Book of Answers Skill

Core Workflow

1. Normalize the user input and resolve intents in this order: welcome, fallback, disabled-history notice, daily fortune, contextual switch-book, duplicate check, normal question.
2. Load per-user state from SQLite before generating any answer.
3. Reuse last_question only when the input is a switch-book follow-up such as 那用电影版的再测一次.
4. Block identical questions asked within 300 seconds unless the request is a switch-book follow-up.
5. Persist only last_question, last_answer, last_book, and last_timestamp after each successful content response.
6. Default to a random choice between 《电影答案之书》、《文学答案之书》 and 《音乐答案之书》 unless the user explicitly specifies a book.
7. Render every answer with both正文 and 出处：....

Project Files

• Use main.py as the runtime entry point and local CLI for manual testing.
• Use service.py for intent orchestration and response composition.
• Use router.py for three-book routing and switch-book detection.
• Use storage.py for SQLite persistence and state recovery.
• Use data/books.json as the built-in answer corpus for the three themed books.
• Use skill.json and interaction_model.json as the portable metadata and utterance definitions.
• Use README.md as the repository-level setup and usage guide.

Runtime Rules

• Detect fallback input when the message contains no Chinese or alphabetic characters and is effectively digits or symbols only.
• Prefer explicit book selection when the user names 电影、文学 or 音乐 in the same request.
• Keep response formatting stable for normal answers:

我感受到了你的困惑。
正在为你翻开 [book_name]...

它给你的指引是：
「 [答案正文] 」
出处：[出处]

Validation

• Run python3 main.py --user-id demo-user "明天会好吗" for a direct single-turn test.
• Run python3 main.py --user-id demo-user --interactive for a local REPL.
• Run python3 -m unittest discover -s tests to validate key intent and state flows.
• Run python3 /Users/shaozhaoru/.codex/skills/.system/skill-creator/scripts/quick_validate.py /Users/shaozhaoru/Documents/book-of-answers-skill after editing skill metadata.