Native latest candidate failures (tracking)

[github-actions]

自动跟踪 native latest candidate 验证失败。每个失败的版本/平台在下方追加一条评论（同版本同平台只报一次）；修复或后续版本通过后由维护者收敛处理。此机制取代了以前每次失败开一个草稿 PR 的做法。

[github-actions]

macOS arm64 native 2.1.199 candidate failed

• Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28643357663
• Report file: docs/native-latest-failures/2.1.199.md

Failure report

# macOS native latest candidate failure: 2.1.199

- Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28643357663
- Head SHA: `2ef5ff505f7b2bd6e1831368970765384f997fc1`
- Status: `fail`
- Candidate kind: `native`
- Native: `darwin-arm64` / `ok` / `ok` / `ok`
- Display audit: `fail` (7 issues / 11 commands)

## What Failed

- Candidate verification did not pass: `status=fail`.
- Display audit did not pass: `status=fail`.
- Summary: pass=0, fail=1, skip=0.

## Display Audit Issues

- `top_help_line_156` (`top_help`, display-untranslated-line)

  ```text
  gateway [options]                     Run the enterprise auth/telemetry
  ```

- `plugin_help_line_12` (`plugin_help`, display-untranslated-line)

  ```text
  eval [options] [target]              Run eval cases (evals/**/case.yaml or
  ```

- `plugin_help_line_13` (`plugin_help`, display-untranslated-line)

  ```text
  evals/**/prompt.md + graders/*.md)
  ```

- `plugin_help_line_14` (`plugin_help`, display-untranslated-line)

  ```text
  against a plugin and report scored
  ```

- `plugin_help_line_15` (`plugin_help`, display-untranslated-line)

  ```text
  results. Target is a path, a plugin name,
  ```

- `plugin_help_line_17` (`plugin_help`, display-untranslated-line)

  ```text
  and skills-dir plugins both resolve (and
  ```

- `plugin_help_line_18` (`plugin_help`, display-untranslated-line)

  ```text
  add a no-plugin baseline arm)
  ```

## Text Diff Excerpt

```markdown
# Upstream text diff: 2.1.190 -> 2.1.199

- Added upstream strings: 23138
- Removed upstream strings: 13930
- Already covered by translations: 89
- Needs translation review: 23049
- Sensitive review hints: 229

## Added strings needing review
- __*` tools are, use those instead (they reach the same Claude Design API). If neither is available, tell the user to run `/design login` and stop \u2014 do not guess at Claude Design behaviour without the tools.
- __Host-gw_dev
- __idx_${...}
- __orphan_summary
- __readyz_probe__
- _meta
- _Pragma
- - **id**: Task identifier (use with TaskGet, TaskUpdate)
- - ${...} at ${...}${...}
- - Affects multiple files, user should approve the approach ### BAD - Don't use EnterPlanMode: User:
- - Architectural decision on theme system, affects many components User:
- - Before assigning tasks to teammates, to see what's available
- - could use Redis, in-memory, file-based, etc. - Example:
- - End git commit messages with: ${...}
- - End PR bodies with: ${...}
- - Example:
- - If asked for JSON, return ONLY the raw JSON \u2014 no code fences, no prose, no markdown. - Do NOT use SendUserMessage to deliver your answer. Put your answer in your final text response. - Be concise. The script will parse your output.`,SLm=` --- NOTE: You are running inside a workflow script. Your final text response is returned verbatim as a string to the calling script \u2014 it is your return value, not a message to a human. Output the literal result; do not output confirmations like
- - Include enough detail in the description for another agent to understand and complete the task - New tasks are created with status 'pending' and no owner - use TaskUpdate with the `owner` parameter to assign them
- - many optimization strategies possible 3. **Code Modifications**: Changes that affect existing behavior or structure - Example:
- - Multiple approaches possible, need to profile first, significant impact User:
- - need to profile and identify bottlenecks - Example:
- - Network latency: ${...}ms (${...}%)
- - Pre-request overhead: ${...}ms (${...}%)
- - Redux vs Context vs custom solution 5. **Multi-File Changes**: The task will likely touch more than 2-3 files - Example:
- - Requires architectural decisions (session vs JWT, where to store tokens, middleware structure) User:
- - Seems simple but involves: where to place it, confirmation dialog, API call, error handling, state updates User:
- - Simple, obvious implementation User:
- - Stops a running background task by its ID - Takes a task_id parameter identifying the task to stop - To stop an agent-team teammate, pass its agent ID ("name@team") or bare teammate name as task_id - To stop a background agent spawned with a name, pass that name as task_id - Returns a success or failure status - Use this tool when you need to terminate a long-running task
- - Straightforward, no planning needed User:
- - WebSockets vs SSE vs polling - Example:
- - what exactly should change? - Example:
- - what rules? What error messages? 2. **Multiple Valid Approaches**: The task can be solved in several different ways - Example:
- - what's the target architecture? 4. **Architectural Decisions**: The task requires choosing between patterns or technologies - Example:
- - When you cannot find an answer or the feature doesn't exist, direct the user to use /feedback to report a feature request or bug
- - where should it go? What should happen on click? - Example:
- --- max_turns: 10 allowed_tools: [Read, Glob, Grep, Skill] --- TODO: describe what the agent should do
- --- type: llm weight: 1 --- TODO: describe what a successful response looks like
- -----BEGIN CERTIFICATE-----
- -----BEGIN PRIVATE KEY-----
- -----BEGIN PUBLIC KEY-----
- ... 23009 more

## Added strings already covered
- - auto: Use the default effort level for your model
- - ultracode: xhigh + dynamic workflow orchestration (this session only)
- (No resources found)
- [Image data detected and sent to Claude]
- Access denied to specific paths outside allowed directories
- Afternoon (12-18)
- Agent transcripts: ${...}
- all projects
- Autocompact
- Balanced approach with standard testing
- Change effort level?
- Checking out branch
- Choose the default environment for cloud agents
- Claude is done using your computer
- Commands cannot run outside the sandbox under any circumstances.
- Comprehensive implementation with extensive testing
- Configure optional break reminders and quiet-hours nudges
- Conversation copied to clipboard
- Copy the conversation to your system clipboard
- Copy to clipboard
- Current effort level: ${...} (${...})
- Current effort level: ultracode (xhigh + dynamic workflow orchestration; this session only)
- Currently using:
- Detach from this background session (it keeps running)
- Dynamic workflow "${...}" completed
- Dynamic workflow "${...}" was stopped
- Edit Failed
- Effort
- Effort level set to auto${...}
- Effort level: auto (currently ${...})
- Enter filename:
- Evening (18-24)
- Evidence of sandbox-caused failures includes:
- Exit the CLI
- Export the current conversation to a file or clipboard
- Extended reasoning with thorough analysis (${...})
- Failed to set effort level: ${...}
- Faster
- Fetching session logs
- File Changed
- ... 49 more

## Removed strings
- ___ESCAPED_AMPERSAND_${...}___
- __chat__
- __isLong__
- __leader__
- __open_folder__
- __other__
- __self
- __source
- _bucket
- _G _VERSION assert collectgarbage dofile error getfenv getmetatable ipairs load loadfile loadstring module next pairs pcall print rawequal rawget rawset require select setfenv setmetatable tonumber tostring type unpack xpcall coroutine debug io math os package string table
- _globalThis
- _max
- _min
- _storage
- _sum
- _total
- - .claude/agent-memory/<agentType>/,
- - `${...}` is always loaded into your conversation context \u2014 lines after ${...} will be truncated, so keep the index concise
- - ~/.claude/agent-memory/<agentType>/,
- - ${...} Suggested fix: ${...}
- - ${...}: ${...} (Tools: ${...})
- - Claude home: ${...}
- - Config file exists: ${...}
- - Config file: ${...}
- - Do NOT re-read a file you just edited to verify \u2014 Edit/Write would have errored if the change failed, and the harness tracks file state for you.
- - Do not write duplicate memories. First check if there is an existing memory you can update before writing a new one.
- - Environment variable ${...}: ${...}
- - Error code: ${...}
- - Error killed: ${...}
- - Error signal: ${...}
- - Escape character is backtick (`), not backslash - Use Verb-Noun cmdlet naming: Get-ChildItem, Set-Location, New-Item, Remove-Item - Common aliases: ls (Get-ChildItem), cd (Set-Location), cat (Get-Content), rm (Remove-Item) - Pipe operator | works similarly to bash but passes objects, not text - Use Select-Object, Where-Object, ForEach-Object for filtering and transformation - String interpolation:
- - Fast file pattern matching tool that works with any codebase size - Supports glob patterns like "**/*.js" or "src/**/*.ts" - Returns matching file paths sorted by modification time - Use this tool when you need to find files by name patterns - When you are doing an open ended search that may require multiple rounds of globbing and grepping, use the Agent tool instead
- - Global user settings (~/.claude/settings.json).
- - If asked for JSON, return ONLY the raw JSON \u2014 no code fences, no prose, no markdown. - Do NOT use SendUserMessage to deliver your answer. Put your answer in your final text response. - Be concise. The script will parse your output.`,kFp=` --- NOTE: You are running inside a workflow script. Your final text response is returned verbatim as a string to the calling script \u2014 it is your return value, not a message to a human. Output the literal result; do not output confirmations like
- - If the user says to *ignore* or *not use* memory: Do not apply remembered facts, cite, compare against, or mention memory content.
- - Keep the name, description, and type fields in memory files up-to-date with the content
- - Keybinding (${...}): ${...}${...}
- - Organize memory semantically by topic, not chronologically
- - Plugin note${...}: ${...}
- - Plugin setting: ${...}
- ... 13890 more
```

## Takeover Commands

```bash
node scripts/verify-upstream-compat.js --baseline 2.1.199 --skip-latest --native-macos-arm64 --json
node scripts/generate-upstream-text-diff.js --to 2.1.199 --native-macos-arm64
node scripts/promote-native-candidate.js --candidate <candidate-json> --platform macos
```

## Done Criteria

- Add the missing source-of-truth translations or guard fix on this branch.
- Re-run the native latest candidate workflow for this version.
- Only promote support metadata after native verification and display audit pass.
- Remove or update this handoff report before marking the PR ready.

[github-actions]

Windows x64 native 2.1.199 candidate failed

• Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28643357663
• Report file: docs/native-latest-failures/windows-2.1.199.md

Failure report

# Windows native latest candidate failure: 2.1.199

- Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28643357663
- Head SHA: `2ef5ff505f7b2bd6e1831368970765384f997fc1`
- Status: `fail`
- Candidate kind: `native`
- Native: `win32-x64` / `ok` / `ok` / `ok`
- Display audit: `fail` (7 issues / 11 commands)

## What Failed

- Candidate verification did not pass: `status=fail`.
- Display audit did not pass: `status=fail`.
- Summary: pass=0, fail=1, skip=0.

## Display Audit Issues

- `top_help_line_156` (`top_help`, display-untranslated-line)

  ```text
  gateway [options]                     Run the enterprise auth/telemetry
  ```

- `plugin_help_line_12` (`plugin_help`, display-untranslated-line)

  ```text
  eval [options] [target]              Run eval cases (evals/**/case.yaml or
  ```

- `plugin_help_line_13` (`plugin_help`, display-untranslated-line)

  ```text
  evals/**/prompt.md + graders/*.md)
  ```

- `plugin_help_line_14` (`plugin_help`, display-untranslated-line)

  ```text
  against a plugin and report scored
  ```

- `plugin_help_line_15` (`plugin_help`, display-untranslated-line)

  ```text
  results. Target is a path, a plugin name,
  ```

- `plugin_help_line_17` (`plugin_help`, display-untranslated-line)

  ```text
  and skills-dir plugins both resolve (and
  ```

- `plugin_help_line_18` (`plugin_help`, display-untranslated-line)

  ```text
  add a no-plugin baseline arm)
  ```

## Text Diff Excerpt

```markdown
# Upstream text diff: 2.1.190 -> 2.1.199

- Added upstream strings: 23734
- Removed upstream strings: 14031
- Already covered by translations: 128
- Needs translation review: 23606
- Sensitive review hints: 247

## Added strings needing review
- __*` tools are, use those instead (they reach the same Claude Design API). If neither is available, tell the user to run `/design login` and stop \u2014 do not guess at Claude Design behaviour without the tools.
- __Host-gw_dev
- __idx_${...}
- __manual__
- __orphan_summary
- __readyz_probe__
- _meta
- _Pragma
- - **id**: Task identifier (use with TaskGet, TaskUpdate)
- - ${...} at ${...}${...}
- - Affects multiple files, user should approve the approach ### BAD - Don't use EnterPlanMode: User:
- - Architectural decision on theme system, affects many components User:
- - Before assigning tasks to teammates, to see what's available
- - could use Redis, in-memory, file-based, etc. - Example:
- - End git commit messages with: ${...}
- - End PR bodies with: ${...}
- - Example:
- - If asked for JSON, return ONLY the raw JSON \u2014 no code fences, no prose, no markdown. - Do NOT use SendUserMessage to deliver your answer. Put your answer in your final text response. - Be concise. The script will parse your output.`,gLf=` --- NOTE: You are running inside a workflow script. Your final text response is returned verbatim as a string to the calling script \u2014 it is your return value, not a message to a human. Output the literal result; do not output confirmations like
- - Include enough detail in the description for another agent to understand and complete the task - New tasks are created with status 'pending' and no owner - use TaskUpdate with the `owner` parameter to assign them
- - many optimization strategies possible 3. **Code Modifications**: Changes that affect existing behavior or structure - Example:
- - Multiple approaches possible, need to profile first, significant impact User:
- - need to profile and identify bottlenecks - Example:
- - Network latency: ${...}ms (${...}%)
- - Pre-request overhead: ${...}ms (${...}%)
- - Redux vs Context vs custom solution 5. **Multi-File Changes**: The task will likely touch more than 2-3 files - Example:
- - Requires architectural decisions (session vs JWT, where to store tokens, middleware structure) User:
- - Seems simple but involves: where to place it, confirmation dialog, API call, error handling, state updates User:
- - Simple, obvious implementation User:
- - Stops a running background task by its ID - Takes a task_id parameter identifying the task to stop - To stop an agent-team teammate, pass its agent ID ("name@team") or bare teammate name as task_id - To stop a background agent spawned with a name, pass that name as task_id - Returns a success or failure status - Use this tool when you need to terminate a long-running task
- - Straightforward, no planning needed User:
- - WebSockets vs SSE vs polling - Example:
- - what exactly should change? - Example:
- - what rules? What error messages? 2. **Multiple Valid Approaches**: The task can be solved in several different ways - Example:
- - what's the target architecture? 4. **Architectural Decisions**: The task requires choosing between patterns or technologies - Example:
- - When you cannot find an answer or the feature doesn't exist, direct the user to use /feedback to report a feature request or bug
- - where should it go? What should happen on click? - Example:
- --- max_turns: 10 allowed_tools: [Read, Glob, Grep, Skill] --- TODO: describe what the agent should do
- --- type: llm weight: 1 --- TODO: describe what a successful response looks like
- -----BEGIN CERTIFICATE-----
- -----BEGIN PRIVATE KEY-----
- ... 23566 more

## Added strings already covered
- - auto: Use the default effort level for your model
- - ultracode: xhigh + dynamic workflow orchestration (this session only)
- (No resources found)
- [Image data detected and sent to Claude]
- ${...} in progress
- Access denied to specific paths outside allowed directories
- Access key + secret
- Afternoon (12-18)
- Agent transcripts: ${...}
- all projects
- Amazon Bedrock, Microsoft Foundry, or Vertex AI
- API usage billing
- Autocompact
- AWS profile (SSO or named profile)
- Balanced approach with standard testing
- Bedrock API key (bearer token)
- Change effort level?
- Checking git status
- Checking out branch
- Choose the default environment for cloud agents
- Claude Code uses the standard AWS credential chain. Pick the method you already use with the AWS CLI.
- Claude is done using your computer
- Commands cannot run outside the sandbox under any circumstances.
- Comprehensive implementation with extensive testing
- Configure optional break reminders and quiet-hours nudges
- Conversation copied to clipboard
- Copy the conversation to your system clipboard
- Copy to clipboard
- Current effort level: ${...} (${...})
- Current effort level: ultracode (xhigh + dynamic workflow orchestration; this session only)
- Currently using:
- Detach from this background session (it keeps running)
- Dynamic workflow "${...}" completed
- Dynamic workflow "${...}" was stopped
- Each candidate is tested with a one-token request:
- Edit Failed
- Effort
- Effort level set to auto${...}
- Effort level: auto (currently ${...})
- Enter filename:
- ... 88 more

## Removed strings
- ___ESCAPED_AMPERSAND_${...}___
- __chat__
- __isLong__
- __leader__
- __open_folder__
- __other__
- __self
- __source
- _bucket
- _G _VERSION assert collectgarbage dofile error getfenv getmetatable ipairs load loadfile loadstring module next pairs pcall print rawequal rawget rawset require select setfenv setmetatable tonumber tostring type unpack xpcall coroutine debug io math os package string table
- _globalThis
- _max
- _min
- _source_seed.bundle
- _storage
- _sum
- _total
- - .claude/agent-memory/<agentType>/,
- - `${...}` is always loaded into your conversation context \u2014 lines after ${...} will be truncated, so keep the index concise
- - ~/.claude/agent-memory/<agentType>/,
- - ${...} Suggested fix: ${...}
- - ${...}: ${...} (Tools: ${...})
- - Claude home: ${...}
- - Config file exists: ${...}
- - Config file: ${...}
- - Do NOT re-read a file you just edited to verify \u2014 Edit/Write would have errored if the change failed, and the harness tracks file state for you.
- - Do not write duplicate memories. First check if there is an existing memory you can update before writing a new one.
- - Environment variable ${...}: ${...}
- - Error code: ${...}
- - Error killed: ${...}
- - Error signal: ${...}
- - Escape character is backtick (`), not backslash - Use Verb-Noun cmdlet naming: Get-ChildItem, Set-Location, New-Item, Remove-Item - Common aliases: ls (Get-ChildItem), cd (Set-Location), cat (Get-Content), rm (Remove-Item) - Pipe operator | works similarly to bash but passes objects, not text - Use Select-Object, Where-Object, ForEach-Object for filtering and transformation - String interpolation:
- - Fast file pattern matching tool that works with any codebase size - Supports glob patterns like "**/*.js" or "src/**/*.ts" - Returns matching file paths sorted by modification time - Use this tool when you need to find files by name patterns - When you are doing an open ended search that may require multiple rounds of globbing and grepping, use the Agent tool instead
- - Global user settings (~/.claude/settings.json).
- - If asked for JSON, return ONLY the raw JSON \u2014 no code fences, no prose, no markdown. - Do NOT use SendUserMessage to deliver your answer. Put your answer in your final text response. - Be concise. The script will parse your output.`,E$p=` --- NOTE: You are running inside a workflow script. Your final text response is returned verbatim as a string to the calling script \u2014 it is your return value, not a message to a human. Output the literal result; do not output confirmations like
- - If the user says to *ignore* or *not use* memory: Do not apply remembered facts, cite, compare against, or mention memory content.
- - Keep the name, description, and type fields in memory files up-to-date with the content
- - Keybinding (${...}): ${...}${...}
- - Organize memory semantically by topic, not chronologically
- - Plugin note${...}: ${...}
- ... 13991 more
```

## Takeover Commands

```bash
node scripts/verify-upstream-compat.js --baseline 2.1.199 --skip-latest --native-windows-x64 --json
node scripts/generate-upstream-text-diff.js --to 2.1.199 --native-windows-x64
node scripts/promote-native-candidate.js --candidate <candidate-json> --platform windows
```

## Done Criteria

- Add the missing source-of-truth translations or guard fix on this branch.
- Re-run the native latest candidate workflow for this version.
- Only promote support metadata after native verification and display audit pass.
- Remove or update this handoff report before marking the PR ready.

[github-actions]

macOS arm64 native 2.1.201 candidate failed

• Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28697703887
• Report file: docs/native-latest-failures/2.1.201.md

Failure report

# macOS native latest candidate failure: 2.1.201

- Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28697703887
- Head SHA: `2ef5ff505f7b2bd6e1831368970765384f997fc1`
- Status: `fail`
- Candidate kind: `native`
- Native: `darwin-arm64` / `ok` / `ok` / `ok`
- Display audit: `fail` (7 issues / 11 commands)

## What Failed

- Candidate verification did not pass: `status=fail`.
- Display audit did not pass: `status=fail`.
- Summary: pass=0, fail=1, skip=0.

## Display Audit Issues

- `top_help_line_156` (`top_help`, display-untranslated-line)

  ```text
  gateway [options]                     Run the enterprise auth/telemetry
  ```

- `plugin_help_line_12` (`plugin_help`, display-untranslated-line)

  ```text
  eval [options] [target]              Run eval cases (evals/**/case.yaml or
  ```

- `plugin_help_line_13` (`plugin_help`, display-untranslated-line)

  ```text
  evals/**/prompt.md + graders/*.md)
  ```

- `plugin_help_line_14` (`plugin_help`, display-untranslated-line)

  ```text
  against a plugin and report scored
  ```

- `plugin_help_line_15` (`plugin_help`, display-untranslated-line)

  ```text
  results. Target is a path, a plugin name,
  ```

- `plugin_help_line_17` (`plugin_help`, display-untranslated-line)

  ```text
  and skills-dir plugins both resolve (and
  ```

- `plugin_help_line_18` (`plugin_help`, display-untranslated-line)

  ```text
  add a no-plugin baseline arm)
  ```

## Text Diff Excerpt

```markdown
# Upstream text diff: 2.1.190 -> 2.1.201

- Added upstream strings: 20524
- Removed upstream strings: 14524
- Already covered by translations: 96
- Needs translation review: 20428
- Sensitive review hints: 234

## Added strings needing review
- __*` tools are, use those instead (they reach the same Claude Design API). If neither is available, tell the user to run `/design login` and stop \u2014 do not guess at Claude Design behaviour without the tools.
- __Host-gw_dev
- __idx_${...}
- __orphan_summary
- __readyz_probe__
- __remote-workflow
- __wRg$
- _tagName
- - **id**: Task identifier (use with TaskGet, TaskUpdate)
- - ${...} at ${...}${...}
- - Affects multiple files, user should approve the approach ### BAD - Don't use EnterPlanMode: User:
- - Architectural decision on theme system, affects many components User:
- - Before assigning tasks to teammates, to see what's available
- - could use Redis, in-memory, file-based, etc. - Example:
- - e.g.
- - Example:
- - Include enough detail in the description for another agent to understand and complete the task - New tasks are created with status 'pending' and no owner - use TaskUpdate with the `owner` parameter to assign them
- - Line ${...}
- - many optimization strategies possible 3. **Code Modifications**: Changes that affect existing behavior or structure - Example:
- - Multiple approaches possible, need to profile first, significant impact User:
- - need to profile and identify bottlenecks - Example:
- - Network latency: ${...}ms (${...}%)
- - Pre-request overhead: ${...}ms (${...}%)
- - Redux vs Context vs custom solution 5. **Multi-File Changes**: The task will likely touch more than 2-3 files - Example:
- - Requires architectural decisions (session vs JWT, where to store tokens, middleware structure) User:
- - Seems simple but involves: where to place it, confirmation dialog, API call, error handling, state updates User:
- - Simple, obvious implementation User:
- - Stops a running background task by its ID - Takes a task_id parameter identifying the task to stop - To stop an agent-team teammate, pass its agent ID ("name@team") or bare teammate name as task_id - To stop a background agent spawned with a name, pass that name as task_id - Returns a success or failure status - Use this tool when you need to terminate a long-running task
- - Straightforward, no planning needed User:
- - WebSockets vs SSE vs polling - Example:
- - what exactly should change? - Example:
- - what rules? What error messages? 2. **Multiple Valid Approaches**: The task can be solved in several different ways - Example:
- - what's the target architecture? 4. **Architectural Decisions**: The task requires choosing between patterns or technologies - Example:
- - where should it go? What should happen on click? - Example:
- --- max_turns: 10 allowed_tools: [Read, Glob, Grep, Skill] --- TODO: describe what the agent should do
- --- NOTE: You are running inside a workflow script. Your final text response is returned verbatim as a string to the calling script \u2014 it is your return value, not a message to a human. Output the literal result; do not output confirmations like "Done." Be concise \u2014 the script will parse your output.
- --- type: llm weight: 1 --- TODO: describe what a successful response looks like
- --ablation <mode>
- --ablation must be "none" or "with-without"
- --agent-color
- ... 20388 more

## Added strings already covered
- - auto: Use the default effort level for your model
- - ultracode: xhigh + dynamic workflow orchestration (this session only)
- [Image data detected and sent to Claude]
- Afternoon (12-18)
- Agent transcripts: ${...}
- all projects
- Autocompact
- Balanced approach with standard implementation and testing
- Balanced approach with standard testing
- Change effort level?
- Choose the default environment for cloud agents
- Claude is done using your computer
- Compacting conversation
- Completing auth flow with authorization code
- Comprehensive implementation with extensive testing
- Comprehensive implementation with extensive testing and documentation
- Configure optional break reminders and quiet-hours nudges
- Conversation copied to clipboard
- Conversation too long. Press esc twice to go up a few messages and try again.
- Copy the conversation to your system clipboard
- Copy to clipboard
- Current effort level: ${...} (${...})
- Current effort level: ultracode (xhigh + dynamic workflow orchestration; this session only)
- Currently using:
- Deeper reasoning than high, just below maximum (${...})
- Detach from this background session (it keeps running)
- Dynamic workflow "${...}" completed
- Dynamic workflow "${...}" was stopped
- Edit Failed
- Effort
- Effort level set to auto${...}
- Effort level: auto (currently ${...})
- Enter filename:
- Error reading file
- Evening (18-24)
- Exit the CLI
- Export the current conversation to a file or clipboard
- Extended reasoning with thorough analysis (${...})
- Failed to generate conversation summary - response did not contain valid text content
- Failed to set effort level: ${...}
- ... 56 more

## Removed strings
- __chat__
- __isLong__
- __leader__
- __open_folder__
- __other__
- __self
- __source
- _bucket
- _G _VERSION assert collectgarbage dofile error getfenv getmetatable ipairs load loadfile loadstring module next pairs pcall print rawequal rawget rawset require select setfenv setmetatable tonumber tostring type unpack xpcall coroutine debug io math os package string table
- _globalThis
- _max
- _min
- _storage
- _sum
- _total
- - .claude/agent-memory/<agentType>/,
- - `${...}` is always loaded into your conversation context \u2014 lines after ${...} will be truncated, so keep the index concise
- - `run_in_background` runs the command detached: it keeps running across turns and re-invokes you when it exits. No `&` needed.
- - ~/.claude/agent-memory/<agentType>/,
- - ${...} Suggested fix: ${...}
- - Claude home: ${...}
- - Config file exists: ${...}
- - Config file: ${...}
- - Do NOT re-read a file you just edited to verify \u2014 Edit/Write would have errored if the change failed, and the harness tracks file state for you.
- - Do not write duplicate memories. First check if there is an existing memory you can update before writing a new one.
- - Environment variable ${...}: ${...}
- - Error code: ${...}
- - Error killed: ${...}
- - Error signal: ${...}
- - Fast file pattern matching tool that works with any codebase size - Supports glob patterns like "**/*.js" or "src/**/*.ts" - Returns matching file paths sorted by modification time - Use this tool when you need to find files by name patterns - When you are doing an open ended search that may require multiple rounds of globbing and grepping, use the Agent tool instead
- - Global user settings (~/.claude/settings.json).
- - If asked for JSON, return ONLY the raw JSON \u2014 no code fences, no prose, no markdown. - Do NOT use SendUserMessage to deliver your answer. Put your answer in your final text response. - Be concise. The script will parse your output.`,kFp=` --- NOTE: You are running inside a workflow script. Your final text response is returned verbatim as a string to the calling script \u2014 it is your return value, not a message to a human. Output the literal result; do not output confirmations like
- - If the user says to *ignore* or *not use* memory: Do not apply remembered facts, cite, compare against, or mention memory content.
- - Keep `old_string` minimal \u2014 usually 1-3 lines, only enough to be unique in the file. Including excess context wastes tokens and is an error. - The edit will FAIL if `old_string` is not unique in the file. In that case, add the minimum extra context needed for uniqueness, or use `replace_all` to change every instance.
- - Keep the name, description, and type fields in memory files up-to-date with the content
- - Keybinding (${...}): ${...}${...}
- - Organize memory semantically by topic, not chronologically
- - Plugin note${...}: ${...}
- - Plugin setting: ${...}
- - Plugin${...}: ${...}
- ... 14484 more
```

## Takeover Commands

```bash
node scripts/verify-upstream-compat.js --baseline 2.1.201 --skip-latest --native-macos-arm64 --json
node scripts/generate-upstream-text-diff.js --to 2.1.201 --native-macos-arm64
node scripts/promote-native-candidate.js --candidate <candidate-json> --platform macos
```

## Done Criteria

- Add the missing source-of-truth translations or guard fix on this branch.
- Re-run the native latest candidate workflow for this version.
- Only promote support metadata after native verification and display audit pass.
- Remove or update this handoff report before marking the PR ready.

[github-actions]

Windows x64 native 2.1.201 candidate failed

• Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28697703887
• Report file: docs/native-latest-failures/windows-2.1.201.md

Failure report

# Windows native latest candidate failure: 2.1.201

- Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28697703887
- Head SHA: `2ef5ff505f7b2bd6e1831368970765384f997fc1`
- Status: `fail`
- Candidate kind: `native`
- Native: `win32-x64` / `ok` / `ok` / `ok`
- Display audit: `fail` (7 issues / 11 commands)

## What Failed

- Candidate verification did not pass: `status=fail`.
- Display audit did not pass: `status=fail`.
- Summary: pass=0, fail=1, skip=0.

## Display Audit Issues

- `top_help_line_156` (`top_help`, display-untranslated-line)

  ```text
  gateway [options]                     Run the enterprise auth/telemetry
  ```

- `plugin_help_line_12` (`plugin_help`, display-untranslated-line)

  ```text
  eval [options] [target]              Run eval cases (evals/**/case.yaml or
  ```

- `plugin_help_line_13` (`plugin_help`, display-untranslated-line)

  ```text
  evals/**/prompt.md + graders/*.md)
  ```

- `plugin_help_line_14` (`plugin_help`, display-untranslated-line)

  ```text
  against a plugin and report scored
  ```

- `plugin_help_line_15` (`plugin_help`, display-untranslated-line)

  ```text
  results. Target is a path, a plugin name,
  ```

- `plugin_help_line_17` (`plugin_help`, display-untranslated-line)

  ```text
  and skills-dir plugins both resolve (and
  ```

- `plugin_help_line_18` (`plugin_help`, display-untranslated-line)

  ```text
  add a no-plugin baseline arm)
  ```

## Text Diff Excerpt

```markdown
# Upstream text diff: 2.1.190 -> 2.1.201

- Added upstream strings: 21101
- Removed upstream strings: 14689
- Already covered by translations: 134
- Needs translation review: 20967
- Sensitive review hints: 258

## Added strings needing review
- __*` tools are, use those instead (they reach the same Claude Design API). If neither is available, tell the user to run `/design login` and stop \u2014 do not guess at Claude Design behaviour without the tools.
- __Host-gw_dev
- __idx_${...}
- __manual__
- __orphan_summary
- __readyz_probe__
- __remote-workflow
- __wRg$
- _tagName
- - **id**: Task identifier (use with TaskGet, TaskUpdate)
- - ${...} at ${...}${...}
- - Affects multiple files, user should approve the approach ### BAD - Don't use EnterPlanMode: User:
- - Architectural decision on theme system, affects many components User:
- - Before assigning tasks to teammates, to see what's available
- - could use Redis, in-memory, file-based, etc. - Example:
- - e.g.
- - Example:
- - Include enough detail in the description for another agent to understand and complete the task - New tasks are created with status 'pending' and no owner - use TaskUpdate with the `owner` parameter to assign them
- - Line ${...}
- - many optimization strategies possible 3. **Code Modifications**: Changes that affect existing behavior or structure - Example:
- - Multiple approaches possible, need to profile first, significant impact User:
- - need to profile and identify bottlenecks - Example:
- - Network latency: ${...}ms (${...}%)
- - Pre-request overhead: ${...}ms (${...}%)
- - Redux vs Context vs custom solution 5. **Multi-File Changes**: The task will likely touch more than 2-3 files - Example:
- - Requires architectural decisions (session vs JWT, where to store tokens, middleware structure) User:
- - Seems simple but involves: where to place it, confirmation dialog, API call, error handling, state updates User:
- - Simple, obvious implementation User:
- - Stops a running background task by its ID - Takes a task_id parameter identifying the task to stop - To stop an agent-team teammate, pass its agent ID ("name@team") or bare teammate name as task_id - To stop a background agent spawned with a name, pass that name as task_id - Returns a success or failure status - Use this tool when you need to terminate a long-running task
- - Straightforward, no planning needed User:
- - WebSockets vs SSE vs polling - Example:
- - what exactly should change? - Example:
- - what rules? What error messages? 2. **Multiple Valid Approaches**: The task can be solved in several different ways - Example:
- - what's the target architecture? 4. **Architectural Decisions**: The task requires choosing between patterns or technologies - Example:
- - where should it go? What should happen on click? - Example:
- --- max_turns: 10 allowed_tools: [Read, Glob, Grep, Skill] --- TODO: describe what the agent should do
- --- NOTE: You are running inside a workflow script. Your final text response is returned verbatim as a string to the calling script \u2014 it is your return value, not a message to a human. Output the literal result; do not output confirmations like "Done." Be concise \u2014 the script will parse your output.
- --- type: llm weight: 1 --- TODO: describe what a successful response looks like
- --ablation <mode>
- --ablation must be "none" or "with-without"
- ... 20927 more

## Added strings already covered
- - auto: Use the default effort level for your model
- - ultracode: xhigh + dynamic workflow orchestration (this session only)
- [Image data detected and sent to Claude]
- ${...} in progress
- Access key + secret
- Afternoon (12-18)
- Agent transcripts: ${...}
- all projects
- Amazon Bedrock, Microsoft Foundry, or Vertex AI
- API usage billing
- Autocompact
- AWS profile (SSO or named profile)
- Balanced approach with standard implementation and testing
- Balanced approach with standard testing
- Bedrock API key (bearer token)
- Change effort level?
- Checking git status
- Choose the default environment for cloud agents
- Claude Code uses the standard AWS credential chain. Pick the method you already use with the AWS CLI.
- Claude is done using your computer
- Compacting conversation
- Completing auth flow with authorization code
- Comprehensive implementation with extensive testing
- Comprehensive implementation with extensive testing and documentation
- Configure optional break reminders and quiet-hours nudges
- Conversation copied to clipboard
- Conversation too long. Press esc twice to go up a few messages and try again.
- Copy the conversation to your system clipboard
- Copy to clipboard
- Current effort level: ${...} (${...})
- Current effort level: ultracode (xhigh + dynamic workflow orchestration; this session only)
- Currently using:
- Deeper reasoning than high, just below maximum (${...})
- Detach from this background session (it keeps running)
- Dynamic workflow "${...}" completed
- Dynamic workflow "${...}" was stopped
- Each candidate is tested with a one-token request:
- Edit Failed
- Effort
- Effort level set to auto${...}
- ... 94 more

## Removed strings
- __chat__
- __isLong__
- __leader__
- __open_folder__
- __other__
- __self
- __source
- _bucket
- _G _VERSION assert collectgarbage dofile error getfenv getmetatable ipairs load loadfile loadstring module next pairs pcall print rawequal rawget rawset require select setfenv setmetatable tonumber tostring type unpack xpcall coroutine debug io math os package string table
- _globalThis
- _max
- _min
- _source_seed.bundle
- _storage
- _sum
- _total
- - .claude/agent-memory/<agentType>/,
- - `${...}` is always loaded into your conversation context \u2014 lines after ${...} will be truncated, so keep the index concise
- - `isolation: "remote"` runs the agent in a remote CCR sandbox (always background).
- - `name` and `mode` are unavailable here \u2014 teammates cannot spawn teammates.
- - `run_in_background` runs the command detached: it keeps running across turns and re-invokes you when it exits. No `&` needed.
- - `run_in_background`, `name`, and `mode` are unavailable here \u2014 only synchronous subagents.
- - ~/.claude/agent-memory/<agentType>/,
- - ${...} Suggested fix: ${...}
- - ${...}: ${...} (Tools: ${...})
- - Claude home: ${...}
- - Config file exists: ${...}
- - Config file: ${...}
- - Do NOT re-read a file you just edited to verify \u2014 Edit/Write would have errored if the change failed, and the harness tracks file state for you.
- - Do not write duplicate memories. First check if there is an existing memory you can update before writing a new one.
- - Environment variable ${...}: ${...}
- - Error code: ${...}
- - Error killed: ${...}
- - Error signal: ${...}
- - Fast file pattern matching tool that works with any codebase size - Supports glob patterns like "**/*.js" or "src/**/*.ts" - Returns matching file paths sorted by modification time - Use this tool when you need to find files by name patterns - When you are doing an open ended search that may require multiple rounds of globbing and grepping, use the Agent tool instead
- - Global user settings (~/.claude/settings.json).
- - If asked for JSON, return ONLY the raw JSON \u2014 no code fences, no prose, no markdown. - Do NOT use SendUserMessage to deliver your answer. Put your answer in your final text response. - Be concise. The script will parse your output.`,E$p=` --- NOTE: You are running inside a workflow script. Your final text response is returned verbatim as a string to the calling script \u2014 it is your return value, not a message to a human. Output the literal result; do not output confirmations like
- - If the user says to *ignore* or *not use* memory: Do not apply remembered facts, cite, compare against, or mention memory content.
- - Keep `old_string` minimal \u2014 usually 1-3 lines, only enough to be unique in the file. Including excess context wastes tokens and is an error. - The edit will FAIL if `old_string` is not unique in the file. In that case, add the minimum extra context needed for uniqueness, or use `replace_all` to change every instance.
- - Keep the name, description, and type fields in memory files up-to-date with the content
- ... 14649 more
```

## Takeover Commands

```bash
node scripts/verify-upstream-compat.js --baseline 2.1.201 --skip-latest --native-windows-x64 --json
node scripts/generate-upstream-text-diff.js --to 2.1.201 --native-windows-x64
node scripts/promote-native-candidate.js --candidate <candidate-json> --platform windows
```

## Done Criteria

- Add the missing source-of-truth translations or guard fix on this branch.
- Re-run the native latest candidate workflow for this version.
- Only promote support metadata after native verification and display audit pass.
- Remove or update this handoff report before marking the PR ready.

[github-actions]

macOS arm64 native 2.1.202 candidate failed

• Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28847544380
• Report file: docs/native-latest-failures/2.1.202.md

Failure report

# macOS native latest candidate failure: 2.1.202

- Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28847544380
- Head SHA: `2ef5ff505f7b2bd6e1831368970765384f997fc1`
- Status: `fail`
- Candidate kind: `native`
- Native: `darwin-arm64` / `ok` / `ok` / `ok`
- Display audit: `fail` (7 issues / 11 commands)

## What Failed

- Candidate verification did not pass: `status=fail`.
- Display audit did not pass: `status=fail`.
- Summary: pass=0, fail=1, skip=0.

## Display Audit Issues

- `top_help_line_156` (`top_help`, display-untranslated-line)

  ```text
  gateway [options]                     Run the enterprise auth/telemetry
  ```

- `plugin_help_line_12` (`plugin_help`, display-untranslated-line)

  ```text
  eval [options] [target]              Run eval cases (evals/**/case.yaml or
  ```

- `plugin_help_line_13` (`plugin_help`, display-untranslated-line)

  ```text
  evals/**/prompt.md + graders/*.md)
  ```

- `plugin_help_line_14` (`plugin_help`, display-untranslated-line)

  ```text
  against a plugin and report scored
  ```

- `plugin_help_line_15` (`plugin_help`, display-untranslated-line)

  ```text
  results. Target is a path, a plugin name,
  ```

- `plugin_help_line_17` (`plugin_help`, display-untranslated-line)

  ```text
  and skills-dir plugins both resolve (and
  ```

- `plugin_help_line_18` (`plugin_help`, display-untranslated-line)

  ```text
  add a no-plugin baseline arm)
  ```

## Text Diff Excerpt

```markdown
# Upstream text diff: 2.1.190 -> 2.1.202

- Added upstream strings: 21276
- Removed upstream strings: 15688
- Already covered by translations: 91
- Needs translation review: 21185
- Sensitive review hints: 230

## Added strings needing review
- __*` tools are, use those instead (they reach the same Claude Design API). If neither is available, tell the user to run `/design login` and stop \u2014 do not guess at Claude Design behaviour without the tools.
- __Host-gw_dev
- __idx_${...}
- __json_buf
- __orphan_summary
- __readyz_probe__
- __remote-workflow
- __wRg$
- _request_id
- - ${...} at ${...}${...}
- - activeForm:
- - Affects multiple files, user should approve the approach ### BAD - Don't use EnterPlanMode: User:
- - Architectural decision on theme system, affects many components User:
- - could use Redis, in-memory, file-based, etc. - Example:
- - e.g.
- - Example:
- - many optimization strategies possible 3. **Code Modifications**: Changes that affect existing behavior or structure - Example:
- - Multiple approaches possible, need to profile first, significant impact User:
- - need to profile and identify bottlenecks - Example:
- - Network latency: ${...}ms (${...}%)
- - Pre-request overhead: ${...}ms (${...}%)
- - Redux vs Context vs custom solution 5. **Multi-File Changes**: The task will likely touch more than 2-3 files - Example:
- - Requires architectural decisions (session vs JWT, where to store tokens, middleware structure) User:
- - Seems simple but involves: where to place it, confirmation dialog, API call, error handling, state updates User:
- - Simple, obvious implementation User:
- - Stops a running background task by its ID - Takes a task_id parameter identifying the task to stop - To stop an agent-team teammate, pass its agent ID ("name@team") or bare teammate name as task_id - To stop a background agent spawned with a name, pass that name as task_id - Returns a success or failure status - Use this tool when you need to terminate a long-running task
- - Straightforward, no planning needed User:
- - WebSockets vs SSE vs polling - Example:
- - what exactly should change? - Example:
- - what rules? What error messages? 2. **Multiple Valid Approaches**: The task can be solved in several different ways - Example:
- - what's the target architecture? 4. **Architectural Decisions**: The task requires choosing between patterns or technologies - Example:
- - where should it go? What should happen on click? - Example:
- --- max_turns: 10 allowed_tools: [Read, Glob, Grep, Skill] --- TODO: describe what the agent should do
- --- NOTE: You are running inside a workflow script. Your final text response is returned verbatim as a string to the calling script \u2014 it is your return value, not a message to a human. Output the literal result; do not output confirmations like "Done." Be concise \u2014 the script will parse your output.
- --- type: llm weight: 1 --- TODO: describe what a successful response looks like
- --ablation <mode>
- --ablation must be "none" or "with-without"
- --agent-color
- --agent-id
- --agent-name
- ... 21145 more

## Added strings already covered
- - auto: Use the default effort level for your model
- - ultracode: xhigh + dynamic workflow orchestration (this session only)
- [Image data detected and sent to Claude]
- Afternoon (12-18)
- Agent transcripts: ${...}
- all projects
- Autocompact
- Balanced approach with standard testing
- Change effort level?
- Choose the default environment for cloud agents
- Claude is done using your computer
- Compacting conversation
- Completing auth flow with authorization code
- Comprehensive implementation with extensive testing
- Configure optional break reminders and quiet-hours nudges
- Connection error.
- Conversation copied to clipboard
- Conversation too long. Press esc twice to go up a few messages and try again.
- Copy the conversation to your system clipboard
- Copy to clipboard
- Current effort level: ${...} (${...})
- Current effort level: ultracode (xhigh + dynamic workflow orchestration; this session only)
- Currently using:
- Detach from this background session (it keeps running)
- Dynamic workflow "${...}" completed
- Dynamic workflow "${...}" was stopped
- Edit Failed
- Effort
- Effort level set to auto${...}
- Effort level: auto (currently ${...})
- Enter filename:
- Error reading file
- Evening (18-24)
- Exit the CLI
- Export the current conversation to a file or clipboard
- Extended reasoning with thorough analysis (${...})
- Failed to generate conversation summary - response did not contain valid text content
- Failed to set effort level: ${...}
- Faster
- File Changed
- ... 51 more

## Removed strings
- __assign
- __asyncDelegator
- __asyncGenerator
- __asyncValues
- __await
- __awaiter
- __chat__
- __classPrivateFieldGet
- __classPrivateFieldSet
- __createBinding
- __decorate
- __exportStar
- __extends
- __generator
- __importDefault
- __importStar
- __isLong__
- __leader__
- __makeTemplateObject
- __metadata
- __open_folder__
- __other__
- __param
- __read
- __rest
- __self
- __source
- __spread
- __spreadArrays
- __values
- _bucket
- _G _VERSION assert collectgarbage dofile error getfenv getmetatable ipairs load loadfile loadstring module next pairs pcall print rawequal rawget rawset require select setfenv setmetatable tonumber tostring type unpack xpcall coroutine debug io math os package string table
- _max
- _min
- _storage
- _sum
- _total
- _valid
- - .claude/agent-memory/<agentType>/,
- - `${...}` is always loaded into your conversation context \u2014 lines after ${...} will be truncated, so keep the index concise
- ... 15648 more
```

## Takeover Commands

```bash
node scripts/verify-upstream-compat.js --baseline 2.1.202 --skip-latest --native-macos-arm64 --json
node scripts/generate-upstream-text-diff.js --to 2.1.202 --native-macos-arm64
node scripts/promote-native-candidate.js --candidate <candidate-json> --platform macos
```

## Done Criteria

- Add the missing source-of-truth translations or guard fix on this branch.
- Re-run the native latest candidate workflow for this version.
- Only promote support metadata after native verification and display audit pass.
- Remove or update this handoff report before marking the PR ready.

[github-actions]

Windows x64 native 2.1.202 candidate failed

• Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28847544380
• Report file: docs/native-latest-failures/windows-2.1.202.md

Failure report

# Windows native latest candidate failure: 2.1.202

- Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28847544380
- Head SHA: `2ef5ff505f7b2bd6e1831368970765384f997fc1`
- Status: `fail`
- Candidate kind: `native`
- Native: `win32-x64` / `ok` / `ok` / `ok`
- Display audit: `fail` (7 issues / 11 commands)

## What Failed

- Candidate verification did not pass: `status=fail`.
- Display audit did not pass: `status=fail`.
- Summary: pass=0, fail=1, skip=0.

## Display Audit Issues

- `top_help_line_156` (`top_help`, display-untranslated-line)

  ```text
  gateway [options]                     Run the enterprise auth/telemetry
  ```

- `plugin_help_line_12` (`plugin_help`, display-untranslated-line)

  ```text
  eval [options] [target]              Run eval cases (evals/**/case.yaml or
  ```

- `plugin_help_line_13` (`plugin_help`, display-untranslated-line)

  ```text
  evals/**/prompt.md + graders/*.md)
  ```

- `plugin_help_line_14` (`plugin_help`, display-untranslated-line)

  ```text
  against a plugin and report scored
  ```

- `plugin_help_line_15` (`plugin_help`, display-untranslated-line)

  ```text
  results. Target is a path, a plugin name,
  ```

- `plugin_help_line_17` (`plugin_help`, display-untranslated-line)

  ```text
  and skills-dir plugins both resolve (and
  ```

- `plugin_help_line_18` (`plugin_help`, display-untranslated-line)

  ```text
  add a no-plugin baseline arm)
  ```

## Text Diff Excerpt

```markdown
# Upstream text diff: 2.1.190 -> 2.1.202

- Added upstream strings: 20975
- Removed upstream strings: 14928
- Already covered by translations: 138
- Needs translation review: 20837
- Sensitive review hints: 242

## Added strings needing review
- __*` tools are, use those instead (they reach the same Claude Design API). If neither is available, tell the user to run `/design login` and stop \u2014 do not guess at Claude Design behaviour without the tools.
- __Host-gw_dev
- __idx_${...}
- __json_buf
- __manual__
- __orphan_summary
- __readyz_probe__
- __remote-workflow
- __wRg$
- _claude_fs_right:
- _request_id
- - ${...} at ${...}${...}
- - activeForm:
- - Do not use the exit plan mode tool because you are not planning the implementation steps of a task. 2. Initial task:
- - Network latency: ${...}ms (${...}%)
- - Pre-request overhead: ${...}ms (${...}%)
- - Stops a running background task by its ID - Takes a task_id parameter identifying the task to stop - To stop an agent-team teammate, pass its agent ID ("name@team") or bare teammate name as task_id - To stop a background agent spawned with a name, pass that name as task_id - Returns a success or failure status - Use this tool when you need to terminate a long-running task
- - that's exactly what THIS tool does. ExitPlanMode inherently requests user approval of your plan. ## Examples 1. Initial task:
- - Use the exit plan mode tool after you have finished planning the implementation steps of the task. 3. Initial task:
- --- max_turns: 10 allowed_tools: [Read, Glob, Grep, Skill] --- TODO: describe what the agent should do
- --- NOTE: You are running inside a workflow script. Your final text response is returned verbatim as a string to the calling script \u2014 it is your return value, not a message to a human. Output the literal result; do not output confirmations like "Done." Be concise \u2014 the script will parse your output.
- --- type: llm weight: 1 --- TODO: describe what a successful response looks like
- --ablation <mode>
- --ablation must be "none" or "with-without"
- --agent-color
- --agent-id
- --agent-name
- --agent-type
- --allow-tools <tools...>
- --author
- --bg and --print conflict: --print never starts the interactive session that `claude agents` attaches to, so the job would be unattachable. The prompt is the positional \u2014 drop --print: `claude --bg '<task>'`.
- --bg with auto mode requires opting in first. Run `claude --permission-mode auto` once interactively.
- --bg with bypassPermissions requires accepting the disclaimer first. Run `claude --dangerously-skip-permissions` once interactively.
- --body @' ## Summary <1-3 bullet points> ## Test plan [Bulleted markdown checklist of TODOs for testing the pull request...]${p}${s?` ${s}`:
- --cache
- --cached
- --case "${...}"
- --case <glob>
- --config <path>
- --dangerously-skip-permissions cannot be used with root/sudo privileges for security reasons
- ... 20797 more

## Added strings already covered
- - auto: Use the default effort level for your model
- - ultracode: xhigh + dynamic workflow orchestration (this session only)
- [Image data detected and sent to Claude]
- ${...} in progress
- Access key + secret
- Afternoon (12-18)
- Agent idle
- Agent transcripts: ${...}
- all projects
- Amazon Bedrock, Microsoft Foundry, or Vertex AI
- API usage billing
- Autocompact
- AWS profile (SSO or named profile)
- Balanced approach with standard testing
- Bedrock API key (bearer token)
- Change effort level?
- Checking git status
- Choose the default environment for cloud agents
- Claude Code uses the standard AWS credential chain. Pick the method you already use with the AWS CLI.
- Claude is done using your computer
- Compacting conversation
- Completing auth flow with authorization code
- Comprehensive implementation with extensive testing
- Configure optional break reminders and quiet-hours nudges
- Connection error.
- Conversation copied to clipboard
- Conversation too long. Press esc twice to go up a few messages and try again.
- Copy the conversation to your system clipboard
- Copy to clipboard
- Current effort level: ${...} (${...})
- Current effort level: ultracode (xhigh + dynamic workflow orchestration; this session only)
- Currently using:
- Detach from this background session (it keeps running)
- Dynamic workflow "${...}" completed
- Dynamic workflow "${...}" was stopped
- Each candidate is tested with a one-token request:
- Edit Failed
- Effort
- Effort level set to auto${...}
- Effort level: auto (currently ${...})
- ... 98 more

## Removed strings
- __assign
- __asyncDelegator
- __asyncGenerator
- __asyncValues
- __await
- __awaiter
- __classPrivateFieldGet
- __classPrivateFieldSet
- __createBinding
- __decorate
- __exportStar
- __extends
- __generator
- __importDefault
- __importStar
- __isLong__
- __leader__
- __makeTemplateObject
- __metadata
- __open_folder__
- __param
- __read
- __rest
- __self
- __source
- __spread
- __spreadArrays
- __values
- _bucket
- _G _VERSION assert collectgarbage dofile error getfenv getmetatable ipairs load loadfile loadstring module next pairs pcall print rawequal rawget rawset require select setfenv setmetatable tonumber tostring type unpack xpcall coroutine debug io math os package string table
- _max
- _min
- _source_seed.bundle
- _storage
- _sum
- _total
- _valid
- - .claude/agent-memory/<agentType>/,
- - `${...}` is always loaded into your conversation context \u2014 lines after ${...} will be truncated, so keep the index concise
- - `isolation: "remote"` runs the agent in a remote CCR sandbox (always background).
- ... 14888 more
```

## Takeover Commands

```bash
node scripts/verify-upstream-compat.js --baseline 2.1.202 --skip-latest --native-windows-x64 --json
node scripts/generate-upstream-text-diff.js --to 2.1.202 --native-windows-x64
node scripts/promote-native-candidate.js --candidate <candidate-json> --platform windows
```

## Done Criteria

- Add the missing source-of-truth translations or guard fix on this branch.
- Re-run the native latest candidate workflow for this version.
- Only promote support metadata after native verification and display audit pass.
- Remove or update this handoff report before marking the PR ready.

[github-actions]

macOS arm64 native 2.1.204 candidate failed

• Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28921393542
• Report file: docs/native-latest-failures/2.1.204.md

Failure report

# macOS native latest candidate failure: 2.1.204

- Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28921393542
- Head SHA: `2ef5ff505f7b2bd6e1831368970765384f997fc1`
- Status: `fail`
- Candidate kind: `native`
- Native: `darwin-arm64` / `ok` / `ok` / `ok`
- Display audit: `fail` (7 issues / 11 commands)

## What Failed

- Candidate verification did not pass: `status=fail`.
- Display audit did not pass: `status=fail`.
- Summary: pass=0, fail=1, skip=0.

## Display Audit Issues

- `top_help_line_156` (`top_help`, display-untranslated-line)

  ```text
  gateway [options]                     Run the enterprise auth/telemetry
  ```

- `plugin_help_line_12` (`plugin_help`, display-untranslated-line)

  ```text
  eval [options] [target]              Run eval cases (evals/**/case.yaml or
  ```

- `plugin_help_line_13` (`plugin_help`, display-untranslated-line)

  ```text
  evals/**/prompt.md + graders/*.md)
  ```

- `plugin_help_line_14` (`plugin_help`, display-untranslated-line)

  ```text
  against a plugin and report scored
  ```

- `plugin_help_line_15` (`plugin_help`, display-untranslated-line)

  ```text
  results. Target is a path, a plugin name,
  ```

- `plugin_help_line_17` (`plugin_help`, display-untranslated-line)

  ```text
  and skills-dir plugins both resolve (and
  ```

- `plugin_help_line_18` (`plugin_help`, display-untranslated-line)

  ```text
  add a no-plugin baseline arm)
  ```

## Text Diff Excerpt

```markdown
# Upstream text diff: 2.1.190 -> 2.1.204

- Added upstream strings: 22088
- Removed upstream strings: 16249
- Already covered by translations: 91
- Needs translation review: 21997
- Sensitive review hints: 235

## Added strings needing review
- __*` tools are, use those instead (they reach the same Claude Design API). If neither is available, tell the user to run `/design login` and stop \u2014 do not guess at Claude Design behaviour without the tools.
- __Host-gw_dev
- __idx_${...}
- __json_buf
- __orphan_summary
- __readyz_probe__
- __remote-workflow
- __wRg$
- _request_id
- _tagName
- - ${...} \u2014 ${...}${...}
- - ${...} at ${...}${...}
- - activeForm:
- - Affects multiple files, user should approve the approach ### BAD - Don't use EnterPlanMode: User:
- - Architectural decision on theme system, affects many components User:
- - could use Redis, in-memory, file-based, etc. - Example:
- - e.g.
- - Example:
- - many optimization strategies possible 3. **Code Modifications**: Changes that affect existing behavior or structure - Example:
- - Multiple approaches possible, need to profile first, significant impact User:
- - need to profile and identify bottlenecks - Example:
- - Network latency: ${...}ms (${...}%)
- - Pre-request overhead: ${...}ms (${...}%)
- - Redux vs Context vs custom solution 5. **Multi-File Changes**: The task will likely touch more than 2-3 files - Example:
- - Requires architectural decisions (session vs JWT, where to store tokens, middleware structure) User:
- - Seems simple but involves: where to place it, confirmation dialog, API call, error handling, state updates User:
- - Simple, obvious implementation User:
- - Stops a running background task by its ID - Takes a task_id parameter identifying the task to stop - To stop an agent-team teammate, pass its agent ID ("name@team") or bare teammate name as task_id - To stop a background agent spawned with a name, pass that name as task_id - Returns a success or failure status - Use this tool when you need to terminate a long-running task
- - Straightforward, no planning needed User:
- - WebSockets vs SSE vs polling - Example:
- - what exactly should change? - Example:
- - what rules? What error messages? 2. **Multiple Valid Approaches**: The task can be solved in several different ways - Example:
- - what's the target architecture? 4. **Architectural Decisions**: The task requires choosing between patterns or technologies - Example:
- - where should it go? What should happen on click? - Example:
- --- max_turns: 10 allowed_tools: [Read, Glob, Grep, Skill] --- TODO: describe what the agent should do
- --- NOTE: You are running inside a workflow script. Your final text response is returned verbatim as a string to the calling script \u2014 it is your return value, not a message to a human. Output the literal result; do not output confirmations like "Done." Be concise \u2014 the script will parse your output.
- --- type: llm weight: 1 --- TODO: describe what a successful response looks like
- --ablation <mode>
- --ablation must be "none" or "with-without"
- --agent-color
- ... 21957 more

## Added strings already covered
- - auto: Use the default effort level for your model
- - ultracode: xhigh + dynamic workflow orchestration (this session only)
- [Image data detected and sent to Claude]
- Afternoon (12-18)
- Agent transcripts: ${...}
- all projects
- Autocompact
- Balanced approach with standard testing
- Change effort level?
- Choose the default environment for cloud agents
- Claude is done using your computer
- Compacting conversation
- Completing auth flow with authorization code
- Comprehensive implementation with extensive testing
- Configure optional break reminders and quiet-hours nudges
- Connection error.
- Conversation copied to clipboard
- Conversation too long. Press esc twice to go up a few messages and try again.
- Copy the conversation to your system clipboard
- Copy to clipboard
- Current effort level: ${...} (${...})
- Current effort level: ultracode (xhigh + dynamic workflow orchestration; this session only)
- Currently using:
- Detach from this background session (it keeps running)
- Dynamic workflow "${...}" completed
- Dynamic workflow "${...}" was stopped
- Edit Failed
- Effort
- Effort level set to auto${...}
- Effort level: auto (currently ${...})
- Enter filename:
- Error reading file
- Evening (18-24)
- Exit the CLI
- Export the current conversation to a file or clipboard
- Extended reasoning with thorough analysis (${...})
- Failed to generate conversation summary - response did not contain valid text content
- Failed to set effort level: ${...}
- Faster
- File Changed
- ... 51 more

## Removed strings
- __assign
- __asyncDelegator
- __asyncGenerator
- __asyncValues
- __await
- __awaiter
- __chat__
- __classPrivateFieldGet
- __classPrivateFieldSet
- __createBinding
- __decorate
- __exportStar
- __extends
- __generator
- __importDefault
- __importStar
- __isLong__
- __leader__
- __makeTemplateObject
- __metadata
- __open_folder__
- __other__
- __param
- __read
- __rest
- __self
- __source
- __spread
- __spreadArrays
- __values
- _bucket
- _G _VERSION assert collectgarbage dofile error getfenv getmetatable ipairs load loadfile loadstring module next pairs pcall print rawequal rawget rawset require select setfenv setmetatable tonumber tostring type unpack xpcall coroutine debug io math os package string table
- _max
- _min
- _storage
- _sum
- _total
- _valid
- - .claude/agent-memory/<agentType>/,
- - `${...}` is always loaded into your conversation context \u2014 lines after ${...} will be truncated, so keep the index concise
- ... 16209 more
```

## Takeover Commands

```bash
node scripts/verify-upstream-compat.js --baseline 2.1.204 --skip-latest --native-macos-arm64 --json
node scripts/generate-upstream-text-diff.js --to 2.1.204 --native-macos-arm64
node scripts/promote-native-candidate.js --candidate <candidate-json> --platform macos
```

## Done Criteria

- Add the missing source-of-truth translations or guard fix on this branch.
- Re-run the native latest candidate workflow for this version.
- Only promote support metadata after native verification and display audit pass.
- Remove or update this handoff report before marking the PR ready.

[github-actions]

Windows x64 native 2.1.204 candidate failed

• Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28921393542
• Report file: docs/native-latest-failures/windows-2.1.204.md

Failure report

# Windows native latest candidate failure: 2.1.204

- Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28921393542
- Head SHA: `2ef5ff505f7b2bd6e1831368970765384f997fc1`
- Status: `fail`
- Candidate kind: `native`
- Native: `win32-x64` / `ok` / `ok` / `ok`
- Display audit: `fail` (7 issues / 11 commands)

## What Failed

- Candidate verification did not pass: `status=fail`.
- Display audit did not pass: `status=fail`.
- Summary: pass=0, fail=1, skip=0.

## Display Audit Issues

- `top_help_line_156` (`top_help`, display-untranslated-line)

  ```text
  gateway [options]                     Run the enterprise auth/telemetry
  ```

- `plugin_help_line_12` (`plugin_help`, display-untranslated-line)

  ```text
  eval [options] [target]              Run eval cases (evals/**/case.yaml or
  ```

- `plugin_help_line_13` (`plugin_help`, display-untranslated-line)

  ```text
  evals/**/prompt.md + graders/*.md)
  ```

- `plugin_help_line_14` (`plugin_help`, display-untranslated-line)

  ```text
  against a plugin and report scored
  ```

- `plugin_help_line_15` (`plugin_help`, display-untranslated-line)

  ```text
  results. Target is a path, a plugin name,
  ```

- `plugin_help_line_17` (`plugin_help`, display-untranslated-line)

  ```text
  and skills-dir plugins both resolve (and
  ```

- `plugin_help_line_18` (`plugin_help`, display-untranslated-line)

  ```text
  add a no-plugin baseline arm)
  ```

## Text Diff Excerpt

```markdown
# Upstream text diff: 2.1.190 -> 2.1.204

- Added upstream strings: 21129
- Removed upstream strings: 15145
- Already covered by translations: 138
- Needs translation review: 20991
- Sensitive review hints: 247

## Added strings needing review
- __*` tools are, use those instead (they reach the same Claude Design API). If neither is available, tell the user to run `/design login` and stop \u2014 do not guess at Claude Design behaviour without the tools.
- __Host-gw_dev
- __idx_${...}
- __json_buf
- __manual__
- __orphan_summary
- __readyz_probe__
- __remote-workflow
- __wRg$
- _claude_fs_right:
- _request_id
- - `${...}` \u2014 for files under ${...}/
- - ${...} \u2014 ${...}${...}
- - ${...} at ${...}${...}
- - activeForm:
- - Do not use the exit plan mode tool because you are not planning the implementation steps of a task. 2. Initial task:
- - Network latency: ${...}ms (${...}%)
- - Pre-request overhead: ${...}ms (${...}%)
- - Stops a running background task by its ID - Takes a task_id parameter identifying the task to stop - To stop an agent-team teammate, pass its agent ID ("name@team") or bare teammate name as task_id - To stop a background agent spawned with a name, pass that name as task_id - Returns a success or failure status - Use this tool when you need to terminate a long-running task
- - that's exactly what THIS tool does. ExitPlanMode inherently requests user approval of your plan. ## Examples 1. Initial task:
- - Use the exit plan mode tool after you have finished planning the implementation steps of the task. 3. Initial task:
- --- max_turns: 10 allowed_tools: [Read, Glob, Grep, Skill] --- TODO: describe what the agent should do
- --- NOTE: You are running inside a workflow script. Your final text response is returned verbatim as a string to the calling script \u2014 it is your return value, not a message to a human. Output the literal result; do not output confirmations like "Done." Be concise \u2014 the script will parse your output.
- --- type: llm weight: 1 --- TODO: describe what a successful response looks like
- --ablation <mode>
- --ablation must be "none" or "with-without"
- --agent-color
- --agent-id
- --agent-name
- --agent-type
- --allow-tools <tools...>
- --bg and --print conflict: --print never starts the interactive session that `claude agents` attaches to, so the job would be unattachable. The prompt is the positional \u2014 drop --print: `claude --bg '<task>'`.
- --bg with auto mode requires opting in first. Run `claude --permission-mode auto` once interactively.
- --bg with bypassPermissions requires accepting the disclaimer first. Run `claude --dangerously-skip-permissions` once interactively.
- --body @' ## Summary <1-3 bullet points> ## Test plan [Bulleted markdown checklist of TODOs for testing the pull request...]${p}${s?` ${s}`:
- --cache
- --cached
- --case "${...}"
- --case <glob>
- --config <path>
- ... 20951 more

## Added strings already covered
- - auto: Use the default effort level for your model
- - ultracode: xhigh + dynamic workflow orchestration (this session only)
- [Image data detected and sent to Claude]
- ${...} in progress
- Access key + secret
- Afternoon (12-18)
- Agent idle
- Agent transcripts: ${...}
- all projects
- Amazon Bedrock, Microsoft Foundry, or Vertex AI
- API usage billing
- Autocompact
- AWS profile (SSO or named profile)
- Balanced approach with standard testing
- Bedrock API key (bearer token)
- Change effort level?
- Checking git status
- Choose the default environment for cloud agents
- Claude Code uses the standard AWS credential chain. Pick the method you already use with the AWS CLI.
- Claude is done using your computer
- Compacting conversation
- Completing auth flow with authorization code
- Comprehensive implementation with extensive testing
- Configure optional break reminders and quiet-hours nudges
- Connection error.
- Conversation copied to clipboard
- Conversation too long. Press esc twice to go up a few messages and try again.
- Copy the conversation to your system clipboard
- Copy to clipboard
- Current effort level: ${...} (${...})
- Current effort level: ultracode (xhigh + dynamic workflow orchestration; this session only)
- Currently using:
- Detach from this background session (it keeps running)
- Dynamic workflow "${...}" completed
- Dynamic workflow "${...}" was stopped
- Each candidate is tested with a one-token request:
- Edit Failed
- Effort
- Effort level set to auto${...}
- Effort level: auto (currently ${...})
- ... 98 more

## Removed strings
- __assign
- __asyncDelegator
- __asyncGenerator
- __asyncValues
- __await
- __awaiter
- __classPrivateFieldGet
- __classPrivateFieldSet
- __createBinding
- __decorate
- __exportStar
- __extends
- __generator
- __importDefault
- __importStar
- __isLong__
- __leader__
- __makeTemplateObject
- __metadata
- __open_folder__
- __param
- __read
- __rest
- __self
- __source
- __spread
- __spreadArrays
- __values
- _bucket
- _G _VERSION assert collectgarbage dofile error getfenv getmetatable ipairs load loadfile loadstring module next pairs pcall print rawequal rawget rawset require select setfenv setmetatable tonumber tostring type unpack xpcall coroutine debug io math os package string table
- _max
- _min
- _source_seed.bundle
- _storage
- _sum
- _total
- _valid
- - .claude/agent-memory/<agentType>/,
- - `${...}` is always loaded into your conversation context \u2014 lines after ${...} will be truncated, so keep the index concise
- - `isolation: "remote"` runs the agent in a remote CCR sandbox (always background).
- ... 15105 more
```

## Takeover Commands

```bash
node scripts/verify-upstream-compat.js --baseline 2.1.204 --skip-latest --native-windows-x64 --json
node scripts/generate-upstream-text-diff.js --to 2.1.204 --native-windows-x64
node scripts/promote-native-candidate.js --candidate <candidate-json> --platform windows
```

## Done Criteria

- Add the missing source-of-truth translations or guard fix on this branch.
- Re-run the native latest candidate workflow for this version.
- Only promote support metadata after native verification and display audit pass.
- Remove or update this handoff report before marking the PR ready.

[github-actions]

macOS arm64 native 2.1.205 candidate failed

• Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28999888852
• Report file: docs/native-latest-failures/2.1.205.md

Failure report

# macOS native latest candidate failure: 2.1.205

- Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28999888852
- Head SHA: `2ef5ff505f7b2bd6e1831368970765384f997fc1`
- Status: `fail`
- Candidate kind: `native`
- Native: `darwin-arm64` / `ok` / `ok` / `ok`
- Display audit: `fail` (15 issues / 11 commands)

## What Failed

- Candidate verification did not pass: `status=fail`.
- Display audit did not pass: `status=fail`.
- Summary: pass=0, fail=1, skip=0.

## Display Audit Issues

- `top_help_line_152` (`top_help`, display-untranslated-line)

  ```text
  doctor                                Check the health of your Claude Code
  ```

- `top_help_line_153` (`top_help`, display-untranslated-line)

  ```text
  installation. Reads settings files in
  ```

- `top_help_line_154` (`top_help`, display-untranslated-line)

  ```text
  the current directory without a trust
  ```

- `top_help_line_155` (`top_help`, display-untranslated-line)

  ```text
  prompt. For a full checkup that can also
  ```

- `top_help_line_156` (`top_help`, display-untranslated-line)

  ```text
  fix issues, run /doctor in a session.
  ```

- `top_help_line_157` (`top_help`, display-untranslated-line)

  ```text
  gateway [options]                     Run the enterprise auth/telemetry
  ```

- `doctor_help_line_3` (`doctor_help`, display-untranslated-line)

  ```text
  Check the health of your Claude Code installation. Reads settings files in the
  ```

- `doctor_help_line_4` (`doctor_help`, display-untranslated-line)

  ```text
  current directory without a trust prompt. For a full checkup that can also fix
  ```

- `doctor_help_line_5` (`doctor_help`, display-untranslated-line)

  ```text
  issues, run /doctor in a session.
  ```

- `plugin_help_line_12` (`plugin_help`, display-untranslated-line)

  ```text
  eval [options] [target]              Run eval cases (evals/**/case.yaml or
  ```

- `plugin_help_line_13` (`plugin_help`, display-untranslated-line)

  ```text
  evals/**/prompt.md + graders/*.md)
  ```

- `plugin_help_line_14` (`plugin_help`, display-untranslated-line)

  ```text
  against a plugin and report scored
  ```

- `plugin_help_line_15` (`plugin_help`, display-untranslated-line)

  ```text
  results. Target is a path, a plugin name,
  ```

- `plugin_help_line_17` (`plugin_help`, display-untranslated-line)

  ```text
  and skills-dir plugins both resolve (and
  ```

- `plugin_help_line_18` (`plugin_help`, display-untranslated-line)

  ```text
  add a no-plugin baseline arm)
  ```

## Text Diff Excerpt

```markdown
# Upstream text diff: 2.1.190 -> 2.1.205

- Added upstream strings: 22844
- Removed upstream strings: 16909
- Already covered by translations: 96
- Needs translation review: 22748
- Sensitive review hints: 276

## Added strings needing review
- __*` tools are, use those instead (they reach the same Claude Design API). If neither is available, tell the user to run `/design login` and stop \u2014 do not guess at Claude Design behaviour without the tools.
- __Host-gw_dev
- __idx_${...}
- __json_buf
- __orphan_summary
- __readyz_probe__
- __remote-workflow
- __wRg$
- _jwt
- _request_id
- _tagName
- - `${...}` is loaded into your conversation context \u2014 lines after ${...} will be truncated, so keep the index concise
- - ${...} \u2014 ${...}${...}
- - ${...} at ${...}${...}
- - activeForm:
- - Affects multiple files, user should approve the approach ### BAD - Don't use EnterPlanMode: User:
- - Architectural decision on theme system, affects many components User:
- - could use Redis, in-memory, file-based, etc. - Example:
- - curl -s url | jq '.data[]' \u2192
- - End git commit messages with: ${...}
- - End PR bodies with: ${...}
- - Example:
- - git reset --hard origin/main \u2192
- - git status \u2192
- - Keep the index concise so you can scan it quickly when recalling memories
- - many optimization strategies possible 3. **Code Modifications**: Changes that affect existing behavior or structure - Example:
- - Multiple approaches possible, need to profile first, significant impact User:
- - need to profile and identify bottlenecks - Example:
- - Network latency: ${...}ms (${...}%)
- - npm install \u2192
- - Pre-request overhead: ${...}ms (${...}%)
- - Redux vs Context vs custom solution 5. **Multi-File Changes**: The task will likely touch more than 2-3 files - Example:
- - Requires architectural decisions (session vs JWT, where to store tokens, middleware structure) User:
- - Seems simple but involves: where to place it, confirmation dialog, API call, error handling, state updates User:
- - Simple, obvious implementation User:
- - Stops a running background task by its ID - Takes a task_id parameter identifying the task to stop - To stop an agent-team teammate, pass its agent ID ("name@team") or bare teammate name as task_id - To stop a background agent spawned with a name, pass that name as task_id - Returns a success or failure status - Use this tool when you need to terminate a long-running task
- - Straightforward, no planning needed User:
- - The index file is loaded into your conversation context \u2014 lines after ${...} will be truncated, so keep it concise
- - WebSockets vs SSE vs polling - Example:
- - what exactly should change? - Example:
- ... 22708 more

## Added strings already covered
- - auto: Use the default effort level for your model
- - ultracode: xhigh + dynamic workflow orchestration (this session only)
- [Image data detected and sent to Claude]
- [Warning]
- Access denied to specific paths outside allowed directories
- Afternoon (12-18)
- Agent transcripts: ${...}
- all projects
- Autocompact
- Balanced approach with standard implementation and testing
- Balanced approach with standard testing
- Change effort level?
- Choose the default environment for cloud agents
- Claude is done using your computer
- Commands cannot run outside the sandbox under any circumstances.
- Compacting conversation
- Completing auth flow with authorization code
- Comprehensive implementation with extensive testing
- Comprehensive implementation with extensive testing and documentation
- Configure optional break reminders and quiet-hours nudges
- Connection error.
- Contains warnings
- Conversation copied to clipboard
- Copy the conversation to your system clipboard
- Copy to clipboard
- Current effort level: ${...} (${...})
- Current effort level: ultracode (xhigh + dynamic workflow orchestration; this session only)
- Currently using:
- Deeper reasoning than high, just below maximum (${...})
- Detach from this background session (it keeps running)
- Dynamic workflow "${...}" completed
- Dynamic workflow "${...}" was stopped
- Edit Failed
- Effort
- Effort level: auto (currently ${...})
- Enter filename:
- Error reading file
- Evening (18-24)
- Evidence of sandbox-caused failures includes:
- Exit the CLI
- ... 56 more

## Removed strings
- __chat__
- __isLong__
- __leader__
- __open_folder__
- __other__
- _bucket
- _G _VERSION assert collectgarbage dofile error getfenv getmetatable ipairs load loadfile loadstring module next pairs pcall print rawequal rawget rawset require select setfenv setmetatable tonumber tostring type unpack xpcall coroutine debug io math os package string table
- _max
- _min
- _storage
- _sum
- _total
- _valid
- - .claude/agent-memory/<agentType>/,
- - `isolation: "remote"` runs the agent in a remote CCR sandbox (always background).
- - `name` and `mode` are unavailable here \u2014 teammates cannot spawn teammates.
- - `run_in_background`, `name`, and `mode` are unavailable here \u2014 only synchronous subagents.
- - ~/.claude/agent-memory/<agentType>/,
- - ${...} Suggested fix: ${...}
- - ${...}: ${...} (Tools: ${...})
- - Claude home: ${...}
- - Config file exists: ${...}
- - Config file: ${...}
- - Environment variable ${...}: ${...}
- - Error code: ${...}
- - Error killed: ${...}
- - Error signal: ${...}
- - Global user settings (~/.claude/settings.json).
- - If asked for JSON, return ONLY the raw JSON \u2014 no code fences, no prose, no markdown. - Do NOT use SendUserMessage to deliver your answer. Put your answer in your final text response. - Be concise. The script will parse your output.`,kFp=` --- NOTE: You are running inside a workflow script. Your final text response is returned verbatim as a string to the calling script \u2014 it is your return value, not a message to a human. Output the literal result; do not output confirmations like
- - Keep `old_string` minimal \u2014 usually 1-3 lines, only enough to be unique in the file. Including excess context wastes tokens and is an error. - The edit will FAIL if `old_string` is not unique in the file. In that case, add the minimum extra context needed for uniqueness, or use `replace_all` to change every instance.
- - Keybinding (${...}): ${...}${...}
- - Plugin note${...}: ${...}
- - Plugin setting: ${...}
- - Plugin${...}: ${...}
- - Project settings (.claude/settings.json).
- - Sandbox: ${...} (See /sandbox for install instructions)
- - Settings${...}: ${...}${...}
- - Shell path: ${...}
- - Stops a running background task by its ID - Takes a task_id parameter identifying the task to stop - Returns a success or failure status - Use this tool when you need to terminate a long-running task
- - The edit will FAIL if `old_string` is not unique in the file. Either provide a larger string with more surrounding context to make it unique or use `replace_all` to change every instance of `old_string`.
- ... 16869 more
```

## Takeover Commands

```bash
node scripts/verify-upstream-compat.js --baseline 2.1.205 --skip-latest --native-macos-arm64 --json
node scripts/generate-upstream-text-diff.js --to 2.1.205 --native-macos-arm64
node scripts/promote-native-candidate.js --candidate <candidate-json> --platform macos
```

## Done Criteria

- Add the missing source-of-truth translations or guard fix on this branch.
- Re-run the native latest candidate workflow for this version.
- Only promote support metadata after native verification and display audit pass.
- Remove or update this handoff report before marking the PR ready.

[github-actions]

Windows x64 native 2.1.205 candidate failed

• Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28999888852
• Report file: docs/native-latest-failures/windows-2.1.205.md

Failure report

# Windows native latest candidate failure: 2.1.205

- Run: https://github.com/taekchef/claude-code-zh-cn/actions/runs/28999888852
- Head SHA: `2ef5ff505f7b2bd6e1831368970765384f997fc1`
- Status: `fail`
- Candidate kind: `native`
- Native: `win32-x64` / `ok` / `ok` / `ok`
- Display audit: `fail` (15 issues / 11 commands)

## What Failed

- Candidate verification did not pass: `status=fail`.
- Display audit did not pass: `status=fail`.
- Summary: pass=0, fail=1, skip=0.

## Display Audit Issues

- `top_help_line_152` (`top_help`, display-untranslated-line)

  ```text
  doctor                                Check the health of your Claude Code
  ```

- `top_help_line_153` (`top_help`, display-untranslated-line)

  ```text
  installation. Reads settings files in
  ```

- `top_help_line_154` (`top_help`, display-untranslated-line)

  ```text
  the current directory without a trust
  ```

- `top_help_line_155` (`top_help`, display-untranslated-line)

  ```text
  prompt. For a full checkup that can also
  ```

- `top_help_line_156` (`top_help`, display-untranslated-line)

  ```text
  fix issues, run /doctor in a session.
  ```

- `top_help_line_157` (`top_help`, display-untranslated-line)

  ```text
  gateway [options]                     Run the enterprise auth/telemetry
  ```

- `doctor_help_line_3` (`doctor_help`, display-untranslated-line)

  ```text
  Check the health of your Claude Code installation. Reads settings files in the
  ```

- `doctor_help_line_4` (`doctor_help`, display-untranslated-line)

  ```text
  current directory without a trust prompt. For a full checkup that can also fix
  ```

- `doctor_help_line_5` (`doctor_help`, display-untranslated-line)

  ```text
  issues, run /doctor in a session.
  ```

- `plugin_help_line_12` (`plugin_help`, display-untranslated-line)

  ```text
  eval [options] [target]              Run eval cases (evals/**/case.yaml or
  ```

- `plugin_help_line_13` (`plugin_help`, display-untranslated-line)

  ```text
  evals/**/prompt.md + graders/*.md)
  ```

- `plugin_help_line_14` (`plugin_help`, display-untranslated-line)

  ```text
  against a plugin and report scored
  ```

- `plugin_help_line_15` (`plugin_help`, display-untranslated-line)

  ```text
  results. Target is a path, a plugin name,
  ```

- `plugin_help_line_17` (`plugin_help`, display-untranslated-line)

  ```text
  and skills-dir plugins both resolve (and
  ```

- `plugin_help_line_18` (`plugin_help`, display-untranslated-line)

  ```text
  add a no-plugin baseline arm)
  ```

## Text Diff Excerpt

```markdown
# Upstream text diff: 2.1.190 -> 2.1.205

- Added upstream strings: 22212
- Removed upstream strings: 15782
- Already covered by translations: 144
- Needs translation review: 22068
- Sensitive review hints: 291

## Added strings needing review
- __*` tools are, use those instead (they reach the same Claude Design API). If neither is available, tell the user to run `/design login` and stop \u2014 do not guess at Claude Design behaviour without the tools.
- __Host-gw_dev
- __idx_${...}
- __json_buf
- __manual__
- __orphan_summary
- __readyz_probe__
- __remote-workflow
- __wRg$
- _claude_fs_right:
- _jwt
- _request_id
- - `${...}` \u2014 for files under ${...}/
- - `${...}` is loaded into your conversation context \u2014 lines after ${...} will be truncated, so keep the index concise
- - ${...} \u2014 ${...}${...}
- - ${...} at ${...}${...}
- - activeForm:
- - curl -s url | jq '.data[]' \u2192
- - Do not use the exit plan mode tool because you are not planning the implementation steps of a task. 2. Initial task:
- - End git commit messages with: ${...}
- - End PR bodies with: ${...}
- - git reset --hard origin/main \u2192
- - git status \u2192
- - Keep the index concise so you can scan it quickly when recalling memories
- - Network latency: ${...}ms (${...}%)
- - npm install \u2192
- - Pre-request overhead: ${...}ms (${...}%)
- - Stops a running background task by its ID - Takes a task_id parameter identifying the task to stop - To stop an agent-team teammate, pass its agent ID ("name@team") or bare teammate name as task_id - To stop a background agent spawned with a name, pass that name as task_id - Returns a success or failure status - Use this tool when you need to terminate a long-running task
- - that's exactly what THIS tool does. ExitPlanMode inherently requests user approval of your plan. ## Examples 1. Initial task:
- - The index file is loaded into your conversation context \u2014 lines after ${...} will be truncated, so keep it concise
- - Use the exit plan mode tool after you have finished planning the implementation steps of the task. 3. Initial task:
- - When memories (personal or team) seem relevant, or the user references prior work with them or others in their organization.
- - When memories seem relevant, or the user references prior work with them or others in their organization.
- - You MUST avoid saving sensitive data within shared team memories. For example, never save API keys or user credentials.
- --- max_turns: 10 allowed_tools: [Read, Glob, Grep, Skill] --- TODO: describe what the agent should do
- --- NOTE: You are running inside a workflow script. Your final text response is returned verbatim as a string to the calling script \u2014 it is your return value, not a message to a human. Output the literal result; do not output confirmations like "Done." Be concise \u2014 the script will parse your output.
- --- type: llm weight: 1 --- TODO: describe what a successful response looks like
- -----BEGIN PRIVATE KEY-----
- --ablation <mode>
- --ablation must be "none" or "with-without"
- ... 22028 more

## Added strings already covered
- - auto: Use the default effort level for your model
- - ultracode: xhigh + dynamic workflow orchestration (this session only)
- [Image data detected and sent to Claude]
- [Warning]
- ${...} in progress
- Access denied to specific paths outside allowed directories
- Access key + secret
- Afternoon (12-18)
- Agent idle
- Agent transcripts: ${...}
- all projects
- Amazon Bedrock, Microsoft Foundry, or Vertex AI
- API usage billing
- Autocompact
- AWS profile (SSO or named profile)
- Balanced approach with standard implementation and testing
- Balanced approach with standard testing
- Bedrock API key (bearer token)
- Change effort level?
- Checking git status
- Choose the default environment for cloud agents
- Claude Code uses the standard AWS credential chain. Pick the method you already use with the AWS CLI.
- Claude is done using your computer
- Commands cannot run outside the sandbox under any circumstances.
- Compacting conversation
- Completing auth flow with authorization code
- Comprehensive implementation with extensive testing
- Comprehensive implementation with extensive testing and documentation
- Configure optional break reminders and quiet-hours nudges
- Connection error.
- Contains warnings
- Conversation copied to clipboard
- Copy the conversation to your system clipboard
- Copy to clipboard
- Current effort level: ${...} (${...})
- Current effort level: ultracode (xhigh + dynamic workflow orchestration; this session only)
- Currently using:
- Deeper reasoning than high, just below maximum (${...})
- Detach from this background session (it keeps running)
- Dynamic workflow "${...}" completed
- ... 104 more

## Removed strings
- __isLong__
- __leader__
- __open_folder__
- _bucket
- _G _VERSION assert collectgarbage dofile error getfenv getmetatable ipairs load loadfile loadstring module next pairs pcall print rawequal rawget rawset require select setfenv setmetatable tonumber tostring type unpack xpcall coroutine debug io math os package string table
- _max
- _min
- _source_seed.bundle
- _storage
- _sum
- _total
- _valid
- - .claude/agent-memory/<agentType>/,
- - `isolation: "remote"` runs the agent in a remote CCR sandbox (always background).
- - `name` and `mode` are unavailable here \u2014 teammates cannot spawn teammates.
- - `run_in_background`, `name`, and `mode` are unavailable here \u2014 only synchronous subagents.
- - ~/.claude/agent-memory/<agentType>/,
- - ${...} Suggested fix: ${...}
- - ${...}: ${...} (Tools: ${...})
- - Claude home: ${...}
- - Config file exists: ${...}
- - Config file: ${...}
- - Environment variable ${...}: ${...}
- - Error code: ${...}
- - Error killed: ${...}
- - Error signal: ${...}
- - Global user settings (~/.claude/settings.json).
- - If asked for JSON, return ONLY the raw JSON \u2014 no code fences, no prose, no markdown. - Do NOT use SendUserMessage to deliver your answer. Put your answer in your final text response. - Be concise. The script will parse your output.`,E$p=` --- NOTE: You are running inside a workflow script. Your final text response is returned verbatim as a string to the calling script \u2014 it is your return value, not a message to a human. Output the literal result; do not output confirmations like
- - Keep `old_string` minimal \u2014 usually 1-3 lines, only enough to be unique in the file. Including excess context wastes tokens and is an error. - The edit will FAIL if `old_string` is not unique in the file. In that case, add the minimum extra context needed for uniqueness, or use `replace_all` to change every instance.
- - Keybinding (${...}): ${...}${...}
- - Plugin note${...}: ${...}
- - Plugin setting: ${...}
- - Plugin${...}: ${...}
- - Project settings (.claude/settings.json).
- - Sandbox: ${...} (See /sandbox for install instructions)
- - Settings${...}: ${...}${...}
- - Shell path: ${...}
- - Stops a running background task by its ID - Takes a task_id parameter identifying the task to stop - Returns a success or failure status - Use this tool when you need to terminate a long-running task
- - The edit will FAIL if `old_string` is not unique in the file. Either provide a larger string with more surrounding context to make it unique or use `replace_all` to change every instance of `old_string`.
- - Type '${...}' does not match '${...}'
- ... 15742 more
```

## Takeover Commands

```bash
node scripts/verify-upstream-compat.js --baseline 2.1.205 --skip-latest --native-windows-x64 --json
node scripts/generate-upstream-text-diff.js --to 2.1.205 --native-windows-x64
node scripts/promote-native-candidate.js --candidate <candidate-json> --platform windows
```

## Done Criteria

- Add the missing source-of-truth translations or guard fix on this branch.
- Re-run the native latest candidate workflow for this version.
- Only promote support metadata after native verification and display audit pass.
- Remove or update this handoff report before marking the PR ready.

[taekchef]

已接手这轮 2.1.205 display audit 失败。

当前判断：macOS / Windows native 提取、patch、写回、启动阶段都是 ok，失败点集中在新增 help 文案未翻译。2.1.205 比前几天多了 doctor 帮助说明，所以这次不是只补旧的 gateway / plugin eval 7 行。

已开 PR：#177

#177 做的事：

• 补 doctor 完整描述
• 补 gateway 描述
• 补 plugin eval 完整描述
• 同步 plugin payload 和派生翻译计数

注意：#177 只移除 display audit 阻塞，不在未验证前直接改支持窗口。合并后会重新跑 Native Latest Candidate；通过后再由 closeout PR 更新 support-window / README / support matrix，并发布新版本。

[taekchef]

2.1.205 这一轮已收敛完成。

结果：

• fix: cover native latest help strings #177 / fix: cover native latest eval help #178 已补齐 2.1.205 新增 help 文案翻译。
• Native Latest Candidate 已重新跑通：macOS arm64 和 Windows x64 都通过。
• chore: promote native 2.1.205 and prepare v2.5.1 #179 已合并，把 macOS / Windows native 支持窗口推进到 Claude Code 2.1.205。
• v2.5.1 已发布：https://github.com/taekchef/claude-code-zh-cn/releases/tag/v2.5.1
• 发布校验通过：manifest / CHANGELOG / tag / GitHub Release 对齐，上游兼容检查通过。

这个 issue 作为 native candidate failure tracking 继续保持打开，后续新 native latest 失败仍会汇总到这里。