Skip to content

qna.md

Latest commit

 

History

History
146 lines (101 loc) · 4.94 KB

qna.md

File metadata and controls

146 lines (101 loc) · 4.94 KB
Raw

Workflow: Concept Q&A

For text-based ICT/SMC questions when no chart is attached.

When this workflow applies

  • User asks "what is X" / "how to identify Y" / "how to enter Z"
  • Topics on ICT / SMC / Wyckoff / VSA / Order Flow / Price Action
  • User mentions specific terms (FVG / OB / Killzone / CISD / OTE / Liquidity / Inducement / Breaker / IFVG, etc.)
  • User asks about trading strategy, market structure, risk management

When NOT to use

  • User attached a chart → switch to analyze.md
  • General greetings / unrelated topics → no skill needed

Special case: data-source / freshness questions

If the user's question is about the data source / pipeline / freshness / upstream vendors — e.g. "数据从哪来 / 数据源 / data source / where is this data from / 你用什么数据 / 实时吗 / 怎么取的数据" — do NOT call kb_retrieve.py. Instead, respond using the canonical data-source disclosure from SKILL.body.md § "Data source disclosure".

If the conversation has already produced an API response in this turn (klines / indicators / chart), substitute its freshness block and exchange/market/symbol fields into the template.

If no API call has been made yet, answer:

本对话尚未发起行情数据请求；如果你接下来问某个资产的行情，数据将通过 Mobius Quant API (api.mobiusquant.ai) 获取。具体上游来源（Mobius 内部接入哪些交易所 / 数据供应商）skill 无法核实，详情见 mobiusquant.ai。

Then stop. Do NOT add SMC analysis or chart in response to a data-source question alone.

Steps

Step 1: Retrieve relevant cards

Extract the core concepts from the user's question (prefer English technical terms), then run:

.venv/bin/python scripts/kb_retrieve.py "<query>" --top-k 5

Variants:

# Case-only retrieval
kb_retrieve.py "BTC reversal liquidity sweep" --type case --top-k 5

# School filter
kb_retrieve.py "smart money concepts market structure" --school ICT --top-k 5

Step 2: Synthesize the answer

The tool returns markdown-formatted cards with: Term / Aliases / School / Definition / Identification rules / Trading implication / Common mistakes / Related concepts.

Strict requirements:

  1. Anchor every claim to the knowledge base — cite specific rule numbers
  2. No vague generalities — give concrete identification steps, entry points, stop placements
  3. If retrieval is insufficient — say "knowledge base does not explicitly cover X, but concept Y may be relevant"
  4. Link related concepts — when discussing FVG, mention PD Array / OTE / CISD relations
  5. Match user's language — Chinese question → Chinese answer; English → English (technical terms stay English per shared rules)

Query optimization tips

  • Use English technical terms for best retrieval (knowledge base is English):
    • "如何识别市场反转" → retrieve "market structure shift trend reversal"
    • "止损放哪" → retrieve "stop loss placement swing point"
  • Join multiple concepts with spaces to let vector search match related clusters
  • For case queries, use concrete features: "4H FVG liquidity sweep entry"

Examples

Example 1 — Concept question

User: "什么是 Fair Value Gap，怎么交易"

Action:

kb_retrieve.py "Fair Value Gap how to trade entry" --top-k 5

Response (in Chinese, technical terms in English):

  • Precise FVG definition (three-candle non-overlap pattern)
  • 3 identification rules (specific bullish/bearish criteria)
  • Entry strategy (wait for CISD confirmation + entry at OTE 0.62-0.79 + stop below swept low)
  • Common mistakes (5 concrete pitfalls, citing the knowledge base)

Example 2 — School overview

User: "ICT 是什么流派，它的核心方法论是什么"

Action:

kb_retrieve.py "ICT methodology smart money concepts" --school ICT --top-k 8

Response:

  • Positioning of ICT (Inner Circle Trader)
  • 4-5 core tools (OB / FVG / Liquidity Sweep / Killzone)
  • Typical workflow: HTF bias → PD Array → CISD → entry → stops/targets
  • Common misapplications

Example 3 — Case query

User: "找一个 BTC 在 FVG 反转的真实案例"

Action:

kb_retrieve.py "BTC bitcoin Fair Value Gap reversal entry" --type case --top-k 3

Response: Extract 1-2 most relevant cases' analysis_steps + lessons.

Output format

Standard prose answer (no JSON, no special structure). The 5-section format with auto-annotation is for chart analysis only — Q&A is free-form structured prose with clear sections like:

## 定义 / Definition
<from card>

## 识别规则 / Identification Rules
1. <rule 1>
2. <rule 2>
...

## 交易意义 / Trading Implication
<from card>

## 常见错误 / Common Mistakes
- <mistake 1>
- <mistake 2>

## 相关概念 / Related
<linked terms>

Adapt section headers to fit the question type (e.g. for a strategy question, lead with "Strategy" rather than "Definition").