| name | stage7-deployment-scenario |
| description | Generate a client-facing deployment scenario and QA set from pipeline outputs (Stages 1-6). Maps consultation categories to resolution methods (RAG/Task) and deployment steps, with test queries per category. Outputs HTML + Markdown, optionally publishes to Notion. |
Stage 7: ๋ฐฐํฌ ์๋๋ฆฌ์ค & QA ์ธํธ ์์ฑ
Overview
Stage 1~6 ์ฐ์ถ๋ฌผ์ ์ฌ๊ตฌ์กฐํํ์ฌ ๊ณ ๊ฐ์ฌ ๊ณต์ ์ฉ ๋ฐฐํฌ ์๋๋ฆฌ์ค + QA ์ธํธ๋ฅผ ์์ฑํฉ๋๋ค.
๊ธฐ์กด ๋ถ์ ๋ฐ์ดํฐ๋ฅผ ์๋ก ์์ฑํ์ง ์๊ณ ์กฐํฉ/์ฌ๊ตฌ์ฑ๋ง ์ํํฉ๋๋ค.
Language: Detect the language from the user's first message and respond in that language throughout. Support Korean (ํ๊ตญ์ด) and Japanese (ๆฅๆฌ่ช). Default to Korean if language is unclear.
ํต์ฌ ์ฐ์ถ๋ฌผ:
- ์๋ด ์นดํ
๊ณ ๋ฆฌ ร ๋น๋ ร ํด๊ฒฐ๋ฐฉ์(RAG/Task) ร ๋ฐฐํฌ ๋จ๊ณ(Step 1/2) ๋งคํ
- ๊ฐ ์นดํ
๊ณ ๋ฆฌ๋ณ ํ
์คํธ ์ฟผ๋ฆฌ(๊ณ ๊ฐ ๋ฐํ ์์ + ๊ธฐ๋ ๋ต๋ณ) ํฌํจ QA ์ธํธ
- HTML ํ์ (๊ณ ๊ฐ์ฌ ๊ณต์ ) + Markdown (๋ก์ปฌ ๋ณด๊ด)
๋ฐฐํฌ ๋จ๊ณ ์ ์:
| ๋จ๊ณ | ๋ฒ์ | ์์ ๊ธฐ๊ฐ | ๋ด๋น | ์ค๋ช
|
|---|
| Step 1-1 | RAG + rules ๋ก ALF๊ฐ ์๊ฒฐํ๋ ํ๋ฆ | 2~3์ผ | CSํ | ์ ๋ณด ์๋ด๋ง์ผ๋ก ์๋ด ์ข
๋ฃ. ์๋ด์ฌ ๊ฐ์
๋ถํ์. |
| Step 1-2 | RAG + rules ์์ ์๋ด์ฌ ์ฐ๊ฒฐ์ด ํ์ํ ํ๋ฆ | 2~3์ผ | CSํ | ALF๊ฐ ์ด๊ธฐ ์๋/์ ๋ณด ์์ง ํ ์๋ด์ฌ๋ก ์ ํ. ๊ณ ๊ฐ์ฌ์ ์กฐ์จ ํ์ โ ์ด๋ค ์์ ์ ์ฐ๊ฒฐํ ์ง, ์ด๋ค ์ ๋ณด๋ฅผ ์ฌ์ ์์งํ ์ง ๊ตฌ์ฒด์ ์ผ๋ก ์ ์ํด์ผ ํจ. |
| Step 2 | API ํธ์ถ ์ฝ๋๋
ธ๋ ํฌํจ task | 1~2์ฃผ | CSํ + ๊ฐ๋ฐํ | ๊ณ ๊ฐ์ฌ ๊ฐ๋ฐ ํ์. |
Step 1-1 vs 1-2 ๊ตฌ๋ถ ๊ธฐ์ค: ๋์ผ ์นดํ
๊ณ ๋ฆฌ ์์์๋ ํ๋ฆ์ ๋ฐ๋ผ ๋๋จ.
์) "AS ๋น์ฉ/๋ณด์ฆ๊ธฐ๊ฐ" โ "๋ฐฐ์ก๋น ์ผ๋ง?" = 1-1 (์ ๋ณด ์๋ด ์๊ฒฐ) / "๋ฌด์์ธ์ง ์ ์์ธ์ง ๋ด์ฃผ์ธ์" = 1-2 (์๋ด์ฌ ํ๋จ ํ์)
์ปค๋ฒ๋ฆฌ์ง ์งํ ์ ์:
| ์งํ | ์ ์ | ์ฐ์ถ ๋ฐฉ์ |
|---|
| ๊ด์ฌ์จ | ์ ์ฒด ์๋ด ์ค ALF๊ฐ ๊ฐ์
ํ๋ ๋น์จ | Step์ ํด๋นํ๋ ์นดํ
๊ณ ๋ฆฌ ๊ฑด์ ํฉ๊ณ / ์ ์ฒด ๊ฑด์ |
| ํด๊ฒฐ์จ | ALF๊ฐ ๊ฐ์
ํ ์๋ด ์ค ์๋ด์ฌ ๊ฐ์
์์ด ์๋ฃํ ๋น์จ | automation_analysis.md์ ๋ณด์์ ์กฐ์ ๊ฐ (์์ ํด๊ฒฐ + ์น์ธ๋
ธ๋) |
| ์ปค๋ฒ๋ฆฌ์ง | ์ ์ฒด ์๋ด ๋๋น ALF๊ฐ ์๋ ์๋ฃํ๋ ๋น์จ | ๊ด์ฌ์จ ร ํด๊ฒฐ์จ |
์์ฝ ํ
์ด๋ธ์๋ ๊ด์ฌ์จ, ํด๊ฒฐ์จ, ์ปค๋ฒ๋ฆฌ์ง(= ๊ด์ฌ์จ ร ํด๊ฒฐ์จ)๋ฅผ ๋ชจ๋ ํ๊ธฐํฉ๋๋ค.
์
๋ ฅ ํ์ผ:
05_sales_report/analysis/cross_analysis.json โ ์นดํ
๊ณ ๋ฆฌ ๋น๋
05_sales_report/alf_setup/rag_items.md โ RAG ํญ๋ชฉ (Priority 1/2)
{company}_api_requirements.md โ API ํ์/๋ถํ์ ํ์คํฌ ๋ถ๋ฅ
05_sales_report/tasks/TASK*.md โ ํ์คํฌ ์ ์
02_extraction/faq.json โ FAQ Q/A ์ (ํ
์คํธ ์ฟผ๋ฆฌ ์์ค)
03_sop/*.sop.md โ SOP Case ์๋๋ฆฌ์ค (ํ
์คํธ ์ฟผ๋ฆฌ ์์ค)
05_sales_report/analysis/automation_analysis.md โ ์ปค๋ฒ๋ฆฌ์ง ์์น
{company}_alf_implementation_guide.md โ ROI ์์น
์ฐ์ถ๋ฌผ:
results/{company}/07_deployment/
โโโ deployment_qa_set.html โ QA ์ธํธ (๊ณ ๊ฐ์ฌ ๊ณต์ ์ฉ, ์ธ๋ผ์ธ CSS)
โโโ deployment_qa_set.md โ QA ์ธํธ (๋ก์ปฌ ๋ณด๊ด์ฉ ๋งํฌ๋ค์ด)
โโโ deployment_flow.html โ ํ ํฝโํด๊ฒฐ๋ฐฉ์โStep ๋งคํ ์๊ฐํ (SVG)
Estimated Time: ~5 minutes
Parameters
Required
- company: ํ์ฌ ์๋ณ์ (์:
ppodeuk, usimsa_v2)
Optional
- output_dir (default:
results/{company}/07_deployment): ์ถ๋ ฅ ๋๋ ํ ๋ฆฌ
- notion_parent_page (default: none): Notion ๋ถ๋ชจ ํ์ด์ง URL/ID. ์ ๊ณต ์ Notion์ ํผ๋ธ๋ฆฌ์ฑ.
Steps
1. Validate Input Files
Actions:
- Scan
results/ for company directories containing 05_sales_report/analysis/cross_analysis.json
- Use
AskUserQuestion to confirm target company
- Verify all input files exist:
cross_analysis.json (REQUIRED)
rag_items.md (REQUIRED)
{company}_api_requirements.md (REQUIRED)
05_sales_report/tasks/TASK*.md (REQUIRED, at least 0 โ some companies may have no tasks)
02_extraction/faq.json (REQUIRED โ for test queries)
03_sop/*.sop.md (REQUIRED โ for test queries)
automation_analysis.md (optional โ for coverage figures)
alf_implementation_guide.md (optional โ for ROI figures)
Constraints:
- If
cross_analysis.json, rag_items.md, or api_requirements.md is missing, STOP and inform user to run Stages 5-6 first
- If
faq.json or SOPs are missing, STOP and inform user to run Stage 2-3 first
Expected Output:
โ
์
๋ ฅ ํ์ผ ํ์ธ ์๋ฃ
- cross_analysis.json: โ
({N} topics, {N} total chats)
- rag_items.md: โ
(Priority 1: {N}ํญ๋ชฉ / Priority 2: {N}ํญ๋ชฉ)
- api_requirements.md: โ
(API ํ์: {N}๊ฐ / ๋ถํ์: {N}๊ฐ)
- tasks: {N}๊ฐ
- faq.json: โ
({N} pairs)
- SOPs: {N}๊ฐ
2. Build Category Mapping Table
Read the source files and build the master mapping of categories to resolution methods and deployment steps.
Actions:
-
Read cross_analysis.json โ extract topic_dialog_cross for per-topic record counts. Calculate each topic's total count and percentage of total_chats. Sort by count descending.
-
Classify each topic's resolution method:
- Read
rag_items.md:
- Topics referenced ONLY in Priority 1 items (์ง์์๋ต/์ ์ฑ
ํ์ธ) โ RAG
- Read
api_requirements.md:
- Section 3 "API ๋ถํ์ ํ์คํฌ" โ RAG (knowledge response only) or RAG + ์๋ด์ฌ (if listed as requiring agent handling)
- Section 1 "API ํ์ ํ์คํฌ" โ Task (API)
- Read
05_sales_report/tasks/TASK*.md file list:
- Topics with a TASK file โ Task
- Check each TASK: if it has API call nodes (dashed orange border in Mermaid, or listed in api_requirements Section 1) โ Task (API) = Step 2
- If TASK has NO API calls โ Task (๋จ์) = Step 1
-
Assign deployment step โ classify at the flow level, not just category level. A single category can have BOTH Step 1-1 and Step 1-2 flows:
- Step 1-1: FAQ/์ ์ฑ
์ ๋ณด ์๋ด๋ง์ผ๋ก ์๋ด ์ข
๋ฃ๋๋ ํ๋ฆ. ํ๋จ/์ฆ๋น/์ฌ์ง ํ์ธ ๋ถํ์.
- Step 1-2: ALF๊ฐ ์ด๊ธฐ ์๋/์ ๋ณด ์์ง ํ ์๋ด์ฌ ์ฐ๊ฒฐ์ด ํ์ํ ํ๋ฆ. ๊ณ ๊ฐ์ฌ์ ์ฌ์ ์กฐ์จ ํ์ โ ์ด๋ค ์์ ์ ์ ํํ ์ง, ์ด๋ค ์ ๋ณด๋ฅผ ์์งํ ์ง ๊ตฌ์ฒด์ ์ผ๋ก ์ ์.
- Step 2: API ํธ์ถ์ด ํ์ํ Task
- ์๋ด์ฌ ์ ํ: Topics classified as
7_์๋ด์ฌ์ ํ dominant in cross_analysis
-
Step 1-1 vs 1-2 ๊ตฌ๋ถ ๋ฐฉ๋ฒ:
- Read
faq.json โ Q/A ์์ answer๊ฐ ์ ๋ณด ์๋ด๋ก ๋๋๋์ง(1-1), "AS ์ ์ ์์ ์์ง" / "์ฌ์ง/์์ ๋ณด๋ด์ฃผ์ธ์" / "์๋ด์ฌ ํ์ธ ํ ์๋ด" ๋ฑ์ผ๋ก ์ด์ด์ง๋์ง(1-2) ํ๋จ
- Read SOP Cases โ Case์ ์ต์ข
๋จ๊ณ๊ฐ "์๋ด ์๋ฃ"๋ฉด 1-1, "์๋ด์ฌ ์ฐ๊ฒฐ" / "์๋ ์ฒ๋ฆฌ" / "์ฆ๋น ํ๋จ"์ด๋ฉด 1-2
- ๋์ผ ์นดํ
๊ณ ๋ฆฌ์ ์๋ก ๋ค๋ฅธ ํ๋ฆ์ ๊ฐ๊ฐ 1-1๊ณผ 1-2์ ๋ฐฐ์น
-
Step 1-2 ์์ธ ์๊ตฌ์ฌํญ (๊ณ ๊ฐ์ฌ ์กฐ์จ ํ์ ํญ๋ชฉ):
๊ฐ 1-2 ํ๋ฆ์ ๋ํด ๋ค์์ ๋ช
์:
- ALF ์ฒ๋ฆฌ ๋ฒ์: ALF๊ฐ ์ด๋๊น์ง ํ๋์ง (์: ์ฆ์ ํ์ธ + ์์ ์์ง)
- ์๋ด์ฌ ์ ํ ์์ : ์ ํํ ์ด๋ค ์กฐ๊ฑด์์ ์๋ด์ฌ๋ก ๋๊ธฐ๋์ง
- ์ฌ์ ์์ง ์ ๋ณด: ์ ํ ์ ALF๊ฐ ์์งํด์ผ ํ ๋ฐ์ดํฐ ๋ชฉ๋ก
- ๊ณ ๊ฐ์ฌ ํ์ธ ํ์ ์ฌํญ: ๊ณ ๊ฐ์ฌ์ ํฉ์ํด์ผ ํ ํฌ์ธํธ (์: ๋ถ๋ ํ๋จ ๊ธฐ์ค, ํ๋ถ ์น์ธ ๊ถํ)
Output format (internal, used for Step 3-4):
[
{ topic: "HT_USAGE_ACCESSORIES", count: 214, pct: 7.5, flows: [
{ flow: "์๋ชจํ ๊ตฌ๋งค ์๋ด", step: "1-1", resolution: "RAG" },
{ flow: "๋ชจ๋ธ ๋ฏธํ์ธ โ ์๋ด์ฌ ํ์ธ", step: "1-2", resolution: "RAG โ ์๋ด์ฌ" }
]},
{ topic: "TS_PRODUCT_DEFECT_TRIAGE", count: 628, pct: 22.1, flows: [
{ flow: "์๊ฐ์ ๊ฒ ์๋ด (์ ์ ๋ฒ์ ์ฆ์)", step: "1-1", resolution: "RAG" },
{ flow: "๋ถ๋ ํ๋จ โ AS ์ ์", step: "1-2", resolution: "RAG โ ์๋ด์ฌ", handoff_point: "์ฌ์ง ์ฆ๋น ํ๋จ", pre_collect: ["์ฑํจ","์ฐ๋ฝ์ฒ","๋ชจ๋ธ๋ช
","์ฆ์","์ฌ์ง"] }
]},
{ topic: "TS_AS_RECEIPT", count: 579, pct: 20.4, flows: [
{ flow: "AS ์ ์ ์๋ํ", step: "2", resolution: "Task (API)", task: "TASK1", apis: ["์ฃผ๋ฌธ์กฐํ","AS์ ์๋ฑ๋ก"] }
]},
...
]
Constraints:
- Every topic in
cross_analysis.json MUST appear in the mapping โ no omissions
- A topic can appear in BOTH Step 1 (RAG part) and Step 2 (Task part) if it has both knowledge items and API tasks
- Resolution method MUST be derived from source files, not guessed
3. Generate Test Queries
For each flow (not just category) in the mapping table, generate 1-2 test queries.
Actions:
For Step 1-1 flows (ALF ์๊ฒฐ):
- Read
02_extraction/faq.json โ find Q/A pairs where the answer is self-contained information
- Select the most representative 1-2 pairs
- Format as:
- ๊ตฌ์ฒด์ ์ฟผ๋ฆฌ: The actual customer utterance from FAQ
- ๊ธฐ๋ ๋ต๋ณ: The expected answer summary (information-only, no agent handoff)
For Step 1-2 flows (์๋ด์ฌ ์ฐ๊ฒฐ) โ DETAILED:
- Read
faq.json + 03_sop/*.sop.md โ find flows that end in agent handoff
- For each flow, document:
- ๊ตฌ์ฒด์ ์ฟผ๋ฆฌ: Customer utterance that triggers this flow
- ALF ์๋ ๋ฒ์: What ALF handles before handoff (e.g., self-check guide, form collection)
- ์๋ด์ฌ ์ ํ ์์ : Exact condition/trigger for agent handoff
- ์ฌ์ ์์ง ์ ๋ณด: Data ALF must collect before handoff (e.g., ์ฑํจ, ์ฐ๋ฝ์ฒ, ๋ชจ๋ธ๋ช
, ์ฆ์, ์ฌ์ง)
- ๊ณ ๊ฐ์ฌ ํ์ธ ํ์: What needs to be agreed with the client (e.g., defect judgment criteria, refund authority)
For Step 2 flows (Task API):
- Read the corresponding
05_sales_report/tasks/TASK*.md โ extract the trigger scenario
- Read the corresponding
03_sop/*.sop.md โ extract Case 1 customer scenario
- Format as:
- ๊ตฌ์ฒด์ ์ฟผ๋ฆฌ: Realistic customer request that triggers this task
- ๊ธฐ๋ ํ๋ก์ฐ: Expected task execution flow (from TASK file summary table)
- Note the required API endpoints
Constraints:
- Test queries MUST use natural customer language, not technical terms
- Expected answers MUST come from SOP/FAQ source data โ no fabrication
- Each category MUST have at least 1 test query
- Maximum 2 test queries per category (1 concrete + 1 abstract if data allows)
4. Generate Output Files
4-A. deployment_qa_set.md (Markdown)
Write the structured document following this format:
# {ํ์ฌ๋ช
} ALF ๋ฐฐํฌ ์๋๋ฆฌ์ค & QA ์ธํธ
> ์์ฑ์ผ: {YYYY-MM-DD} | ๋ถ์ ๊ท๋ชจ: {N}๊ฑด / {K}๊ฐ ์นดํ
๊ณ ๋ฆฌ
> Step 1 ์์ ์์: 2~3์ผ | Step 2 ์์ ์์: +1~2์ฃผ
---
## ์์ฝ
| ๊ตฌ๋ถ | ํญ๋ชฉ ์ | ๊ด์ฌ์จ | ํด๊ฒฐ์จ | ์ปค๋ฒ๋ฆฌ์ง (๊ด์ฌ์จ ร ํด๊ฒฐ์จ) |
|------|--------|--------|--------|--------------------------|
| **Step 1** (RAG + ๊ท์น + ๋จ์ ํ์คํฌ) | {N}๊ฐ ์นดํ
๊ณ ๋ฆฌ | {A}% | {B}% | **{AรB/100}%** |
| **Step 2** (API ์ฐ๋ ํ์คํฌ) | {N}๊ฐ ์นดํ
๊ณ ๋ฆฌ | ๋์ {C}% | {D}% | ๋์ **{CรD/100}%** |
| ์๋ด์ฌ ์ ํ | {N}๊ฐ ์นดํ
๊ณ ๋ฆฌ | {E}% | โ | โ |
> **๊ด์ฌ์จ** = ํด๋น Step ์นดํ
๊ณ ๋ฆฌ ๊ฑด์ / ์ ์ฒด ๊ฑด์
> **ํด๊ฒฐ์จ** = ALF ๊ฐ์
์๋ด ์ค ์๋ด์ฌ ์์ด ์๋ฃ ๋น์จ (automation_analysis.md ๋ณด์์ ๊ธฐ์ค)
> **์ปค๋ฒ๋ฆฌ์ง** = ๊ด์ฌ์จ ร ํด๊ฒฐ์จ
---
## Step 1: RAG + ๊ท์น + ๋จ์ ํ์คํฌ (2~3์ผ)
> CSํ๋ง์ผ๋ก ์ธํ
๊ฐ๋ฅ. ๊ท์น ๋ฑ๋ก + RAG ์ง์ ๋ฑ๋ก + ๋จ์ ํ์คํฌ ์ค์ .
### {ํ ํฝ ํ๊ธ๋ช
} โ {N}๊ฑด ({X}%) | RAG
| # | ์ ํ | ๊ณ ๊ฐ ๋ฐํ ์์ | ๊ธฐ๋ ๋ต๋ณ | QA ๊ฒฐ๊ณผ |
|---|------|-------------|----------|---------|
| 1 | ๊ตฌ์ฒด | "{์ค์ ๊ณ ๊ฐ ๋ฐํ}" | {๊ธฐ๋ ๋ต๋ณ ์์ฝ} | |
| 2 | ์ถ์ | "{์งง์/๋ค๋ฅธ ํํ}" | {๊ธฐ๋ ๋ต๋ณ ์์ฝ} | |
### {ํ ํฝ ํ๊ธ๋ช
} โ {N}๊ฑด ({X}%) | Task (๋จ์)
| # | ์ ํ | ๊ณ ๊ฐ ๋ฐํ ์์ | ๊ธฐ๋ ํ๋ก์ฐ | QA ๊ฒฐ๊ณผ |
|---|------|-------------|-----------|---------|
| 1 | ๊ตฌ์ฒด | "{์ค์ ๊ณ ๊ฐ ์์ฒญ}" | {๊ธฐ๋ ์ฒ๋ฆฌ ํ๋ฆ} | |
---
## Step 2: API ์ฐ๋ ํ์คํฌ (+1~2์ฃผ)
> ๊ณ ๊ฐ์ฌ ๊ฐ๋ฐํ API ๊ฐ๋ฐ ํ์. ๊ฐ๋ฐ ์๋ฃ ํ ํ์คํฌ ์ฝ๋๋
ธ๋ ์ฐ๊ฒฐ.
### {ํ ํฝ ํ๊ธ๋ช
} โ {N}๊ฑด ({X}%) | Task (API: {API๋ช
})
> ํ์ API: {endpoint ๋ชฉ๋ก}
| # | ์ ํ | ๊ณ ๊ฐ ๋ฐํ ์์ | ๊ธฐ๋ ํ๋ก์ฐ | QA ๊ฒฐ๊ณผ |
|---|------|-------------|-----------|---------|
| 1 | ๊ตฌ์ฒด | "{์ค์ ๊ณ ๊ฐ ์์ฒญ}" | {๊ธฐ๋ ์ฒ๋ฆฌ ํ๋ฆ} | |
---
## ์ ์ฒด QA ์์ฝ
| # | ์นดํ
๊ณ ๋ฆฌ | ๋น๋ | ํด๊ฒฐ ๋ฐฉ์ | Step | QA ํญ๋ชฉ ์ | ๊ฒฐ๊ณผ |
|---|---------|------|----------|------|-----------|------|
| 1 | {ํ ํฝ๋ช
} | {N}๊ฑด ({X}%) | RAG | 1 | {N} | |
| 2 | {ํ ํฝ๋ช
} | {N}๊ฑด ({X}%) | Task (API) | 2 | {N} | |
| ... | | | | | | |
| **ํฉ๊ณ** | | **{total}๊ฑด** | | | **{total_qa}** | |
4-B. deployment_qa_set.html (HTML)
Convert the Markdown to a standalone HTML file with these features:
- Inline CSS only โ no external dependencies, single file
- Clean, professional design: Light gray background, white card sections, subtle borders
- Collapsible sections: Step 1 and Step 2 sections use
<details><summary> for fold/unfold
- QA result checkboxes: Each QA row has an interactive checkbox (
<input type="checkbox">)
- Print-friendly:
@media print styles that expand all sections and show checkboxes as boxes
- Color coding: Step 1 items in green accent, Step 2 in orange accent, ์๋ด์ฌ์ ํ in red accent
- Summary stats at top: Total categories, Step 1/2 counts, coverage percentages
- Responsive: Works on both desktop and mobile screens
- Korean font stack:
"Pretendard", "Noto Sans KR", sans-serif
Constraints:
- HTML file MUST be completely self-contained (no CDN, no external CSS/JS)
- File size should be reasonable (~50-100KB max)
- Tables MUST be properly formatted with borders and padding
- The HTML MUST render correctly when opened directly in a browser (
file:// protocol)
4-C. deployment_flow.html (์๊ฐํ)
Generate a standalone HTML file with an inline SVG chart showing the 3-column flow:
3-Column Layout:
- ์ผ์ชฝ โ ์๋ด ํ ํฝ (๋น๋): Colored bars with height proportional to record count. Show topic name + count + percentage.
- ๊ฐ์ด๋ฐ โ ํด๊ฒฐ ๋ฐฉ์: RAG/Rules box + Task (API) box. Show aggregate counts.
- ์ค๋ฅธ์ชฝ โ ๋ฐฐํฌ ๋จ๊ณ: Step 1-1 (green), Step 1-2 (blue), Step 2 (orange) boxes. Show flow counts + coverage + timeline.
Flow Lines:
- Curved paths (
<path> with cubic bezier) connecting topics โ resolution โ steps
- Line thickness proportional to record count
- Color matches the destination step
- Opacity 0.3-0.6 for readability
Right-side Annotations:
- For topics that split across Step 1-1 and 1-2, show annotation boxes listing which flows go where
Design Requirements:
- Inline SVG within HTML (no external images)
viewBox based sizing for responsiveness
- Same CSS conventions as 4-B (font stack, background, print styles)
- Legend at bottom explaining colors
Reference: Use results/์ค์/07_deployment/deployment_flow.html as the structural reference. Adapt the data (topic names, counts, flow mappings) from the category mapping table built in Step 2.
5. Notion Publishing (optional)
Skip if: notion_parent_page is not provided.
Actions:
- Use
AskUserQuestion to get the Notion parent page URL or ID (if not already provided as parameter)
- If user says 'skip', skip this step entirely
- Fetch the parent page using
mcp__claude_ai_Notion__notion-fetch to validate access
- Create a hub page:
{ํ์ฌ๋ช
} ALF ๋ฐฐํฌ ์๋๋ฆฌ์ค under the parent page using mcp__claude_ai_Notion__notion-create-pages
- Include the full deployment scenario content (converted to Notion markdown)
- Create a QA tracking database under the hub page using
mcp__claude_ai_Notion__notion-create-database:
CREATE TABLE (
"์นดํ
๊ณ ๋ฆฌ" TITLE,
"๋น๋" NUMBER,
"ํด๊ฒฐ ๋ฐฉ์" SELECT('RAG':blue, 'Task (๋จ์)':green, 'Task (API)':orange, '์๋ด์ฌ':red),
"๋ฐฐํฌ ๋จ๊ณ" SELECT('Step 1':green, 'Step 2':orange),
"ํ
์คํธ ์ฟผ๋ฆฌ" RICH_TEXT,
"๊ธฐ๋ ๋ต๋ณ" RICH_TEXT,
"QA ๊ฒฐ๊ณผ" SELECT('Pass':green, 'Fail':red, '๋ฏธํ
์คํธ':gray),
"๋ฉ๋ชจ" RICH_TEXT
)
- Create one row per QA item using
mcp__claude_ai_Notion__notion-create-pages
Constraints:
- Notion page content MUST follow Notion's enhanced markdown spec
- If Notion API fails, skip gracefully and note in the summary โ local files are the primary output
6. Verification + Summary
Verify:
- Completeness: Every topic from
cross_analysis.json appears in the output
- Classification consistency: Step assignments match
api_requirements.md
- Test query coverage: Every category has at least 1 test query
- File integrity:
.md, .html, and deployment_flow.html all exist and are non-empty
Present verification report:
๐ ๊ฒ์ฆ ๊ฒฐ๊ณผ
์นดํ
๊ณ ๋ฆฌ ์: cross_analysis {N}๊ฐ โ QA ์ธํธ {N}๊ฐ โ โ
์ผ์น
Step 1 ํญ๋ชฉ: {N}๊ฐ (RAG {N} + ๋จ์ Task {N})
Step 2 ํญ๋ชฉ: {N}๊ฐ (API Task {N})
์๋ด์ฌ ์ ํ: {N}๊ฐ
ํ
์คํธ ์ฟผ๋ฆฌ: ์ด {N}๊ฐ (์นดํ
๊ณ ๋ฆฌ๋น ํ๊ท {X}๊ฐ)
Notion: {โ
ํผ๋ธ๋ฆฌ์ฑ ์๋ฃ / โญ๏ธ ์คํต}
์์ ์ด ํ์ํ ๋ถ๋ถ์ด ์์ผ๋ฉด ๋ง์ํด์ฃผ์ธ์.
After user approval, present final summary:
โ
Stage 7 ์๋ฃ โ ๋ฐฐํฌ ์๋๋ฆฌ์ค & QA ์ธํธ
๐ Output:
- results/{company}/07_deployment/deployment_qa_set.md
- results/{company}/07_deployment/deployment_flow.html
- results/{company}/07_deployment/deployment_qa_set.html
{- Notion: {page_url} (์๋ ๊ฒฝ์ฐ)}
๐ ์์ฝ:
- ์ ์ฒด ์นดํ
๊ณ ๋ฆฌ: {N}๊ฐ
- Step 1 (2~3์ผ): {N}๊ฐ โ ๊ด์ฌ์จ {A}% ร ํด๊ฒฐ์จ {B}% = ์ปค๋ฒ๋ฆฌ์ง {X}%
- Step 2 (+1~2์ฃผ): {N}๊ฐ โ ๋์ ๊ด์ฌ์จ {C}% ร ํด๊ฒฐ์จ {D}% = ๋์ ์ปค๋ฒ๋ฆฌ์ง {Y}%
- QA ์๋๋ฆฌ์ค: ์ด {N}๊ฐ
Notes
Data Flow
cross_analysis.json โโโ ์นดํ
๊ณ ๋ฆฌ ๋น๋
rag_items.md โโโโโโโโโโโ RAG ๋ถ๋ฅ (Priority 1 = Step 1)
api_requirements.md โโโโ Task ๋ถ๋ฅ (API ์ ๋ฌด โ Step 1/2)
faq.json โโโโโโโโโโโโโโโ RAG ํ
์คํธ ์ฟผ๋ฆฌ (Q/A ์)
03_sop/*.sop.md โโโโโโโโ Task ํ
์คํธ ์ฟผ๋ฆฌ (Case ์๋๋ฆฌ์ค)
automation_analysis.md โ ์ปค๋ฒ๋ฆฌ์ง ์์น
alf_implementation_guide.md โ ROI ์์น
Category Resolution Logic
Topic in api_requirements Section 3 ("API ๋ถํ์")
โ RAG or RAG + ์๋ด์ฌ โ Step 1
Topic in api_requirements Section 1 ("API ํ์")
โ Task (API) โ Step 2
Topic not in api_requirements
โ Check rag_items.md Priority 1 โ RAG โ Step 1
โ Check dominant dialog type in cross_analysis:
1_์ง์์๋ต / 4_์ ์ฑ
ํ์ธ โ RAG โ Step 1
7_์๋ด์ฌ์ ํ โ ์๋ด์ฌ ์ ํ
Otherwise โ Step 1 (conservative default)
Handling Topics That Span Both Steps
Some topics may have BOTH a RAG component (Step 1) and a Task component (Step 2). For example:
- "์๋น์ค ํด์ง" has a RAG knowledge item (ํด์ง ๋ฐฉ๋ฒ ์๋ด) AND a Task (TASK2 ํด์ง ์ฒ๋ฆฌ API)
- In this case, list the topic in BOTH Step 1 (RAG part) and Step 2 (Task part) with clear labels
HTML Design Principles
- Professional, clean design suitable for client presentation
- No JavaScript frameworks โ vanilla HTML/CSS only
- Checkboxes for QA tracking work without JS (native
<input type="checkbox">)
<details>/<summary> for collapsible sections (native HTML5, no JS)
- Print styles for PDF export