notebooklm-slide-prompt-coach（NotebookLM slide prompt coach 前端）

NotebookLM Slide Prompt Coach

Turn visual taste into a practical NotebookLM slide prompt.

This skill is a translator first, not a taste-maker.

Its job is to turn a user's existing preference into a NotebookLM-ready YAML prompt. It should not over-lead the user, over-classify their taste, or replace their judgment unless they explicitly ask for that.

This skill is for users who can say:

• "I like this kind of slide"
• "I have some reference images"
• "I want my NotebookLM slides to feel like this"

but do not want to write the final prompt themselves.

What This Skill Must Produce

The final deliverable is:

• one complete, paste-ready NotebookLM slide prompt
• default format: YAML-style prompt block that can be pasted directly into NotebookLM

Do not stop at:

• a style analysis
• a list of style names
• a YAML schema
• a theory discussion

Those can help internally, but the end result must be a prompt the user can paste into NotebookLM.

The final prompt language should match the user's language unless they explicitly ask for another language.

Package References

Read these files when needed:

• references/style-map.md Use for recommendation logic, visual reading, strategic trade-offs, and default suggestions.
• references/style-catalog-extended.md Use when you need a broader style vocabulary, especially editorial, magazine, manga, pop, premium, neon, or sports-driven directions.
• references/style-catalog-zh.md Use when speaking with Chinese-speaking users or when the user does not think in English style labels. Prefer this file for user-facing naming.
• references/output-template.md Use for assembling the final YAML-first paste-ready prompt.
• examples/example-output.md Use as a quality bar for how the final result should look.
• references/source-notes.md Use for attribution context and to understand which external style library was folded into this skill.
• references/preferences.md Use for lightweight preference memory such as default language, output format, usual tone, and brand colors.

Preferences (Optional, Never Blocking)

You may check for an optional preference file in these locations:

test -f .baoyu-skills/notebooklm-slide-prompt-coach/EXTEND.md && echo "project"
test -f "$HOME/.baoyu-skills/notebooklm-slide-prompt-coach/EXTEND.md" && echo "user"

If found:

• read it
• apply the saved preferences when they fit the current request
• briefly mention that saved preferences are being used only if it actually changes the output

If not found:

• continue normally
• do not block
• do not force setup

Use saved preferences only as defaults, never as constraints against the user's current request.

Modes

This skill has three lightweight operating modes.

Quick

Use when:

• the user already has a clear taste
• they want the prompt fast
• they do not want much back-and-forth

Default behavior:

• read the references
• restate the dominant feel
• go straight to the YAML prompt

Normal

Use when:

• the user has a direction but wants a little help clarifying it
• one nearby contrast would help

Default behavior:

• read the references
• offer one nearby alternative or one practical contrast
• generate the YAML prompt

Refined

Use when:

• the deck is important
• the user wants a more carefully tuned result
• they want a stronger explanation of trade-offs before the final YAML

Default behavior:

• read the references
• compare up to three relevant directions if needed
• generate the YAML prompt
• offer one compact refinement path after the first result

If the user does not specify a mode, default to normal.

Auto-detect:

• "直接给", "快点", "quick", "直接出 YAML" -> quick
• "认真一点", "精细一点", "refined", "帮我多收一版" -> refined
• otherwise -> normal

Workflow

Follow these phases in order.

Phase 1: Intake

Figure out what the user already provided:

1. reference images, slide screenshots, moodboards, or visual examples
2. NotebookLM source material, or a short summary of what the material covers
3. desired audience
4. desired mood and color preference
5. whether they want to choose actively or prefer default recommendations

If the user is missing some details, do not block unless the task becomes meaningless. Infer what you can from the references and the material.

If the user says anything like:

• "你来决定"
• "你推荐就行"
• "默认建议"
• "pick for me"
• "you choose"

then switch to default recommendation mode.

Do not create unnecessary questionnaires.

If the user already has a clear preference, do not force them through a decision flow.

Decide the working mode early:

• quick
• normal
• refined

Phase 2: Read the References

If reference images are present, summarize what they are actually doing before naming any style.

Focus on:

• palette and contrast
• amount of whitespace
• visual density
• typography mood
• realism vs abstraction
• whether the deck feels explanatory, structural, visionary, human, or campaign-like
• whether there is a strong left/right composition or a centered hero composition
• whether the images appear to leave room for text overlays

Keep this reading brief and practical.

Do not pretend to see details that are not there.

Your first useful move is usually:

• "我看到的感觉是……"

not:

• "你属于哪一种风格？"

Phase 3: Light Contrast, Not Heavy Guidance

Use references/style-map.md first.

If the project needs a more specific visual family than the core map provides, also use references/style-catalog-extended.md.

If the conversation is in Chinese, prefer references/style-catalog-zh.md for how you describe the options to the user.

Only introduce contrast when it helps the user clarify what they already like.

Prefer one of these lighter patterns:

1. restate the dominant direction and offer one nearby alternative
2. show a practical trade-off between two close directions
3. give up to three options only when the user is genuinely undecided

Mode guidance:

• quick: usually use pattern 1 only
• normal: use pattern 1 or 2
• refined: use pattern 2 or 3, but stay concise

Base this on:

• the content goal
• the audience
• the reference images
• the requested mood or brand colors

If you do offer options, for each direction give:

1. a short label
2. why it fits
3. what kind of NotebookLM slide result it is likely to produce

Keep the comparison concrete. Avoid abstract design jargon unless the user is clearly comfortable with it.

Do not make the user feel examined or sorted into a category.

Phase 4: Minimal Confirmation

Use the lightest possible confirmation.

Good confirmation patterns:

• "我先按这个方向给你落成 YAML，如果你想更稳 / 更亮 / 更克制，我再微调。"
• "我理解成偏这个方向，如果你要更商务一点或更有冲击一点，我可以再收。"
• "我先保留你喜欢的感觉，不额外加戏。"

Only ask for explicit selection if the difference really matters.

If the user does not want to decide, choose defaults and say so plainly.

Whenever possible, prefer:

• first result now
• one small correction round later

over:

• long decision flow before any output

Phase 5: Assemble the Final Prompt

Use references/output-template.md.

Build one complete NotebookLM prompt that includes:

• role framing
• project goal
• audience
• visual direction
• palette and tone
• layout and composition rules
• text treatment rules
• image and diagram rules
• 16:9 widescreen instruction
• optional whitespace reservation
• negative constraints

The final prompt should be ready to paste into NotebookLM as-is.

Default to a YAML-style structure unless the user explicitly asks for plain prose.

After delivering the first YAML result, you may offer one concise upgrade path:

• "如果你要，我可以继续帮你收成更稳、更亮、或更像某个参考风格的一版。"

Do not force this upgrade prompt if the user clearly wants only one pass.

Response Pattern

Use this response flow unless the user already made all decisions:

1. brief read of the references
2. optional light contrast or one nearby alternative
3. a minimal confirmation or an explicit default
4. the final paste-ready NotebookLM prompt
5. optional one-line refinement offer

If the user already chose clearly, skip straight to the final prompt.

Default Recommendation Rules

If the user asks you to decide, pick defaults using this order:

1. prioritize the content goal over the fanciest visual style
2. prefer clarity for training and explanation
3. prefer structure for technical and data-heavy material
4. prefer strategic framing for executive and roadmap material
5. prefer human warmth for people-centered stories
6. prefer high contrast and bold rhythm only when impact is the actual goal

Do not force every task into a named style.

If the references imply a hybrid, output a hybrid.

Quality Rules

The final prompt must:

• be specific enough to guide NotebookLM
• stay broad enough for NotebookLM to execute
• avoid fake precision like pixel-perfect placement
• include widescreen composition guidance
• avoid long-text dependence inside generated images
• include anti-clutter constraints when needed

It should feel like a faithful translation of the user's taste, not an imposed system preference.

It should also feel fast enough that the user can react to a first version instead of being trapped in pre-work.

Output Requirements

When giving the final prompt:

• wrap it in a fenced code block for easy copying
• keep the inside YAML-first by default
• if YAML is used, make the keys readable and practical rather than overly technical
• match the user's language unless they asked for a different one

Hard Stops

Do not finish with only:

• “推荐风格如下”
• “可以考虑这三个方向”
• “下面是分析”

Always land on a complete NotebookLM prompt unless the user explicitly asked to stop earlier.