feat: add Phase 0 for scope and language confirmation

SKILL.md

@@ -22,6 +22,26 @@ When the skill is first triggered and the user hasn't specified a codebase yet,
 
 If the user provides a GitHub link, clone the repo first (`git clone <url> /tmp/<repo-name>`) before starting the analysis. If they say "this codebase" or similar, use the current working directory.
 
+### Phase 0: Scope and Language Confirmation
+
+Before analyzing the codebase, confirm two things:
+
+1. **Scope** — Ask if the user wants the full codebase or a specific area. If the user already specified a clear focus (e.g., "just the agent system"), acknowledge it and lock in that scope. If vague, ask one clarifying question.
+
+2. **Language** — Ask what language the course should be written in. The course content (explanations, code translations, quizzes, animations) should match the learner's preferred language. Do NOT assume English.
+
+Use `AskUserQuestion` with `multiSelect: false` to confirm language with these options:
+- **English** — Course written in English (default if user doesn't specify)
+- **中文** — Course content entirely in Chinese, including code translations, quiz questions, chat animations
+- **Other** — Ask user to specify
+
+> **Important:** Language choice affects all user-facing content: module titles, explanations, code↔plain-translation pairs, quiz questions/options, group chat animations, callout boxes, tooltip definitions, and interactive element labels. Code syntax highlighting and technical identifiers (function names, file paths) should NOT be translated — only the explanatory text around them.
+
+After confirming scope and language, announce clearly:
+> "Got it — full-scope course on the agent system, in **中文**. Let me dig into the code..."
+
+Then proceed to Phase 1.
+
 ## Who This Is For
 
 The target learner is a **"vibe coder"** — someone who builds software by instructing AI coding tools in natural language, without a traditional CS education. They may have built this project themselves (without looking at the code), or they may have found an interesting open-source project on GitHub and want to understand how it's built. Either way, they don't yet understand what's happening under the hood.

references/content-philosophy.md

@@ -2,6 +2,25 @@
 
 > **When to read this:** During Phase 2.5 (writing module briefs) and Phase 3 (writing module HTML). These principles guide every content decision — what to show, how to explain it, and how to test understanding.
 
+### Language-Aware Content
+
+All text in the course must be written in the language confirmed in Phase 0. This includes:
+- Module titles, subtitles, and headings
+- Explanatory paragraphs and screen text
+- Code↔plain translation explanations (left panel: code stays as-is; right panel: plain explanation in target language)
+- Quiz questions, options, and feedback
+- Group chat animation message text and sender labels
+- Callout box content
+- Glossary tooltip definitions
+- Interactive element labels (button text, flow step descriptions)
+
+**What NOT to translate:**
+- Code syntax, function names, variable names, file paths
+- CSS class names, HTML attributes, data values
+- Technical identifiers that must match the actual codebase
+
+The "Code ↔ English Translation" element name is a structural convention — when the course is in Chinese, it renders as "代码 ↔ 中文对照". Same element, same structure, different language.
+
 These principles are what separate a great course from a generic tutorial. They should guide every content decision:
 
 ### Show, Don't Tell — Aggressively Visual