk/06-hooks/README.md b/uk/06-hooks/README.md index a5b216d..6d35577 100644 --- a/uk/06-hooks/README.md +++ b/uk/06-hooks/README.md @@ -177,6 +177,8 @@ LLM оцінює промпт і повертає структуроване р > **Примітка**: Agent-хуки є експериментальними і можуть змінитися. +> **Примітка**: Починаючи з v2.1.280, agent-хуки не запускаються на `PermissionRequest` — їхня відповідь ніколи не могла дозволити чи відхилити запит. Використовуйте там хук типу command, http, mcp_tool або prompt. + **Ключові властивості:** - `"type": "agent"` — ідентифікує як agent-хук - `"prompt"` — опис завдання для субагента @@ -999,7 +1001,7 @@ python3 09-advanced-features/setup-auto-mode-permissions.py "hooks": [ { "type": "command", - "command": "${CLAUDE_PLUGIN_ROOT}/scripts/validate.sh" + "command": "\"${CLAUDE_PLUGIN_ROOT}\"/scripts/validate.sh" } ] } @@ -1218,10 +1220,11 @@ chmod +x ~/.claude/hooks/*.sh - **[Посібник з памʼяті](../02-memory/)** — конфігурація постійного контексту --- -**Останнє оновлення**: 19 вересня 2026 -**Версія Claude Code**: 2.1.278 +**Останнє оновлення**: 26 вересня 2026 +**Версія Claude Code**: 2.1.283 **Джерела**: - https://code.claude.com/docs/en/hooks +- https://code.claude.com/docs/en/plugins/manifest-reference#quoting-and-path-separators - https://code.claude.com/docs/en/permissions - https://code.claude.com/docs/en/tools-reference#task-tool-availability **Сумісні моделі**: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 diff --git a/uk/09-advanced-features/README.md b/uk/09-advanced-features/README.md index 5c0f481..7a28f3a 100644 --- a/uk/09-advanced-features/README.md +++ b/uk/09-advanced-features/README.md @@ -674,7 +674,7 @@ export CLAUDE_CODE_DISABLE_CRON=1 | `bypassPermissions` | Усі дії, без перевірки дозволів (небезпечно) | | `dontAsk` | Виконуються лише попередньо затверджені інструменти; всі інші відхиляються | -Перемикайте режими через `Shift+Tab` у CLI. Встановіть режим за замовчуванням через прапорець `--permission-mode` або налаштування `permissions.defaultMode`. +Перемикайте режими через `Shift+Tab` у CLI. Встановіть режим за замовчуванням через прапорець `--permission-mode` або налаштування `permissions.defaultMode`. Починаючи з v2.1.283, інтерактивна сесія у стороннього провайдера (Bedrock, Agent Platform від Google Cloud, Foundry) або з вимкненою телеметрією запускається в режимі auto, якщо режим дозволів не налаштовано; `permissions.defaultMode` і далі має пріоритет. ### Методи активації @@ -1026,7 +1026,7 @@ Claude Code підтримує клавіатурні скорочення дл ``` **Синтаксис клавіш**: -- **Модифікатори**: `ctrl`, `alt` (або `opt`), `shift`, `meta` (або `cmd`) +- **Модифікатори**: `ctrl` (або `control`), `alt` (або `opt`, `option`, `meta`), `shift` і `cmd` (або `command`, `super`, `win`) — `cmd` є окремим модифікатором, а не псевдонімом `meta`, і розпізнається лише в терміналах, які передають клавішу Super - **Великі літери включають Shift**: `K` еквівалентно `shift+k` - **Спеціальні клавіші**: `escape`, `enter`, `return`, `tab`, `space`, `backspace`, `delete`, клавіші стрілок @@ -1932,10 +1932,13 @@ claude --teammate-mode in-process - [Офіційна документація Agent Teams](https://code.claude.com/docs/en/agent-teams) --- -**Останнє оновлення**: 19 вересня 2026 -**Версія Claude Code**: 2.1.278 +**Останнє оновлення**: 26 вересня 2026 +**Версія Claude Code**: 2.1.283 **Джерела**: - https://code.claude.com/docs/en/permissions - https://code.claude.com/docs/en/auto-mode-classifier-billing - https://code.claude.com/docs/en/commands +- https://code.claude.com/docs/en/keybindings#modifiers +- https://code.claude.com/docs/en/permission-modes +- https://github.com/anthropics/claude-code/releases/tag/v2.1.283 **Сумісні моделі**: Claude Sonnet 4.6, Claude Opus 4.6, Claude Haiku 4.5 diff --git a/uk/10-cli/README.md b/uk/10-cli/README.md index ce760ce..72387e0 100644 --- a/uk/10-cli/README.md +++ b/uk/10-cli/README.md @@ -141,7 +141,7 @@ claude --model opusplan "design and implement the caching layer" | Прапорець | Опис | Приклад | |-----------|------|---------| | `--system-prompt` | Замінити весь стандартний промпт | `claude --system-prompt "You are a Python expert"` | -| `--system-prompt-file` | Завантажити промпт з файлу (print mode) | `claude -p --system-prompt-file ./prompt.txt "query"` | +| `--system-prompt-file` | Завантажити промпт з файлу, замінивши промпт за замовчуванням | `claude -p --system-prompt-file ./prompt.txt "query"` | | `--append-system-prompt` | Додати до стандартного промпта | `claude --append-system-prompt "Always use TypeScript"` | ### Приклади системного промпта @@ -162,10 +162,10 @@ claude -p --system-prompt-file ./prompts/code-reviewer.txt "review main.py" | Прапорець | Поведінка | Інтерактивний | Print | |-----------|----------|---------------|-------| | `--system-prompt` | Замінює весь стандартний системний промпт | ✅ | ✅ | -| `--system-prompt-file` | Замінює промптом з файлу | ❌ | ✅ | +| `--system-prompt-file` | Замінює промптом з файлу | ✅ | ✅ | | `--append-system-prompt` | Додає до стандартного системного промпта | ✅ | ✅ | -**Використовуйте `--system-prompt-file` лише в print-режимі. Для інтерактивного режиму використовуйте `--system-prompt` або `--append-system-prompt`.** +**Усі прапорці системного промпту працюють як в інтерактивному, так і в print-режимі (файлові форми — з v2.1.69). `--system-prompt` і `--system-prompt-file` замінюють промпт за замовчуванням; прапорці append додають до нього.** ## Управління інструментами та дозволами @@ -837,8 +837,8 @@ claude -p --output-format json "query" *Частина серії посібників [Claude How To](../)* --- -**Останнє оновлення**: 2 вересня 2026 -**Версія Claude Code**: 2.1.257 +**Останнє оновлення**: 26 вересня 2026 +**Версія Claude Code**: 2.1.283 **Джерела**: - https://code.claude.com/docs/en/cli-reference **Сумісні моделі**: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 diff --git a/uk/CATALOG.md b/uk/CATALOG.md index f35dfb0..3c78566 100644 --- a/uk/CATALOG.md +++ b/uk/CATALOG.md @@ -45,7 +45,7 @@ | `/diff` | Інтерактивний перегляд змін | Перегляд змін | | `/config` | Переглянути/змінити конфігурацію | Налаштування поведінки | | `/status` | Показати статус сесії | Перевірка поточного стану | -| `/agents` | Список доступних агентів | Перегляд варіантів делегування | +| `/agents` | Видалено: майстер прибрали у v2.1.198, а команда зникла з меню `/` і `/help` у v2.1.281 — попросіть Claude створити або налаштувати субагентів чи відредагуйте `.claude/agents/` | Керування субагентами через запит до Claude | | `/skills` | Список доступних навичок | Перегляд автовикликів | | `/hooks` | Список налаштованих хуків | Налагодження автоматизації | | `/insights` | Аналіз патернів сесії | Оптимізація сесії | @@ -532,8 +532,8 @@ chmod +x ~/.claude/hooks/*.sh --- -**Останнє оновлення**: 19 вересня 2026 -**Версія Claude Code**: 2.1.278 +**Останнє оновлення**: 26 вересня 2026 +**Версія Claude Code**: 2.1.283 **Джерела**: - https://code.claude.com/docs/en/tools-reference#task-tool-availability - https://code.claude.com/docs/en/commands diff --git a/uk/claude_concepts_guide.md b/uk/claude_concepts_guide.md index 46b4489..a51ec0f 100644 --- a/uk/claude_concepts_guide.md +++ b/uk/claude_concepts_guide.md @@ -2784,7 +2784,7 @@ graph TD ### Тиждень 3: Розподіліть роботу - Створіть першого субагента для конкретної ролі -- Використовуйте команду `/agents` +- Попросіть Claude створити його або додайте файл `.claude/agents/.md` (майстер `/agents` видалено) - Протестуйте делегування з простим завданням ### Тиждень 4: Автоматизуйте все @@ -3137,8 +3137,9 @@ User: \ *Тепер включає: хуки, контрольні точки, режим планування, розширене мислення, фонові завдання, режими дозволів (6 режимів), headless-режим, управління сесіями, авто-пам'ять, команди агентів, заплановані завдання, інтеграцію з Chrome, канали, голосовий ввід та комплектні навички* --- -**Останнє оновлення**: 19 вересня 2026 -**Версія Claude Code**: 2.1.278 +**Останнє оновлення**: 26 вересня 2026 +**Версія Claude Code**: 2.1.283 **Джерела**: - https://code.claude.com/docs/en/tools-reference#task-tool-availability - https://code.claude.com/docs/en/hooks +- https://code.claude.com/docs/en/sub-agents diff --git a/vi/01-slash-commands/README.md b/vi/01-slash-commands/README.md index 22a0145..e41c043 100644 --- a/vi/01-slash-commands/README.md +++ b/vi/01-slash-commands/README.md @@ -23,7 +23,7 @@ Lệnh tích hợp sẵn là các lối tắt cho các hành động phổ biế | Lệnh | Mục Đích | |---------|---------| | `/add-dir ` | Thêm thư mục làm việc | -| `/agents` | Quản lý cấu hình tác nhân | +| `/agents` | Không còn là lệnh hoạt động: trình hướng dẫn tương tác đã bị gỡ bỏ ở v2.1.198, và từ v2.1.281 `/agents` không còn xuất hiện trong menu `/` hay `/help` — gõ lệnh này chỉ giải thích trình hướng dẫn đã được chuyển đi đâu. Hãy yêu cầu Claude tạo hoặc quản lý subagent, hoặc chỉnh sửa trực tiếp các tệp trong `.claude/agents/` | | `/branch [name]` | Chuyển sang một bản sao của cuộc hội thoại tại thời điểm này, giữ nguyên bản gốc (quay lại bằng `/resume`) | | `/fork [prompt]` | Sao chép cuộc hội thoại hiện tại thành một **phiên nền** mới và bạn vẫn tiếp tục ở đây; từ thời điểm đó hai phiên độc lập nhau và bản sao có dòng riêng trong `claude agents` (v2.1.212+) | | `/subtask ` | Spawn một **subagent được fork** kế thừa toàn bộ cuộc hội thoại và làm việc trên tác vụ trong khi bạn tiếp tục; kết quả trả về cuộc hội thoại này khi hoàn thành (v2.1.212+) | @@ -557,11 +557,12 @@ Nếu cả hai tồn tại với cùng tên, **skill sẽ được ưu tiên**. --- -**Cập Nhật Lần Cuối**: Ngày 19 tháng 9 năm 2026 -**Phiên Bản Claude Code**: 2.1.278 +**Cập Nhật Lần Cuối**: Ngày 26 tháng 9 năm 2026 +**Phiên Bản Claude Code**: 2.1.283 **Nguồn**: - https://code.claude.com/docs/en/commands - https://code.claude.com/docs/en/output-styles +- https://github.com/anthropics/claude-code/releases/tag/v2.1.281 **Các Mô Hình Tương Thích**: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 *Phần của series hướng dẫn [Claude How To](../)* diff --git a/vi/03-skills/README.md b/vi/03-skills/README.md index ae05a91..44b2051 100644 --- a/vi/03-skills/README.md +++ b/vi/03-skills/README.md @@ -136,7 +136,7 @@ Hiển thị các ví dụ cụ thể về việc sử dụng Skill này. ### Các Trường Khuyến Nghị - **description** (khuyến nghị): những gì Skill làm VÀ khi nào sử dụng nó. Nếu bỏ trống, Claude Code dùng đoạn văn đầu tiên của nội dung markdown. Nội dung `description` + `when_to_use` gộp lại bị cắt ở **1.536 ký tự** trong danh sách skill (có thể điều chỉnh qua `skillListingMaxDescChars`). Đây là nội dung Claude đối chiếu để quyết định khi nào kích hoạt skill. -- **name** (tùy chọn): mặc định lấy theo **tên thư mục** của skill. Khi được khai báo, nó đặt tên hiển thị — chỉ chữ thường, số, gạch ngang (tối đa 64 ký tự), và không thể chứa "anthropic" hoặc "claude". Với skill trong plugin, `name` cũng đặt phân đoạn cuối của lệnh. +- **name** (tùy chọn): mặc định lấy theo **tên thư mục** của skill. Khi được khai báo, nó đặt tên hiển thị — chỉ chữ thường, số, gạch ngang (tối đa 64 ký tự). Trong Claude Code, `anthropic-skills` và `synced` là tên dành riêng và không được tải bên ngoài plugin; quy tắc chặt hơn cấm "anthropic" hoặc "claude" áp dụng cho các skill được tải lên claude.ai hoặc Skills API. Với skill trong plugin, `name` cũng đặt phân đoạn cuối của lệnh. Mọi trường frontmatter trong SKILL.md đều là tùy chọn; `description` là trường duy nhất được khuyến nghị. @@ -166,7 +166,7 @@ hooks: # Hooks theo phạm vi skill | Trường | Mô Tả | |-------|-------------| -| `name` | Chỉ chữ thường, số, gạch ngang (tối đa 64 ký tự). Không thể chứa "anthropic" hoặc "claude". | +| `name` | Chỉ chữ thường, số, gạch ngang (tối đa 64 ký tự). `anthropic-skills` và `synced` là tên dành riêng trong Claude Code. | | `description` | Những gì Skill làm VÀ khi nào sử dụng nó. Nội dung `description` + `when_to_use` gộp lại bị cắt ở 1.536 ký tự trong danh sách skill (có thể điều chỉnh qua `skillListingMaxDescChars`). Quan trọng cho phù hợp kích hoạt tự động. | | `when_to_use` | Bối cảnh bổ sung về thời điểm Claude nên gọi skill. Được nối vào `description` trong danh sách skill và tính vào giới hạn 1.536 ký tự. | | `argument-hint` | Gợi ý hiển thị trong menu autocomplete `/` (ví dụ: `"[filename] [format]"`). | @@ -808,8 +808,8 @@ Khi bạn bắt đầu xây dựng skills một cách nghiêm túc, hai điều --- -**Cập Nhật Lần Cuối**: Ngày 2 tháng 9 năm 2026 -**Phiên Bản Claude Code**: 2.1.257 +**Cập Nhật Lần Cuối**: Ngày 26 tháng 9 năm 2026 +**Phiên Bản Claude Code**: 2.1.283 **Nguồn**: - https://code.claude.com/docs/en/skills **Các Mô Hình Tương Thích**: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 diff --git a/vi/03-skills/claude-md/SKILL.md b/vi/03-skills/claude-md/SKILL.md index 2c74c5e..0bc1af3 100644 --- a/vi/03-skills/claude-md/SKILL.md +++ b/vi/03-skills/claude-md/SKILL.md @@ -193,7 +193,7 @@ Trước khi hoàn thành, xác minh: Nếu người dùng yêu cầu tạo/cập nhật AGENTS.md: -**Từ v2.1.277, Claude Code đọc `AGENTS.md` trực tiếp như hướng dẫn dự án** — nhưng chỉ khi thư mục làm việc và mọi thư mục phía trên nó đều không có `CLAUDE.md`, `.claude/CLAUDE.md`, hay `CLAUDE.local.md`. `~/.claude/CLAUDE.md`, CLAUDE.md được quản lý, và `.claude/rules/` không tính vào kiểm tra đó và vẫn được tải song song. Hành vi này được điều khiển bởi **Project instructions** trong `/config`: `claude-md-or-agents-md` (mặc định), `claude-md-and-agents-md`, `claude-md`, hoặc `managed-only`. Ở nơi không hỗ trợ đọc trực tiếp — Bedrock/Vertex/Foundry, telemetry bị tắt, phiên đầu tiên sau khi nâng cấp, hoặc `disableAllHooks`/`allowManagedHooksOnly` — hãy dùng cách dự phòng: import từ CLAUDE.md bằng `@AGENTS.md`, hoặc symlink `CLAUDE.md` tới nó. +**Từ v2.1.277, Claude Code đọc `AGENTS.md` trực tiếp như hướng dẫn dự án** — nhưng chỉ khi thư mục làm việc và mọi thư mục phía trên nó đều không có `CLAUDE.md`, `.claude/CLAUDE.md`, hay `CLAUDE.local.md`. `~/.claude/CLAUDE.md`, CLAUDE.md được quản lý, và `.claude/rules/` không tính vào kiểm tra đó và vẫn được tải song song. Hành vi này được điều khiển bởi **Project instructions** trong `/config`: `claude-md-or-agents-md` (mặc định), `claude-md-and-agents-md`, `claude-md`, hoặc `managed-only`. Ở nơi không hỗ trợ đọc trực tiếp — phiên bản trước v2.1.277 (trước v2.1.281 trên Bedrock/Vertex/Foundry, LLM gateway, hoặc khi telemetry bị tắt), hoặc phiên đầu tiên sau khi nâng cấp — hãy dùng cách dự phòng: import từ CLAUDE.md bằng `@AGENTS.md`, hoặc symlink `CLAUDE.md` tới nó. AGENTS.md là file bối cảnh dự án dùng chung giữa nhiều công cụ — cùng *loại* tài liệu với CLAUDE.md, không phải định dạng định nghĩa tác nhân. Nó tồn tại để nhiều coding agent có thể dùng chung một bộ quy ước dự án: - Các lệnh build, test, và lint diff --git a/vi/04-subagents/README.md b/vi/04-subagents/README.md index c1ba1f9..ff7885f 100644 --- a/vi/04-subagents/README.md +++ b/vi/04-subagents/README.md @@ -47,7 +47,7 @@ Tác nhân con cho phép thực thi tác vụ được ủy quyền trong Claude Mỗi tác nhân con hoạt động độc lập với một trang sạch, chỉ nhận bối cảnh cụ thể cần thiết cho tác vụ của họ, sau đó trả về kết quả cho tác nhân chính để tổng hợp. -**Bắt Đầu Nhanh**: Sử dụng lệnh `/agents` để tạo, xem, chỉnh sửa, và quản lý các tác nhân con của bạn một cách tương tác. +**Bắt Đầu Nhanh**: Hãy yêu cầu Claude tạo tác nhân con cho bạn ("tạo một tác nhân con để review bảo mật"), hoặc thêm trực tiếp một file `.claude/agents/.md` — xem [Quản Lý Tác Nhân Con](#quản-lý-tác-nhân-con) bên dưới. --- @@ -286,16 +286,13 @@ Claude Code bao gồm một số tác nhân con được tích hợp sẵn luôn ### Sử Dụng Lệnh `/agents` (Khuyến Nghị) -```bash -/agents +> **Lưu ý**: Trình hướng dẫn tương tác `/agents` đã bị gỡ bỏ trong v2.1.198, và kể từ v2.1.281, `/agents` không còn xuất hiện trong menu `/` hoặc `/help`. Hãy yêu cầu Claude tạo hoặc quản lý tác nhân con (ví dụ: "tạo một tác nhân con để review bảo mật"), hoặc thêm trực tiếp một file `.claude/agents/.md`. + +```text +Tạo một tác nhân con để review code tìm lỗ hổng bảo mật. ``` -Điều này cung cấp menu tương tác để: -- Xem tất cả các tác nhân con có sẵn (tích hợp, người dùng, và dự án) -- Tạo các tác nhân con mới với thiết lập có hướng dẫn -- Chỉnh sửa các tác nhân con tùy chỉnh hiện có và truy cập công cụ -- Xóa các tác nhân con tùy chỉnh -- Xem các tác nhân con nào đang hoạt động khi có bản sao trùng lặp +Claude sẽ viết file `.claude/agents/.md` cho bạn, chọn frontmatter phù hợp (công cụ, model, mô tả). Sau đó bạn có thể chỉnh sửa file thủ công hoặc yêu cầu Claude điều chỉnh. ### Quản Lý File Trực Tiếp @@ -480,7 +477,7 @@ description: Performs long-running analysis tasks in the background | Phím Tắt | Hành Động | |----------|--------| | `Ctrl+B` | Chạy nền một tác vụ tác nhân con đang chạy | -| `Ctrl+F` | Giết tất cả các tác nhân nền (nhấn hai lần để xác nhận) | +| `Ctrl+X` `Ctrl+K` | Dừng tất cả các tác nhân con nền đang chạy (nhấn hai lần trong vòng 3 giây để xác nhận; tổ hợp này thay thế `Ctrl+F`) | ### Vô Hiệu hóa Tác Vụ Nền @@ -1016,16 +1013,14 @@ Thư mục này chứa các ví dụ tác nhân con đã sẵn sàng sử dụng ### Phương Pháp 1: Sử Dụng Lệnh /agents (Khuyến Nghị) -```bash -/agents +> **Lưu ý**: Trình hướng dẫn tương tác `/agents` đã bị gỡ bỏ trong v2.1.198, và kể từ v2.1.281, `/agents` không còn xuất hiện trong menu `/` hoặc `/help`. Hãy yêu cầu Claude tạo hoặc quản lý tác nhân con (ví dụ: "tạo một tác nhân con để review bảo mật"), hoặc thêm trực tiếp một file `.claude/agents/.md`. + +```text +Tạo một tác nhân con cấp dự án để chạy test và sửa các lỗi thất bại. +Cấp cho nó quyền truy cập Bash, Read, Edit và Grep. ``` -Sau đó: -1. Chọn 'Create New Agent' -2. Chọn cấp dự án hoặc cấp người dùng -3. Mô tả tác nhân con của bạn chi tiết -4. Chọn các công cụ để cấp quyền truy cập (hoặc bỏ trống để kế thừa tất cả) -5. Lưu và sử dụng +Claude sẽ viết `.claude/agents/.md` với frontmatter phù hợp. Hãy xem lại file được tạo, rồi sử dụng. ### Phương Pháp 2: Sao Chép Vào Dự Án @@ -1061,13 +1056,13 @@ cp /path/to/04-subagents/debugger.md ~/.claude/agents/ ### Xác Minh -Sau khi cài đặt, xác minh các tác nhân được nhận diện: +Sau khi cài đặt, xác minh các tác nhân được nhận diện bằng cách liệt kê thư mục: ```bash -/agents +ls .claude/agents/ ``` -Bạn sẽ thấy các tác nhân đã cài của bạn được liệt kê cùng với các tác nhân được tích hợp sẵn. +Bạn cũng có thể hỏi Claude những tác nhân con nào có sẵn trong phiên hiện tại, và Claude sẽ báo cáo các tác nhân tích hợp sẵn và tùy chỉnh mà nó có thể ủy quyền. --- @@ -1138,8 +1133,10 @@ graph TD --- -**Cập Nhật Lần Cuối**: Ngày 25 tháng 8 năm 2026 -**Phiên Bản Claude Code**: 2.1.245 +**Cập Nhật Lần Cuối**: Ngày 26 tháng 9 năm 2026 +**Phiên Bản Claude Code**: 2.1.283 **Nguồn**: - https://code.claude.com/docs/en/sub-agents +- https://code.claude.com/docs/en/commands +- https://code.claude.com/docs/en/interactive-mode **Các Mô Hình Tương Thích**: Claude Sonnet 4.6, Claude Opus 4.6, Claude Haiku 4.5 diff --git a/vi/05-mcp/README.md b/vi/05-mcp/README.md index 9136d3a..1e6821b 100644 --- a/vi/05-mcp/README.md +++ b/vi/05-mcp/README.md @@ -216,16 +216,17 @@ sequenceDiagram ## Tìm Kiếm Công Cụ MCP / MCP Tool Search -Khi mô tả công cụ MCP vượt quá 10% cửa sổ ngữ cảnh, Claude Code tự động kích hoạt tìm kiếm công cụ để chọn công cụ phù hợp một cách hiệu quả mà không làm bội ngữ cảnh mô hình. +Tìm kiếm công cụ được bật theo mặc định: các công cụ MCP được hoãn lại và khám phá khi cần, nên định nghĩa của chúng không lấp đầy ngữ cảnh mô hình. Claude Code tắt tính năng này khi `ANTHROPIC_BASE_URL` trỏ tới một host không phải của bên thứ nhất, vì hầu hết các proxy không chuyển tiếp các khối mà nó cần. | Cài Đặt | Giá Trị | Mô Tả | |---------|-------|-------------| -| `ENABLE_TOOL_SEARCH` | `auto` (mặc định) | Tự động kích hoạt khi mô tả công cụ vượt quá 10% ngữ cảnh | -| `ENABLE_TOOL_SEARCH` | `auto:` | Tự động kích hoạt tại ngưỡng tùy chỉnh của `N` công cụ | -| `ENABLE_TOOL_SEARCH` | `true` | Luôn được kích hoạt bất kể số lượng công cụ | +| `ENABLE_TOOL_SEARCH` | (không đặt — mặc định) | Tất cả công cụ MCP được hoãn lại và tải khi cần | +| `ENABLE_TOOL_SEARCH` | `auto` | Chế độ ngưỡng: công cụ được tải trước cho đến khi định nghĩa của chúng đạt 10% cửa sổ ngữ cảnh, sau đó tất cả được hoãn lại | +| `ENABLE_TOOL_SEARCH` | `auto:` | Chế độ ngưỡng với tỷ lệ phần trăm tùy chỉnh (0-100) — ví dụ, `auto:5` cho 5% | +| `ENABLE_TOOL_SEARCH` | `true` | Tất cả công cụ MCP được hoãn lại | | `ENABLE_TOOL_SEARCH` | `false` | Đã tắt; tất cả mô tả công cụ được gửi đầy đủ | -> **Lưu Ý:** Tìm kiếm công cụ yêu cầu Sonnet 4 hoặc mới hơn, hoặc Opus 4 hoặc mới hơn. Các mô hình Haiku không được hỗ trợ cho tìm kiếm công cụ. +> **Lưu Ý:** Tìm kiếm công cụ yêu cầu một mô hình hỗ trợ các khối `tool_reference`: Claude Sonnet 4.5, Claude Haiku 4.5, Claude Opus 4.5, và các mô hình mới hơn. ## Cập Nhật Công Cụ Động / Dynamic Tool Updates @@ -1157,9 +1158,10 @@ export GITHUB_TOKEN="your_token" --- -**Cập Nhật Lần Cuối**: Ngày 6 tháng 9 năm 2026 -**Phiên Bản Claude Code**: 2.1.263 +**Cập Nhật Lần Cuối**: Ngày 26 tháng 9 năm 2026 +**Phiên Bản Claude Code**: 2.1.283 **Nguồn**: - https://code.claude.com/docs/en/mcp - https://code.claude.com/docs/en/managed-mcp +- https://code.claude.com/docs/en/mcp#configure-tool-search **Các Mô Hình Tương Thích**: Claude Sonnet 4.6, Claude Opus 4.6, Claude Haiku 4.5 diff --git a/vi/06-hooks/README.md b/vi/06-hooks/README.md index 2cfaba1..2878b43 100644 --- a/vi/06-hooks/README.md +++ b/vi/06-hooks/README.md @@ -165,6 +165,8 @@ Hooks xác thực dựa trên subagent mà spawn một agent chuyên dụng đ > **Lưu ý**: Agent hooks là tính năng thử nghiệm và có thể thay đổi. +> **Lưu ý**: Kể từ v2.1.280, agent hooks không chạy trên `PermissionRequest` — câu trả lời của chúng không bao giờ có thể cho phép hoặc từ chối yêu cầu. Hãy dùng hook command, http, mcp_tool, hoặc prompt ở đó. + **Các thuộc tính chính:** - `"type": "agent"` -- xác định đây là một agent hook - `"prompt"` -- mô tả nhiệm vụ cho subagent @@ -756,7 +758,7 @@ Plugins có thể bao gồm hooks trong file `hooks/hooks.json` của chúng: "hooks": [ { "type": "command", - "command": "${CLAUDE_PLUGIN_ROOT}/scripts/validate.sh" + "command": "\"${CLAUDE_PLUGIN_ROOT}\"/scripts/validate.sh" } ] } @@ -878,9 +880,10 @@ Chỉnh sửa `~/.claude/settings.json` hoặc `.claude/settings.json` với c --- -**Cập Nhật Lần Cuối**: Ngày 19 tháng 9 năm 2026 -**Phiên Bản Claude Code**: 2.1.278 +**Cập Nhật Lần Cuối**: Ngày 26 tháng 9 năm 2026 +**Phiên Bản Claude Code**: 2.1.283 **Nguồn**: - https://code.claude.com/docs/en/hooks - https://code.claude.com/docs/en/tools-reference#task-tool-availability +- https://code.claude.com/docs/en/plugins/manifest-reference#quoting-and-path-separators **Các Mô Hình Tương Thích**: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 diff --git a/vi/09-advanced-features/README.md b/vi/09-advanced-features/README.md index 0679bc0..74077e4 100644 --- a/vi/09-advanced-features/README.md +++ b/vi/09-advanced-features/README.md @@ -672,7 +672,7 @@ Permission modes control what actions Claude can take without explicit approval. | `bypassPermissions` | All actions, no permission checks (dangerous) | | `dontAsk` | Only pre-approved tools execute; all others denied | -Cycle through modes with `Shift+Tab` in the CLI. Set a default with the `--permission-mode` flag or the `permissions.defaultMode` setting. +Cycle through modes with `Shift+Tab` in the CLI. Set a default with the `--permission-mode` flag or the `permissions.defaultMode` setting. Since v2.1.283, an interactive session on a third-party provider (Bedrock, Google Cloud's Agent Platform, Foundry) or with telemetry off starts in auto mode when no permission mode is configured; `permissions.defaultMode` still overrides it. ### Activation Methods @@ -1024,7 +1024,7 @@ Keybindings support chord sequences (multi-key combinations): ``` **Keystroke syntax**: -- **Modifiers**: `ctrl`, `alt` (or `opt`), `shift`, `meta` (or `cmd`) +- **Modifiers**: `ctrl` (or `control`), `alt` (or `opt`, `option`, `meta`), `shift`, and `cmd` (or `command`, `super`, `win`) — `cmd` is a separate modifier, not an alias of `meta`, and is only detected in terminals that report the Super key - **Uppercase implies Shift**: `K` is equivalent to `shift+k` - **Special keys**: `escape`, `enter`, `return`, `tab`, `space`, `backspace`, `delete`, arrow keys @@ -1869,9 +1869,12 @@ For more information about Claude Code and related features: --- -**Cập Nhật Lần Cuối**: Ngày 19 tháng 9 năm 2026 -**Phiên Bản Claude Code**: 2.1.278 +**Cập Nhật Lần Cuối**: Ngày 26 tháng 9 năm 2026 +**Phiên Bản Claude Code**: 2.1.283 **Nguồn**: - https://code.claude.com/docs/en/permissions - https://code.claude.com/docs/en/auto-mode-classifier-billing +- https://code.claude.com/docs/en/keybindings#modifiers +- https://code.claude.com/docs/en/permission-modes +- https://github.com/anthropics/claude-code/releases/tag/v2.1.283 **Các Mô Hình Tương Thích**: Claude Sonnet 4.6, Claude Opus 4.6, Claude Haiku 4.5 diff --git a/vi/10-cli/README.md b/vi/10-cli/README.md index 8ea7924..3fac0a4 100644 --- a/vi/10-cli/README.md +++ b/vi/10-cli/README.md @@ -137,7 +137,7 @@ claude --model opusplan "design and implement the caching layer" | Cờ | Mô Tả | Ví Dụ | |------|-------------|---------| | `--system-prompt` | Thay thế toàn bộ prompt mặc định | `claude --system-prompt "You are a Python expert"` | -| `--system-prompt-file` | Tải prompt từ file (chế độ in) | `claude -p --system-prompt-file ./prompt.txt "query"` | +| `--system-prompt-file` | Tải prompt từ file, thay thế prompt mặc định | `claude -p --system-prompt-file ./prompt.txt "query"` | | `--append-system-prompt` | Append vào prompt mặc định | `claude --append-system-prompt "Always use TypeScript"` | ### Ví Dụ System Prompt / System Prompt Examples @@ -158,10 +158,10 @@ claude -p --system-prompt-file ./prompts/code-reviewer.txt "review main.py" | Cờ | Hành Vi | Interactive | Print | |------|----------|-------------|-------| | `--system-prompt` | Thay thế toàn bộ system prompt | ✅ | ✅ | -| `--system-prompt-file` | Thay thế bằng prompt từ file | ❌ | ✅ | +| `--system-prompt-file` | Thay thế bằng prompt từ file | ✅ | ✅ | | `--append-system-prompt` | Thêm vào system prompt mặc định | ✅ | ✅ | -**Chỉ sử dụng `--system-prompt-file` trong chế độ in. Trong chế độ tương tác, sử dụng `--system-prompt` hoặc `--append-system-prompt`.** +**Tất cả các cờ system prompt đều hoạt động trong cả chế độ tương tác và chế độ in (các dạng file từ v2.1.69). `--system-prompt` và `--system-prompt-file` thay thế prompt mặc định; các cờ append thêm vào prompt đó.** ## Công Cụ & Quản Lý Quyền / Tool & Permission Management @@ -834,8 +834,8 @@ claude -p --output-format json "query" --- -**Cập Nhật Lần Cuối**: Ngày 2 tháng 9 năm 2026 -**Phiên Bản Claude Code**: 2.1.257 +**Cập Nhật Lần Cuối**: Ngày 26 tháng 9 năm 2026 +**Phiên Bản Claude Code**: 2.1.283 **Nguồn**: - https://code.claude.com/docs/en/cli-reference **Các Mô Hình Tương Thích**: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 diff --git a/vi/CATALOG.md b/vi/CATALOG.md index 80b708c..83f7796 100644 --- a/vi/CATALOG.md +++ b/vi/CATALOG.md @@ -41,7 +41,7 @@ Commands là các lệnh tắt do người dùng gọi thực hiện các hành | `/diff` | Trình xem diff tương tác | Review thay đổi | | `/config` | Xem/chỉnh sửa cấu hình | Tùy chỉnh hành vi | | `/status` | Hiển thị trạng thái session | Kiểm tra trạng thái hiện tại | -| `/agents` | Liệt kê các agents có sẵn | Xem tùy chọn ủy quyền | +| `/agents` | Đã bị gỡ bỏ: wizard bị gỡ trong v2.1.198 và lệnh đã rời khỏi menu `/` và `/help` trong v2.1.281 — hãy yêu cầu Claude tạo hoặc quản lý subagents, hoặc chỉnh sửa `.claude/agents/` | Quản lý subagents bằng cách yêu cầu Claude | | `/skills` | Liệt kê các skills có sẵn | Xem khả năng auto-invoke | | `/hooks` | Liệt kê các hooks đã cấu hình | Debug tự động hóa | | `/insights` | Phân tích patterns của session | Tối ưu hóa session | @@ -522,8 +522,8 @@ chmod +x ~/.claude/hooks/*.sh --- -**Cập Nhật Lần Cuối**: Ngày 19 tháng 9 năm 2026 -**Phiên Bản Claude Code**: 2.1.278 +**Cập Nhật Lần Cuối**: Ngày 26 tháng 9 năm 2026 +**Phiên Bản Claude Code**: 2.1.283 **Nguồn**: - https://code.claude.com/docs/en/tools-reference#task-tool-availability - https://code.claude.com/docs/en/commands diff --git a/zh/01-slash-commands/README.md b/zh/01-slash-commands/README.md index 20ae53d..a4006f3 100644 --- a/zh/01-slash-commands/README.md +++ b/zh/01-slash-commands/README.md @@ -23,7 +23,7 @@ Claude Code 目前提供 55+ 个内置命令和 5 个内置 Skills。你可以 | 命令 | 作用 | |---------|---------| | `/add-dir ` | 添加工作目录 | -| `/agents` | 管理 agent 配置 | +| `/agents` | 已不再是可用命令：交互式向导已在 v2.1.198 中移除，且从 v2.1.281 起 `/agents` 不再出现在 `/` 菜单或 `/help` 中 — 输入它只会说明向导的去向。请让 Claude 创建或管理 subagent，或直接编辑 `.claude/agents/` 中的文件 | | `/branch [name]` | 切换到当前对话的一个副本，原对话保持不变 | | `/fork [prompt]` | 把当前对话复制到一个新的**后台会话**，你可以继续在这里工作；从这一刻起两者互相独立，副本会在 `claude agents` 里占一行（v2.1.212+） | | `/subtask ` | 派生一个**forked subagent**，它继承完整对话并去执行该任务，你可以继续工作；任务完成后结果会返回到本对话（v2.1.212+） | @@ -548,11 +548,12 @@ allowed-tools: Bash(npm *), Bash(git *) --- -**最后更新**: 2026 年 9 月 19 日 -**Claude Code 版本**: 2.1.278 +**最后更新**: 2026 年 9 月 26 日 +**Claude Code 版本**: 2.1.283 **来源**: - https://code.claude.com/docs/en/commands - https://code.claude.com/docs/en/output-styles +- https://github.com/anthropics/claude-code/releases/tag/v2.1.281 **兼容模型**: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Sonnet 4.6, Claude Opus 4.8, Claude Haiku 4.5 --- diff --git a/zh/03-skills/claude-md/SKILL.md b/zh/03-skills/claude-md/SKILL.md index f7583c5..fd44a28 100644 --- a/zh/03-skills/claude-md/SKILL.md +++ b/zh/03-skills/claude-md/SKILL.md @@ -193,7 +193,7 @@ agent_docs/ 如果用户请求创建或更新 AGENTS.md： -**从 v2.1.277 起，Claude Code 会直接把 `AGENTS.md` 作为项目指令读取。** 但前提是工作目录以及它上面的每一级目录都没有 `CLAUDE.md`、`.claude/CLAUDE.md` 或 `CLAUDE.local.md`。`~/.claude/CLAUDE.md`、受管理的 CLAUDE.md 和 `.claude/rules/` 不计入这项检查，它们仍会一并加载。该行为由 `/config` 中的 **Project instructions** 控制：`claude-md-or-agents-md`（默认）、`claude-md-and-agents-md`、`claude-md`、`managed-only`。在无法直接读取的环境中 — Bedrock/Vertex/Foundry、关闭遥测、升级后的第一个会话，或 `disableAllHooks`/`allowManagedHooksOnly` — 退回到在 CLAUDE.md 中用 `@AGENTS.md` 导入，或者把 `CLAUDE.md` 软链接到它。 +**从 v2.1.277 起，Claude Code 会直接把 `AGENTS.md` 作为项目指令读取。** 但前提是工作目录以及它上面的每一级目录都没有 `CLAUDE.md`、`.claude/CLAUDE.md` 或 `CLAUDE.local.md`。`~/.claude/CLAUDE.md`、受管理的 CLAUDE.md 和 `.claude/rules/` 不计入这项检查，它们仍会一并加载。该行为由 `/config` 中的 **Project instructions** 控制：`claude-md-or-agents-md`（默认）、`claude-md-and-agents-md`、`claude-md`、`managed-only`。在无法直接读取的环境中 — v2.1.277 之前的版本（在 Bedrock/Vertex/Foundry、LLM 网关或关闭遥测的情况下为 v2.1.281 之前），或升级后的第一个会话 — 退回到在 CLAUDE.md 中用 `@AGENTS.md` 导入，或者把 `CLAUDE.md` 软链接到它。 AGENTS.md 是一个跨工具的项目上下文文件 — 和 CLAUDE.md 属于同一*类*文档，而不是 agent 定义格式。它的存在是为了让多个编码 agent 共用同一套项目约定： - 构建、测试和 lint 命令 diff --git a/zh/04-subagents/README.md b/zh/04-subagents/README.md index b438162..33e135d 100644 --- a/zh/04-subagents/README.md +++ b/zh/04-subagents/README.md @@ -73,10 +73,14 @@ Claude Code 自带一些常见角色，例如： ### 使用 `/agents` 命令（推荐） -```bash -/agents +> **注意**：`/agents` 交互式向导已在 v2.1.198 中移除，且自 v2.1.281 起 `/agents` 不再出现在 `/` 菜单或 `/help` 中。请让 Claude 创建或管理 subagent（例如“创建一个审查安全问题的 subagent”），或直接添加 `.claude/agents/.md` 文件。 + +```text +创建一个审查代码安全漏洞的 subagent。 ``` +Claude 会为你写好 `.claude/agents/.md` 文件，并选择合适的 frontmatter（工具、模型、描述）。之后你可以手动调整该文件，或让 Claude 修改。 + ### 直接管理文件 #### 创建项目 subagent @@ -333,10 +337,15 @@ subagent 在隔离上下文里工作，但仍能把结果汇总回主会话。 ### 方法 1：使用 `/agents` -```bash -/agents +> **注意**：`/agents` 交互式向导已在 v2.1.198 中移除，且自 v2.1.281 起 `/agents` 不再出现在 `/` 菜单或 `/help` 中。请让 Claude 创建或管理 subagent（例如“创建一个审查安全问题的 subagent”），或直接添加 `.claude/agents/.md` 文件。 + +```text +创建一个项目级 subagent，用于运行测试并修复失败。 +给它 Bash、Read、Edit 和 Grep 的访问权限。 ``` +Claude 会写入带有合适 frontmatter 的 `.claude/agents/.md`。检查生成的文件后即可使用。 + ### 方法 2：复制到项目 ```bash @@ -353,8 +362,9 @@ cp 04-subagents/*.md ~/.claude/agents/ ### 验证 -- 运行 `/agents` -- 检查对应 agent 是否出现在列表中 +- 让 Claude 创建一个 subagent，或直接添加 `.claude/agents/.md` 文件 +- 列出目录，确认 agent 文件已就位：`ls .claude/agents/` +- 也可以询问 Claude 当前会话中有哪些可用的 subagent，它会报告可委派的内置与自定义 agent - 试着让 Claude 委派一个简单任务 ## 文件结构 @@ -378,3 +388,11 @@ cp 04-subagents/*.md ~/.claude/agents/ - [Memory 中文指南](../02-memory/README.md) - [Hooks 中文指南](../06-hooks/README.md) - [Plugins 中文指南](../07-plugins/README.md) + +--- + +**最后更新**: 2026 年 9 月 26 日 +**Claude Code 版本**: 2.1.283 +**来源**: +- https://code.claude.com/docs/en/sub-agents +- https://code.claude.com/docs/en/commands diff --git a/zh/10-cli/README.md b/zh/10-cli/README.md index 9de2b01..7a97bc4 100644 --- a/zh/10-cli/README.md +++ b/zh/10-cli/README.md @@ -137,7 +137,7 @@ claude --model opusplan "设计并实现缓存层" | 参数 | 说明 | 示例 | |------|-------------|---------| | `--system-prompt` | 替换整段默认系统提示词 | `claude --system-prompt "You are a Python expert"` | -| `--system-prompt-file` | 从文件加载提示词（仅打印模式） | `claude -p --system-prompt-file ./prompt.txt "query"` | +| `--system-prompt-file` | 从文件加载提示词，替换默认提示词 | `claude -p --system-prompt-file ./prompt.txt "query"` | | `--append-system-prompt` | 在默认提示词后追加内容 | `claude --append-system-prompt "Always use TypeScript"` | ### 系统提示词示例 @@ -158,10 +158,10 @@ claude -p --system-prompt-file ./prompts/code-reviewer.txt "review main.py" | 参数 | 行为 | 交互模式 | 打印模式 | |------|----------|-------------|-------| | `--system-prompt` | 替换整个默认系统提示词 | ✅ | ✅ | -| `--system-prompt-file` | 用文件中的提示词替换 | ❌ | ✅ | +| `--system-prompt-file` | 用文件中的提示词替换 | ✅ | ✅ | | `--append-system-prompt` | 追加到默认系统提示词后 | ✅ | ✅ | -**仅在打印模式中使用 `--system-prompt-file`。交互模式请使用 `--system-prompt` 或 `--append-system-prompt`。** +**所有系统提示词标志都可在交互模式和打印模式中使用（文件形式自 v2.1.69 起）。`--system-prompt` 和 `--system-prompt-file` 会替换默认提示词；append 类标志则在其基础上追加。** ## 工具与权限管理 @@ -642,7 +642,7 @@ claude --working-directory ./backend --mcp-config ./mcp.json --- -**最后更新**: 2026 年 9 月 2 日 -**Claude Code 版本**: 2.1.257 +**最后更新**: 2026 年 9 月 26 日 +**Claude Code 版本**: 2.1.283 **来源**: - https://code.claude.com/docs/en/cli-reference diff --git a/zh/CATALOG.md b/zh/CATALOG.md index 6bf2d7e..ef5661e 100644 --- a/zh/CATALOG.md +++ b/zh/CATALOG.md @@ -41,7 +41,7 @@ | `/diff` | 交互式 diff 查看器 | 审查变更 | | `/config` | 查看/编辑配置 | 自定义行为 | | `/status` | 显示会话状态 | 检查当前状态 | -| `/agents` | 列出可用 agents | 查看委派选项 | +| `/agents` | 已移除：向导在 v2.1.198 中移除，该命令在 v2.1.281 中从 `/` 菜单和 `/help` 中移除 — 请让 Claude 创建或管理 subagents，或编辑 `.claude/agents/` | 让 Claude 管理 subagents | | `/skills` | 列出可用 skills | 查看可自动触发的能力 | | `/hooks` | 列出已配置的 hooks | 调试自动化 | | `/insights` | 分析会话模式 | 优化会话使用 | @@ -531,8 +531,8 @@ claude mcp add github -- npx -y @modelcontextprotocol/server-github --- -**最后更新**: 2026 年 9 月 19 日 -**Claude Code 版本**: 2.1.278 +**最后更新**: 2026 年 9 月 26 日 +**Claude Code 版本**: 2.1.283 **来源**: - https://code.claude.com/docs/en/tools-reference#task-tool-availability - https://code.claude.com/docs/en/commands diff --git a/zh/claude_concepts_guide.md b/zh/claude_concepts_guide.md index e547c79..44ffdfb 100644 --- a/zh/claude_concepts_guide.md +++ b/zh/claude_concepts_guide.md @@ -1279,7 +1279,7 @@ graph TD ### 第 3 周：分发工作 - 创建第一个针对角色的 Subagent -- 使用 `/agents` +- 让 Claude 创建，或添加 `.claude/agents/.md` 文件（`/agents` 向导已移除） - 用简单任务测试委派 ### 第 4 周：全面自动化 @@ -1536,7 +1536,8 @@ claude -r "Feature" *适用于 Claude Fable 5、Claude Opus 5、Claude Sonnet 5、Claude Sonnet 4.6、Claude Opus 4.8、Claude Haiku 4.5* *现已覆盖：Hooks、Checkpoints、Planning Mode、Extended Thinking、Background Tasks、Permission Modes、Headless Mode、Session Management、Auto Memory、Agent Teams、Scheduled Tasks、Chrome Integration、Bundled Skills 等概念。* -**最后更新**：2026 年 9 月 2 日 -**Claude Code 版本**：2.1.257 +**最后更新**：2026 年 9 月 26 日 +**Claude Code 版本**：2.1.283 **来源**： - https://code.claude.com/docs/en/hooks +- https://code.claude.com/docs/en/sub-agents