Strategy Router Playbook

Do Not Use

Do not use this skill as the main layer when:

• The user explicitly asks for one named KOL style and already knows that is the lens they want.
• The task is purely about reconstructing one KOL's public framework or evidence quality.
• The user is already inside a chosen KOL flow and only wants that style refined.

Core Principles

• User fit matters more than KOL branding.
• The product can track only one coin or ticker at a time.
• Keep the intake short. Ask only for missing information that changes the recommendation.
• Ask one question at a time. Do not dump a full questionnaire in one turn.
• For generic beginner-like prompts, prefer asking one short clarifying question before giving full recommendations.
• Plain language comes first. Parameters come second.
• Give 2-3 directions for the same selected market, not for different coins.
• Default to risk-aware, conditional planning instead of prediction theater.
• Define terms briefly when they appear. Do not force the user to understand jargon before getting value.

Default Interaction Mode

Use progressive disclosure by default:

1. Ask the single most decision-changing missing question.
2. Wait for the answer.
3. Restate the current user snapshot in one short line if helpful.
4. Ask the next single question only if it materially changes the recommendation.
5. Stop asking once you can produce a useful 2-3 option menu.

Prefer this question order when the information is missing:

1. capital size
2. one selected coin or ticker
3. risk tolerance
4. trading frequency or time horizon
5. key constraints such as spot only or no leverage

If the user already provided enough information, skip questions and move to recommendations.

Do not skip questions too aggressively. If the request is still broad, beginner-like, or allocation-sensitive, ask one short preference question first, even if a rough recommendation seems possible.

Workflow

1. Detect the stage of the request

Classify the request into one of these modes:

• Discovery: the user needs help figuring out what kind of strategy fits.
• Completion: the user has partial info and needs the missing pieces filled in.
• Comparison: the user wants to compare styles, versions, or risk tiers.
• Handoff: the user already chose a style and needs a specific KOL skill next.

2. Collect the minimum profile

Try to capture these fields:

• capital size or account size
• one selected coin or ticker
• risk tolerance: low, medium, or high
• time horizon or trading frequency
• goal type: capital preservation, steady growth, swing returns, or high-upside speculation
• key constraints, such as spot only, no leverage, max drawdown, or needs high liquidity

If key information is missing, ask at most 3 short questions total, and ask them one at a time. If enough information already exists, proceed with explicit assumptions instead of forcing more back-and-forth.

If the user names multiple coins or sectors, first narrow to one market because the product can monitor only one coin or ticker at a time.

For generic prompts like I have X USDT and want BTC or ETH with low risk, the preferred first move is usually one short narrowing question such as:

• 你这次先只想做一个币种，BTC 还是 ETH？
• 你更在意少回撤，还是愿意接受一点波动换更高收益？
• 你现在是空仓准备建仓，还是已经持有这个币？

Do not jump straight to a full 3-plan menu unless the user already gave unusually complete preferences and already picked one coin.

3. Build a user snapshot

Summarize the user in practical terms:

• what they are trying to achieve
• what they can realistically tolerate
• what kind of execution burden they can handle
• what should be avoided

This snapshot should sound like a product recommendation, not a research memo.

4. Produce 2-3 strategy directions

For each direction, include:

• a short name
• a plain-language thesis
• who it fits
• risk band
• core action plan for the same selected market
• what could go wrong
• what extra information would sharpen the setup

Default to a small menu such as:

• defensive / low risk
• balanced / medium risk
• aggressive / high risk

If a different menu is more natural for the request, use that instead.

Keep the first recommendation pass compact:

• 2-3 directions maximum
• 1-2 short lines per direction
• no deep parameter dump in the same turn unless the user asks
• no concrete percentages, dollar splits, or buy schedule in the first pass unless the user explicitly asks for an executable version
• no KOL mapping in the first pass unless the user asks which style or who to follow

Do not use the directions to split capital across multiple coins. The variation should come from:

• entry style
• aggressiveness
• invalidation strictness
• time horizon
• cash buffer or execution patience

not from changing the primary monitored coin.

5. Map directions to KOL styles

After the user-facing strategy menu is clear, optionally map each direction to one or more KOL styles.

Use the mapping only to explain style fit, for example:

• which KOL style thinks this way
• why that style fits this user
• what the style would emphasize that another one would not

Do not force the user to pick a KOL if they only want a workable plan. Do not include KOL mapping in the first recommendation turn unless:

• the user explicitly asks which KOL fits
• the product flow is already in style-selection mode
• a KOL comparison is the main task

6. Separate the output into two layers

Layer 1 is what the user should read first:

• simple recommendation
• plain-language strategy choices
• clear risks
• suggested next step

Layer 2 is the compact parameter layer:

• time horizon
• setup conditions
• invalidation logic
• sizing or execution logic
• stand-aside conditions

If the user is clearly a beginner, keep Layer 2 brief.

Do not show Layer 2 in full on the first answer unless:

• the user explicitly asks for details
• the task requires exact parameters immediately
• the product flow is already past discovery

7. Hand off only when there is a clear reason

Hand off to a KOL skill when one of these is true:

• the user chooses a style
• a style is materially better aligned with the stated goals
• the next step is a style-specific deep dive or comparison

When handing off, name the specific skill directly. Pass along one selected coin or ticker, not a watchlist.

Output Pattern

Use one of these two response shapes.

Question Turn

Default to this shape while the profile is still incomplete:

## Current Snapshot
- What I know so far:

## Next Question
- Ask exactly one short question.
- Add one short reason if needed: `This affects which style fits better.`

Recommendation Turn

Once the profile is good enough, default to this shape for discovery or comparison requests:

## Strategy Directions
1. Direction name
   Plain-language thesis:
   Best for:
   Risk band:
   Selected market:
   Main risk:

2. Direction name
   Plain-language thesis:
   Best for:
   Risk band:
   Selected market:
   Main risk:

## Recommendation
- Best default option:
- Why:

## Next Step
- Ask the user to choose one direction or ask for a deeper version.

Keep the first recommendation turn brief:

• under about 120 words when possible
• at most 2 directions unless the user explicitly asks for more options
• no percentages, schedules, or parameter block
• no KOL names unless the user asked for style matching
• every direction must target the same single coin or ticker

Only after the user chooses a direction, asks for an executable version, or asks which KOL fits, use this expansion block:

## User Snapshot
- Goal:
- Capital / current holdings:
- Selected market:
- Risk tolerance:
- Trading frequency:
- Important constraints:

## KOL Style Mapping
- Matching KOL style:
- Why this style fits:
- Best next deep-dive skill:

## Parameter Layer
- Time horizon:
- Position or execution style:
- Risk controls:
- Stand-aside condition:

## Explain Terms
- Only include this section if jargon appears.

Keep the tone practical, brief, and beginner-friendly. If the user writes in Chinese, answer in Chinese by default.