[dunvu0]

Adds an optional English UI to ARTEX. Chinese remains the default; English is an explicit choice in the language selector.

This update addresses the review feedback on the first version:

• Incomplete coverage: every screen is now localized, including nested dialogs, tooltips, toasts, empty and error states. Covered areas: tasks (all tabs), chat, findings and cases, assets, traffic and evidence, Repeater, schedules, asset sync, workspace, Agents, LLM, MCP, skills, tools, notifications, interception and approvals, logs and settings.
• Dropdown display: the language selector in the sidebar no longer stretches to the full sidebar width. It keeps a compact width on desktop and mobile.

What is translated, what stays verbatim

• Interface labels, placeholders, validation and status labels come from web/messages/{zh-CN,en}.json (4,174 semantic messages each, ICU plurals for counts).
• Known backend error texts (writeErr in server/*.go) and backend-generated system labels are translated at the display boundary. Unknown errors and error details are shown unchanged.
• Bundled Agent and rule names/descriptions are translated only while they still match the seeded Chinese text. Custom or edited names stay as written.
• Prompts, model output, saved reports, raw HTTP and evidence, and stored values are never translated or rewritten.

Agent output language

Switching the selector to English also saves output_language=en in settings (GET/PUT /api/settings). Agents then receive an English output directive at runtime, and the task Report tab uses English headings. Stored prompt versions are not modified, and Chinese (zh) remains the default.

Implementation

• next-intl with semantic keys and a client provider compatible with the static export. No locale routes or middleware.
• The UI language preference is stored in localStorage; missing or invalid values fall back to Chinese.
• Language switches preserve drafts, selections, editor state, streams and timezone/schedule behavior.
• web/I18N.md documents coverage, how to add messages, the checks, and translation attribution (ScopeWeaver, artex-global).

Rebased on current main (d36e1c3), including the new DeepSeek Search API settings and the Jev model-list change.

Validation

• npm run i18n:check: 4,174 messages, matching keys and ICU arguments.
• npm run test:i18n: 22 tests, including drift checks against backend error texts, Agent seeds and system labels.
• npm run build:static and TypeScript: all static pages export.
• npm run test:i18n:browser: 60 Chromium tests against the static export with mocked APIs, including every screen rendered in English without missing messages.
• Go: TestOutputLanguageSettings (PostgreSQL) plus agent and report language tests.

An earlier build was also checked on a live test instance (language selector, English Agent output and report). The rebased build was verified with the automated checks above; the browser tests use mocked APIs.

🤖 Generated with Claude Code

[RuoJi6]

师傅，有几个问题，师傅有空的话修完之后，我在师傅合并pr

• 界面很多地方都未pr完全。
• 下拉框的UI显示有问题。

[dunvu0]

@RuoJi6 Thanks for the review. I've updated the PR to address both points:

• Incomplete coverage: every screen is now translated, including nested dialogs, tooltips, toasts, empty/error states and known backend error messages. Chinese is still the default.
• Dropdown display: the language selector in the sidebar no longer stretches to the full sidebar width. It now has a compact width on both desktop and mobile.

The branch is rebased on the current main (d36e1c3), and the new DeepSeek Search and Jev settings are translated too. Switching to English also makes Agents and the task report write in English. Stored prompts are not modified.

The PR description has the full details and validation steps. The CI workflow needs your approval to run because this PR comes from a fork

🤖 Generated with Claude Code