Thanks to visit codestin.com
Credit goes to github.com

Skip to content
/ PCIP Public

[System Prompt] Parent-Child Instruction Processing (PCIP) Framework with conversational learning, external knowledge integration, and dynamic expert leadership for optimal coding assistance.

saramjh/PCIP

Folders and files

NameName
Last commit message
Last commit date

Latest commit

ย 

History

9 Commits
ย 
ย 
ย 
ย 
ย 
ย 

Repository files navigation

ํ•œ๊ตญ์–ด

PCIP Framework

AI ์ฝ”๋”ฉ ์–ด์‹œ์Šคํ„ดํŠธ๋ฅผ ์œ„ํ•œ Parent-Child Instruction Processing ํ”„๋ ˆ์ž„์›Œํฌ

License: MIT Version

๊ฐœ์š”

PCIP (Parent-Child Instruction Processing) Framework๋Š” AI ์ฝ”๋”ฉ ์–ด์‹œ์Šคํ„ดํŠธ์˜ ์ปจํ…์ŠคํŠธ ์ดํ•ด์™€ ์ฝ”๋“œ ํ’ˆ์งˆ์„ ํ–ฅ์ƒ์‹œํ‚ค๋Š” 3๊ณ„์ธต ์•„ํ‚คํ…์ฒ˜ ์‹œ์Šคํ…œ์ž…๋‹ˆ๋‹ค.

ํ•ต์‹ฌ ํŠน์ง•

  • 3๊ณ„์ธต ์•„ํ‚คํ…์ฒ˜: PM(๋ฉ”ํƒ€) โ†’ Parent(์ „๋ฌธ๊ฐ€) โ†’ Child(์‹คํ–‰)
  • ๋™์  ์ „๋ฌธ๊ฐ€ ์„ ํƒ: ์š”๊ตฌ์‚ฌํ•ญ์— ๋”ฐ๋ฅธ ์‹ค์‹œ๊ฐ„ ์ „๋ฌธ๊ฐ€ ๋ฐฐ์ •
  • ์œ„ํ—˜๋„ ๊ธฐ๋ฐ˜ ์‹คํ–‰: 5๋‹จ๊ณ„ ์œ„ํ—˜๋„ ํ‰๊ฐ€ ์‹œ์Šคํ…œ
  • ์™ธ๋ถ€ ์ง€์‹ ํ†ตํ•ฉ: RAG/MCP ๋“ฑ์„ ํ†ตํ•œ ์ตœ์‹  ์ •๋ณด ํ™œ์šฉ
  • ๋งฅ๋ฝ ๊ธฐ๋ฐ˜ ์ฒ˜๋ฆฌ: ํ”„๋กœ์ ํŠธ ์•„ํ‚คํ…์ฒ˜ยท๋„๋ฉ”์ธ ๋งฅ๋ฝ ์ž๋™ ํŒŒ์•…

3์ค„ ์š”์•ฝ

๊ณ„์ธตํ™”๋œ AI: PM-Parent(์ „๋ฌธ๊ฐ€)-Child(์‹คํ–‰์ž) ๊ตฌ์กฐ. PM์ด ์ž‘์—…์„ ํŒ๋‹จํ•˜๊ณ  ์•Œ๋งž์€ Parent๋ฅผ ๋ฐฐ์ •ํ•ฉ๋‹ˆ๋‹ค.

๋‘๋‡Œ๊ฐ€ ํ•„์š”ํ•  ๋•Œ(Explicit Mode): ๊ณ ๋‚œ๋„ยท๊ณ ์œ„ํ—˜ ์ž‘์—…์—๋Š” PM์ด Parent๋ฅผ ํˆฌ์ž…ํ•ด ๊ฒฌ๊ณ ํ•œ ์ „๋žต์„ ์ˆ˜๋ฆฝํ•ฉ๋‹ˆ๋‹ค.

๋น ๋ฅด๊ฒŒ ๋๋‚ผ ๋•Œ(Silent Mode): ๋‹จ์ˆœํ•œ ์ž‘์—…์€ Child๊ฐ€ ๋‹จ๋…์œผ๋กœ ์ฒ˜๋ฆฌํ•ฉ๋‹ˆ๋‹ค. Parent ๊ฐœ์ž… ์—†์ด ๋งค์šฐ ๋น ๋ฆ…๋‹ˆ๋‹ค.


๋ฌธ์ œ ์ •์˜

๊ธฐ์กด AI ์ฝ”๋”ฉ ์–ด์‹œ์Šคํ„ดํŠธ์˜ ๋งฅ๋ฝ ์ธ์‹ ํ•œ๊ณ„

ํ˜„์žฌ AI ์ฝ”๋”ฉ ์–ด์‹œ์Šคํ„ดํŠธ๋“ค์˜ ์ฃผ์š” ๋ฌธ์ œ์ :

1) ๋งฅ๋ฝ ๋ฌด์‹œ

  • ํ”„๋กœ์ ํŠธ์˜ ์ „์ฒด ์•„ํ‚คํ…์ฒ˜๋ฅผ ๊ณ ๋ คํ•˜์ง€ ์•Š์€ ์ฝ”๋“œ ์ƒ์„ฑ
  • ๋„๋ฉ”์ธ๋ณ„ ์š”๊ตฌ์‚ฌํ•ญ(๋ณด์•ˆ, ์„ฑ๋Šฅ, ๊ทœ์ •์ค€์ˆ˜ ๋“ฑ) ๊ฐ„๊ณผ
  • ๊ธฐ์กด ์ฝ”๋“œ๋ฒ ์ด์Šค์™€์˜ ์ผ๊ด€์„ฑ ๋ถ€์กฑ

2) ๋ถˆ์™„์ „ํ•œ ๊ตฌํ˜„

์š”์ฒญ: "๊ฒฐ์ œ ์‹œ์Šคํ…œ ๊ตฌํ˜„"
์ „ํ˜•์  ์‹คํŒจ: ๋‹ค์Œ์ด ๋น ์ง„ ๊ธฐ๋ณธ ํผ ์ˆ˜์ค€ ๊ตฌํ˜„
- ๋ณด์•ˆ ๊ณ ๋ ค(PCI DSS ๋“ฑ)
- ์˜ค๋ฅ˜ ์ฒ˜๋ฆฌ ๋ฐ ๋กค๋ฐฑ
- ์ด์ƒ ๊ฑฐ๋ž˜ ํƒ์ง€/์‚ฌ๊ธฐ ๋ฐฉ์ง€
- ๊ฐ์‚ฌ ๋กœ๊น…
- ํ™˜๋ถˆ ์ฒ˜๋ฆฌ

3) ๋ฐ˜๋ณต ๋ณด์ • ์˜ค๋ฒ„ํ—ค๋“œ

  • ์ดˆ๊ธฐ ๊ตฌํ˜„ โ†’ ๋ฌธ์ œ ๋ฐœ๊ฒฌ โ†’ ์ˆ˜์ • ์š”์ฒญ โ†’ ์žฌ๊ตฌํ˜„ ์‚ฌ์ดํด ๋ฐ˜๋ณต
  • ๊ฐœ๋ฐœ์ž๊ฐ€ ๋ชจ๋“  ์„ธ๋ถ€์‚ฌํ•ญ์„ ๋ช…์‹œ์ ์œผ๋กœ ์ง€์‹œํ•ด์•ผ ํ•˜๋Š” ๋ถ€๋‹ด

ํ•ด๊ฒฐ ์•„ํ‚คํ…์ฒ˜

๊ตฌ์กฐํ™”๋œ ์ง€๋„ ์›์น™์—์„œ์˜ ์˜๊ฐ

PCIP Framework๋Š” ๊ตฌ์กฐํ™”๋œ ์ง€๋„ ์›์น™์—์„œ ์˜๊ฐ์„ ์–ป์–ด ์„ค๊ณ„๋˜์—ˆ์Šต๋‹ˆ๋‹ค:

  1. ์ƒํ™ฉ ํŒŒ์•…: ์š”์ฒญ์˜ ๋ณต์žก๋„์™€ ํŠน์„ฑ ๋ถ„์„
  2. ์œ„ํ—˜ ํ‰๊ฐ€: ์ž‘์—…์˜ ์ž ์žฌ์  ๋ฆฌ์Šคํฌ ํ‰๊ฐ€
  3. ๋งฅ๋ฝ ๊ณ ๋ ค: ํ”„๋กœ์ ํŠธ ํŠน์„ฑ, ๊ธฐ์ˆ  ์Šคํƒ, ์‚ฌ์šฉ์ž์ธต ๊ณ ๋ ค
  4. ๊ฐ€์ด๋“œ ์ œ๊ณต: ์ ์ ˆํ•œ ์ „๋ฌธ๊ฐ€ ๊ด€์ ์—์„œ ๋ฐฉํ–ฅ ์ œ์‹œ
  5. ์ง€์† ๊ด€์ฐฐ: ๊ฒฐ๊ณผ๋ฌผ์˜ ํ’ˆ์งˆ ๊ธฐ์ค€ ๊ฒ€์ฆ

3๊ณ„์ธต ์•„ํ‚คํ…์ฒ˜ ๊ฐœ์š”

๊ธฐ์กด ๋ฐฉ์‹

์‚ฌ์šฉ์ž ์š”์ฒญ โ†’ AI โ†’ ์ฝ”๋“œ ์ถœ๋ ฅ
"๊ฒฐ์ œ ์‹œ์Šคํ…œ ๊ตฌํ˜„" โ†’ ์ฆ‰ํฅ ๊ตฌํ˜„ โ†’ ์ž ์žฌ์ ์œผ๋กœ ์œ„ํ—˜ํ•œ ์ฝ”๋“œ

PCIP ๋ฐฉ์‹

์‚ฌ์šฉ์ž ์š”์ฒญ โ†’ PM(๋ถ„์„) โ†’ Parent(๊ฐ€์ด๋“œ) โ†’ Child(์‹คํ–‰)
"๊ฒฐ์ œ ์‹œ์Šคํ…œ" โ†’ ์œ„ํ—˜๋„ ํ‰๊ฐ€ โ†’ ๊ฒฐ์ œ+๋ณด์•ˆ ์ „๋ฌธ๊ฐ€ โ†’ ๊ทœ์ • ์ค€์ˆ˜ยท์•ˆ์ „ํ•œ ์ฝ”๋“œ

๊ณ„์ธต๋ณ„ ์—ญํ• 

๋ฉ”ํƒ€ ๋ ˆ์ด์–ด: PM (ํ”„๋กœ์ ํŠธ ๋งค๋‹ˆ์ €)

  • ์—ญํ• : ๋Œ€ํ™” ๋ถ„์„, ์ „๋ฌธ๊ฐ€ ๋ฐฐ์ •
  • ์ฑ…์ž„:
    • 5๋‹จ๊ณ„ ์œ„ํ—˜๋„ ํ‰๊ฐ€
    • ๋™์  ์ „๋ฌธ๊ฐ€ ํŒ€ ๊ตฌ์„ฑ
    • ์™ธ๋ถ€ ๋ฆฌ์†Œ์Šค ์—ฐ๋™
    • ํ”„๋กœ์ ํŠธ ์ปจํ…์ŠคํŠธ ๊ด€๋ฆฌ

Parent ๋ ˆ์ด์–ด: ๋„๋ฉ”์ธ ์ „๋ฌธ๊ฐ€

  • ์—ญํ• : ํŠนํ™”๋œ ๊ฐ€์ด๋“œ์™€ ํ’ˆ์งˆ ๊ธฐ์ค€ ์ œ์‹œ
  • ์ „๋ฌธ๊ฐ€ ์œ ํ˜•:
    • ๋ณด์•ˆ ์—”์ง€๋‹ˆ์–ด
    • ์„ฑ๋Šฅ ์—”์ง€๋‹ˆ์–ด
    • UX ๋””์ž์ด๋„ˆ
    • ๊ฒฐ์ œ ์‹œ์Šคํ…œ ์ „๋ฌธ๊ฐ€
    • DevOps ์—”์ง€๋‹ˆ์–ด
    • ๋ชจ๋ฐ”์ผ ์ŠคํŽ˜์…œ๋ฆฌ์ŠคํŠธ

Child ๋ ˆ์ด์–ด: ์‹คํ–‰ ์—”์ง„

  • ์—ญํ• : ๊ฐ€์ด๋“œ์— ๋”ฐ๋ฅธ ์ •ํ™•ํ•œ ๊ตฌํ˜„
  • ์—ญ๋Ÿ‰:
    • ๋ฌธ๋ฒ•ยท์ •ํ•ฉ์„ฑ ๊ฒ€์ฆ
    • ๋ชจ๋ฒ” ์‚ฌ๋ก€ ์ ์šฉ
    • ๊ธฐ์กด ์ฝ”๋“œ๋ฒ ์ด์Šค์™€์˜ ํ†ตํ•ฉ

์„ค๊ณ„ ์ฒ ํ•™

"์ข‹์€ ๋ถ€๋ชจ" ํŒจ๋Ÿฌ๋‹ค์ž„

PCIP Framework๋Š” "์ข‹์€ ๋ถ€๋ชจ" ๊ฐœ๋…์—์„œ ์˜๊ฐ์„ ๋ฐ›์•„ ์„ค๊ณ„๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์ข‹์€ ๋ถ€๋ชจ๋Š” ์•„์ด๊ฐ€ ๋ฌด์–ธ๊ฐ€๋ฅผ ํ•˜๊ณ  ์‹ถ์–ดํ•  ๋•Œ ๋‹จ์ˆœํžˆ "์ข‹์•„, ํ•ด๋ด!"๋ผ๊ณ  ๋งํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๋Œ€์‹  ๋‹จ๊ธฐ์  ์˜ํ–ฅ, ์ž ์žฌ์  ๋ฌธ์ œ, ์žฅ๊ธฐ์  ๊ฒฐ๊ณผ๋ฅผ ์˜ˆ์ธกํ•˜๊ณ , ์•„์ด๊ฐ€ ๊ฐ๋‹นํ•  ์ˆ˜ ์žˆ๋‹ค๊ณ  ํŒ๋‹จ๋˜๋ฉด ์•ˆ์ „ํ•œ ๊ฐ€์ด๋“œ๋ผ์ธ์„ ์ œ๊ณตํ•˜์—ฌ ๊ทธ ๊ฒฝ๊ณ„ ๋‚ด์—์„œ ์ž์œ ๋กญ๊ฒŒ ํ™œ๋™ํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•ฉ๋‹ˆ๋‹ค.

์ข‹์€ ๋ถ€๋ชจ ์ถ”๋ก  ๋ฃจํ”„

์ข‹์€ ๋ถ€๋ชจ๋Š” "๋ช…๋ น"์ด ์•„๋‹ˆ๋ผ "์ถ”๋ก "์œผ๋กœ ๊ฐ€์ด๋“œํ•ฉ๋‹ˆ๋‹ค. PCIP์˜ Parent๋Š” ์•„๋ž˜ ์ˆœํ™˜์„ ๋”ฐ๋ฆ…๋‹ˆ๋‹ค.

  • ๊ด€์ฐฐ: ํ˜„์žฌ ์ƒํƒœ์™€ ์ฝ”๋“œ๋ฒ ์ด์Šค, ์ œ์•ฝ, ๋ชฉํ‘œ๋ฅผ ๊ด€์ฐฐ
  • ๊ฐ€์„ค ์ˆ˜๋ฆฝ: ์˜๋„์™€ ๋ฆฌ์Šคํฌ์— ๋Œ€ํ•œ ๊ฐ€์„ค์„ ์„ธ์›€
  • ์งˆ๋ฌธ/ํ™•์ธ: ๋ชจํ˜ธํ•œ ๋ถ€๋ถ„์„ ์งˆ๋ฌธํ•˜๊ฑฐ๋‚˜ ํ…Œ์ŠคํŠธ๋กœ ํ™•์ธ
  • ์™ธ๋ถ€ ์ง€์‹ ๋ณด๊ฐ•: ์ตœ์‹  ๋ฌธ์„œ/ํ‘œ์ค€์„ ์ฐธ์กฐํ•ด ๊ฐ€์„ค์„ ๋ณด๊ฐ•
  • ๊ฐ€๋“œ๋ ˆ์ผ ์„ค์ •: ์•ˆ์ „ํ•œ ๊ฒฝ๊ณ„(๋ถˆ๋ณ€ ์กฐ๊ฑด, ๊ธˆ์ง€์‚ฌํ•ญ, ์ฒดํฌ๋ฆฌ์ŠคํŠธ) ์ •์˜
  • ์œ„์ž„: Child๊ฐ€ ์ž์œจ์ ์œผ๋กœ ์‹คํ–‰ํ•˜๋„๋ก ๊ตฌ์ฒด์  ๊ฒฝ๊ณ„์™€ ์„ฑ๊ณต ์กฐ๊ฑด์„ ์ „๋‹ฌ
  • ๋ชจ๋‹ˆํ„ฐ๋ง: ์‹คํ–‰ ๊ฒฐ๊ณผ์™€ ๋กœ๊ทธ/ํ…Œ์ŠคํŠธ๋กœ ๊ฒ€์ฆ
  • ์กฐ์ •: ํŽธ์ฐจ๊ฐ€ ์žˆ์œผ๋ฉด ๊ฒฝ๊ณ„์™€ ์ „๋žต์„ ์—…๋ฐ์ดํŠธ

๊ฐ„๋‹จ ์˜์‚ฌ์ฝ”๋“œ

observe()
hypothesis = formHypothesis(context, risks)
questions = identifyUnknowns(hypothesis)
resolve(questions) using tests/docs/tools
guardrails = defineBoundaries(hypothesis, standards)
delegateToChild(task, guardrails, successCriteria)
result = verify(outputs, tests, logs)
adjust(guardrails, plan) if driftDetected(result)

์งง์€ ์˜ˆ์‹œ

์š”์ฒญ: "๋กœ๊ทธ์ธ ๋กœ์ง ์ตœ์ ํ™”"
๊ด€์ฐฐ: ์‹คํŒจ์œจ ์ฆ๊ฐ€, DB ๋ถ€ํ•˜ ์ƒ์Šน
๊ฐ€์„ค: ๋ ˆ์ดํŠธ๋ฆฌ๋ฐ‹ยท์„ธ์…˜ํšŒ์ „ ๋ˆ„๋ฝ + N+1 ์ฟผ๋ฆฌ
์™ธ๋ถ€์ง€์‹: OWASP ASVS
๊ฐ€๋“œ๋ ˆ์ผ: 
  - ๋น„๋ฐ€๋ฒˆํ˜ธ ์ทจ๊ธ‰ ๊ธˆ์ง€, ์‹คํŒจ ํšŸ์ˆ˜ ์ž„๊ณ„, ์„ธ์…˜ ํ† ํฐ ํšŒ์ „
  - ORM ์ˆ˜์ค€์—์„œ eager loading ์ ์šฉ, ์ธ๋ฑ์Šค ๊ฒ€์ฆ
์œ„์ž„: Child๊ฐ€ ๋ฆฌํŒฉํ„ฐ + ํ…Œ์ŠคํŠธ ์ถ”๊ฐ€
๋ชจ๋‹ˆํ„ฐ๋ง: ์—๋Ÿฌ์œจ/์ง€์—ฐ/์ฟผ๋ฆฌ์ˆ˜ ๋ฉ”ํŠธ๋ฆญ ํ™•์ธ
์กฐ์ •: ์ž„๊ณ„๊ฐ’ ์ƒํ–ฅยท์ง€์—ฐ ์Šฌ๋ผ์ด๋”ฉ ์œˆ๋„์šฐ ์ ์šฉ

์‚ฌ์ผ๋ŸฐํŠธ ๋ชจ๋“œ(Silent Mode) vs ์ต์Šคํ”Œ๋ฆฌ์‹ฏ ๋ชจ๋“œ(Explicit Mode): ํ•ต์‹ฌ ์ฐจ์ด

์‚ฌ์ผ๋ŸฐํŠธ ๋ชจ๋“œ: ๋‹จ์ˆœ ์ž‘์—…์˜ ์ตœ์  ์†๋„

๋‹จ์ˆœ ์ž‘์—…์€ ์ •ํ™•์„ฑ๊ณผ ์†๋„๋ฅผ ๋™์‹œ์— ๋ชฉํ‘œ๋กœ ํ•ฉ๋‹ˆ๋‹ค. ๊ธฐ๋ณธ ์ƒ์„ฑ ์ž‘์—…์€ ๊ธฐ์กด SOTA ๋ชจ๋ธ๋งŒ์œผ๋กœ๋„ ์ถฉ๋ถ„ํžˆ ํšจ์œจ์ ์œผ๋กœ ์ฒ˜๋ฆฌ๋ฉ๋‹ˆ๋‹ค.

์ต์Šคํ”Œ๋ฆฌ์‹ฏ ๋ชจ๋“œ: ๋ณต์žกํ•œ ์‹œ์Šคํ…œ์„ ์œ„ํ•œ ์ข…ํ•ฉ ๋ถ„์„

Explicit Mode๋Š” ๊ณ ๋‚œ๋„ยท๊ณ ์œ„ํ—˜ยท๋ณต์žก ์—…๋ฌด๋ฅผ ์œ„ํ•ด ์„ค๊ณ„๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ๋‹ค์Œ ์ƒํ™ฉ์—์„œ ์ง„๊ฐ€๋ฅผ ๋ฐœํœ˜ํ•ฉ๋‹ˆ๋‹ค.

  • ํ”„๋กœ๋•์…˜๊ธ‰ ์ฝ”๋“œ๋ฒ ์ด์Šค์˜ ๋ณต์žกํ•œ ์˜์กด์„ฑ
  • ์ƒ์†ยท์ฝœ๋ฐฑ ๊ด€๊ณ„๊ฐ€ ์–ฝํžŒ ๊ตฌ์„ฑ์š”์†Œ
  • ํฌ๋กœ์Šค ํŒŒ์ผ ์ž„ํฌํŠธ/์—ฐ๊ฒฐ์„ฑ
  • ๊ทœ์ • ์ค€์ˆ˜ ์š”๊ตฌ(PCI DSS, HIPAA ๋“ฑ)

์ด๋Ÿฐ ๊ฒฝ์šฐ์—๋Š” ๋ˆˆ์•ž์˜ ์ง€์‹œ๋งŒ ์ฒ˜๋ฆฌํ•ด์„œ๋Š” ์•ˆ ๋˜๊ณ , ์ƒํ˜ธ ์—ฐ๊ฒฐ๋œ ์š”์†Œ๋“ค์„ ํ•จ๊ป˜ ๊ณ ๋ คํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.

PM-Parent-Child ํ˜‘์—… ํ”„๋กœ์„ธ์Šค

1) PM ๋ถ„์„

PM์€ ๋‹ค์Œ์„ ๋ถ„์„ํ•ฉ๋‹ˆ๋‹ค.

  • ์š”์ฒญ์˜ ๋ณต์žก๋„์™€ ๋ฒ”์œ„
  • ๊ธฐ์กด ์ฝ”๋“œ๋ฒ ์ด์Šค ์•„ํ‚คํ…์ฒ˜
  • ์˜ํ–ฅ ๊ฐ€๋Šฅ ์˜์—ญ
  • ์š”๊ตฌ ์ „๋ฌธ ๋ถ„์•ผ

2) Parent ๋ฐฐ์ • ๋ฐ ์ „๋žต ์ˆ˜๋ฆฝ

Parent๊ฐ€ ์„ ์ •๋˜๋ฉด ๋‹ค์Œ์„ ์ˆ˜ํ–‰ํ•ฉ๋‹ˆ๋‹ค.

  • ๋ฒ”์œ„ ์‹๋ณ„: ์˜ํ–ฅ ๋ฒ”์œ„ ์ „๋ฐ˜์„ ์‹๋ณ„
  • ์ „๋žต ์ˆ˜๋ฆฝ: ์ „๋ฌธ๊ฐ€ ๊ด€์ ์—์„œ ์ ‘๊ทผ๋ฒ• ์„ค๊ณ„
  • ์ง€์‹ ๋ณด๊ฐ•: ํ•„์š” ์‹œ ์™ธ๋ถ€ ์ž๋ฃŒ(๊ฒ€์ƒ‰, MCP ๋“ฑ) ํ™œ์šฉ
  • ๊ฐ€๋“œ๋ ˆ์ผ ์ •์˜: Child ์‹คํ–‰์„ ์œ„ํ•œ ์•ˆ์ •์  ๊ฒฝ๊ณ„ ์ œ์‹œ

3) Child ์‹คํ–‰

Child๋Š” ๊ฐ€๋“œ๋ ˆ์ผ ๋‚ด์—์„œ ์‹คํ–‰ํ•˜๋ฉฐ ํ’ˆ์งˆ๊ณผ ์•ˆ์ „์„ ์œ ์ง€ํ•œ ์ฑ„ SOTA ์—ญ๋Ÿ‰์„ ๊ทน๋Œ€ํ™”ํ•ฉ๋‹ˆ๋‹ค.

์™œ ์ค‘์š”ํ•œ๊ฐ€

๊ธฐ์กด AI: "๊ฒฐ์ œ ์‹œ์Šคํ…œ" โ†’ ๋ณด์•ˆ ๋ฏธ๊ณ ๋ ค ๊ธฐ๋ณธ ํผ ์ˆ˜์ค€ PCIP: "๊ฒฐ์ œ ์‹œ์Šคํ…œ" โ†’ ๊ฒฐ์ œ+๋ณด์•ˆ ์ „๋ฌธ๊ฐ€ ํˆฌ์ž… โ†’ PCI ์ค€์ˆ˜ยท์‚ฌ๊ธฐ ๋ฐฉ์ง€ยท๊ฐ์‚ฌ ๋กœ๊น…ยท์—๋Ÿฌ ์ฒ˜๋ฆฌ๊นŒ์ง€ ํฌํ•จํ•œ ์™„๊ฒฐ ํ•ด๋ฒ•


์œ„ํ—˜๋„ ํ‰๊ฐ€ ์‹œ์Šคํ…œ

5๋‹จ๊ณ„ ์œ„ํ—˜๋„ ๋ถ„๋ฅ˜

์ˆ˜์ค€ ์œ„ํ—˜๋„ ๊ธฐ์ค€ ์˜ˆ์‹œ ์‹คํ–‰ ๋ชจ๋“œ
5 ๋งค์šฐ ๋‚ฎ์Œ ๋‹จ์ผ ํŒŒ์ผ, ์˜ํ–ฅ๋„ ๋งค์šฐ ๋‚ฎ์Œ CSS ์Šคํƒ€์ผ ๋ณ€๊ฒฝ, ์ฝ˜์†” ๋กœ๊ทธ ์ถ”๊ฐ€ ์‚ฌ์ผ๋ŸฐํŠธ
4 ๋‚ฎ์Œ 2โ€“3๊ฐœ ํŒŒ์ผ, ๋‹จ์ผ ๋„๋ฉ”์ธ ์ปดํฌ๋„ŒํŠธ ์ˆ˜์ •, ๋‹จ์ˆœ ๊ฒ€์ฆ ์‚ฌ์ผ๋ŸฐํŠธ
3 ๋ณดํ†ต ๋‹ค์ค‘ ํŒŒ์ผ ๋˜๋Š” ๊ต์ฐจ ๋„๋ฉ”์ธ ์‹ ๊ทœ ๊ธฐ๋Šฅ, API ๋ณ€๊ฒฝ ์กฐ๊ฑด๋ถ€
2 ๋†’์Œ ์ค‘์š” ์‹œ์Šคํ…œ ๋ณ€๊ฒฝ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค ์Šคํ‚ค๋งˆ, ๊ฒฐ์ œ ์—ฐ๋™ ์ต์Šคํ”Œ๋ฆฌ์‹ฏ
1 ์น˜๋ช…์  ์‹œ์Šคํ…œ ์ „๋ฐ˜ ๊ฐœํŽธ ์•„ํ‚คํ…์ฒ˜ ๋งˆ์ด๊ทธ๋ ˆ์ด์…˜, ํ”„๋ ˆ์ž„์›Œํฌ ๋ณ€๊ฒฝ ์ต์Šคํ”Œ๋ฆฌ์‹ฏ

์‹คํ–‰ ๋ชจ๋“œ

์‚ฌ์ผ๋ŸฐํŠธ ๋ชจ๋“œ (๋ ˆ๋ฒจ 5, 4)

  • ์‚ฌ์šฉ์ž ์ƒํ˜ธ์ž‘์šฉ ์ตœ์†Œํ™”, ์ฆ‰์‹œ ์‹คํ–‰
  • ๊ฐ„๋‹จ ๊ตฌํ˜„ ์š”์•ฝ ์ œ๊ณต
  • ๊ฐœ๋ฐœ ์†๋„ ์ตœ์ ํ™”

์ต์Šคํ”Œ๋ฆฌ์‹ฏ ๋ชจ๋“œ (๋ ˆ๋ฒจ 1, 2, ๋ณต์žกํ•œ ๋ ˆ๋ฒจ 3)

  • ์ƒ์„ธ ๋ถ„์„๊ณผ ๊ตฌํ˜„ ๊ณ„ํš ์ œ์‹œ
  • ์‹คํ–‰ ์ „ ์‚ฌ์šฉ์ž ์Šน์ธ ํ•„์š”
  • ๋ณ€๊ฒฝ์‚ฌํ•ญ์— ๋Œ€ํ•œ ํฌ๊ด„์  ๋ฌธ์„œํ™”

๊ตฌ์„ฑ

๋™์  ์ „๋ฌธ๊ฐ€ ์„ ํƒ

๋Œ€ํ™” ํ‚ค์›Œ๋“œ ๊ธฐ๋ฐ˜ ์ „๋ฌธ๊ฐ€ ๋ฐฐ์ •

ํ‚ค์›Œ๋“œ ๋ฐฐ์ • ์ „๋ฌธ๊ฐ€
"๋กœ๊ทธ์ธ", "๋ณด์•ˆ", "ํ•ดํ‚น ๋ฐฉ์ง€" ๋ณด์•ˆ ์—”์ง€๋‹ˆ์–ด + ๋ฐฑ์—”๋“œ ์•„ํ‚คํ…ํŠธ
"๋А๋ ค", "์ตœ์ ํ™”", "์†๋„" ์„ฑ๋Šฅ ์—”์ง€๋‹ˆ์–ด + ๋„๋ฉ”์ธ ์ „๋ฌธ๊ฐ€
"์‚ฌ์šฉํ•˜๊ธฐ ์–ด๋ ค์›Œ", "๋””์ž์ธ", "UI" UX ๋””์ž์ด๋„ˆ + ํ”„๋ก ํŠธ์—”๋“œ ์ŠคํŽ˜์…œ๋ฆฌ์ŠคํŠธ
"๊ฒฐ์ œ", "์‡ผํ•‘", "์ƒ๊ฑฐ๋ž˜" ๊ฒฐ์ œ ์‹œ์Šคํ…œ ์ „๋ฌธ๊ฐ€ + ๋ณด์•ˆ ์—”์ง€๋‹ˆ์–ด
"๋ชจ๋ฐ”์ผ", "๋ฐ˜์‘ํ˜•", "์•ฑ" ๋ชจ๋ฐ”์ผ ์ŠคํŽ˜์…œ๋ฆฌ์ŠคํŠธ + UX ๋””์ž์ด๋„ˆ

์™ธ๋ถ€ ์ง€์‹ ํ†ตํ•ฉ

3๋‹จ๊ณ„ ์‹ ๋ขฐ๋„ ์‹œ์Šคํ…œ

  • ๋†’์Œ: ๋‚ด๋ถ€ ์ง€์‹์œผ๋กœ ์ถฉ๋ถ„ โ†’ ์ฆ‰์‹œ ์‘๋‹ต
  • ์ค‘๊ฐ„: ์™ธ๋ถ€ ์ž๋ฃŒ ๊ต์ฐจ๊ฒ€์ฆ โ†’ ๊ฒ€์ฆ๋œ ์‘๋‹ต
  • ๋‚ฎ์Œ: ์ „๋ฌธ ์ž๋ฃŒ ์กฐํšŒ โ†’ ์ •ํ™•ํ•œ ์ •๋ณด

์ง€์› ์™ธ๋ถ€ ์†Œ์Šค

  • ๊ธฐ์ˆ  ๋ฌธ์„œ ๋ฐ API ๋ ˆํผ๋Ÿฐ์Šค
  • ์‚ฐ์—… ํ‘œ์ค€ ๋ฐ ์ปดํ”Œ๋ผ์ด์–ธ์Šค ์š”๊ตฌ
  • ์‹ค์‹œ๊ฐ„ ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ ๋ฒ„์ „ ๋ฐ ๋ณด์•ˆ ๊ณต์ง€
  • ํ”„๋ ˆ์ž„์›Œํฌ ๋ฒ ์ŠคํŠธ ํ”„๋ž™ํ‹ฐ์Šค

ํ’ˆ์งˆ ๋ณด์ฆ

3์ค‘ ๊ฒ€์ฆ ์‹œ์Šคํ…œ

  1. Child ๋ ˆ์ด์–ด: ๋ฌธ๋ฒ•์  ์ •ํ™•์„ฑ ๋ฐ ๊ธฐ๋ณธ ๋™์ž‘ ํ™•์ธ
  2. Parent ๋ ˆ์ด์–ด: ๋„๋ฉ”์ธ๋ณ„ ํ’ˆ์งˆ ๊ธฐ์ค€ ๊ฒ€ํ† 
  3. PM ๋ ˆ์ด์–ด: ํ”„๋กœ์ ํŠธ ์ผ๊ด€์„ฑ ๋ฐ ์•„ํ‚คํ…์ฒ˜ ์ •ํ•ฉ์„ฑ ํ™•์ธ

์„ฑ๋Šฅ ๋น„๊ต

๊ฐœ๋ฐœ ํ”„๋กœ์„ธ์Šค

๋‹จ๊ณ„ ๊ธฐ์กด AI PCIP
์ดˆ๊ธฐ ์š”์ฒญ ๋น ๋ฅธ ๊ตฌํ˜„ ๋ถ„์„ ๋ฐ ๊ณ„ํš
๊ฒฐ๊ณผ ํ’ˆ์งˆ ๋ถ€๋ถ„์ /๋ถˆ์™„์ „ ์™„๊ฒฐยท๋งฅ๋ฝ ์ธ์ง€
๋ฐ˜๋ณต ์ˆ˜์ • ๋‹ค์ˆ˜์˜ ์žฌ์ˆ˜์ • ํ•„์š” ์ตœ์†Œํ•œ์˜ ๋ณด์ •
์ „์ฒด ์†Œ์š” 1์‹œ๊ฐ„ ์ž‘์—… โ†’ 1์ฃผ ์ดˆ๋ฐ˜ ์•ฝ๊ฐ„ ์ง€์—ฐ โ†’ ์ „์ฒด์ ์œผ๋กœ ๋” ๋น ๋ฅธ ์™„๋ฃŒ

์žฅ์ 

โœ… ์„ค๋ช… ๋ถ€๋‹ด ๊ฐ์†Œ: ๋งฅ๋ฝ ์ธ์ง€ ์ฒ˜๋ฆฌ๋กœ ๋ฐ˜๋ณต ์ง€์‹œ ์ตœ์†Œํ™”
โœ… ์ „๋ฌธ๊ฐ€ ๊ด€์  ์ ์šฉ: ๋„๋ฉ”์ธ ์ „๋ฌธ์„ฑ ์ž๋™ ๋ฐ˜์˜
โœ… ๋†’์€ ์ฝ”๋“œ ํ’ˆ์งˆ: ๋‹ค์ธต ๊ฒ€์ฆ์œผ๋กœ ๊ฒฌ๊ณ ํ•œ ๊ฒฐ๊ณผ
โœ… ์‚ฌํ›„ ๋ฌธ์ œ ๊ฐ์†Œ: ์‚ฌ์ „์  ์—ฃ์ง€์ผ€์ด์Šค ๊ณ ๋ ค
โœ… ํ”„๋กœ์ ํŠธ ์ผ๊ด€์„ฑ: ์•„ํ‚คํ…์ฒ˜ ์ •ํ•ฉ์„ฑ ์œ ์ง€

ํ•œ๊ณ„

โŒ ์ดˆ๊ธฐ ๋ณต์žก๋„: ๋‹จ์ˆœ ์ž‘์—…์—” ์„ค์ • ๋ถ€๋‹ด(ํ•˜์ง€๋งŒ Silent Mode๋กœ ์™„ํ™”)
โŒ ํ•™์Šต ๊ณก์„ : ์œ„ํ—˜๋„ ์‹œ์Šคํ…œ ์ดํ•ด ํ•„์š”
โŒ AI ํ•œ๊ณ„: ๋ณธ์งˆ์  ํ•œ๊ณ„๋Š” ์กด์žฌ


๋ณ€๊ฒฝ ๋กœ๊ทธ

  • 2.1: SystemPrompt.md/SystemPromptEN.md ์—…๋ฐ์ดํŠธ(์ง€์‹œ๋ฌธ ๋ช…ํ™•ํ™”, ์šฉ์–ด ํ†ต์ผ, ๋ชจ๋“œ ์ „ํ™˜ ๊ธฐ์ค€ ์ •๊ตํ™”). README ๋ฒ„์ „ ๋ฐฐ์ง€ ๊ฐฑ์‹ .
  • 2.0: ๋‹ค๊ตญ์–ด README ์ •๋ฆฌ, ์œ„ํ—˜๋„/๋ชจ๋“œ ํ‘œ์ค€ํ™”, ํ’ˆ์งˆ ๋ณด์ฆ ์ฒด๊ณ„ ๋ช…์‹œ.

์„ค์น˜ ๋ฐ ์„ค์ •

์‚ฌ์ „ ์ค€๋น„๋ฌผ

  • ์‹œ์Šคํ…œ ํ”„๋กฌํ”„ํŠธ๋ฅผ ์ง€์›ํ•˜๋Š” AI ์–ด์‹œ์Šคํ„ดํŠธ(Claude, ChatGPT ๋“ฑ)
  • ๊ฐœ๋ฐœ ํ™˜๊ฒฝ(Cursor, VS Code ๋“ฑ)

๋น ๋ฅธ ์‹œ์ž‘

  1. ์‹œ์Šคํ…œ ํ”„๋กฌํ”„ํŠธ ๊ฐ€์ ธ์˜ค๊ธฐ

    SystemPrompt.md ๋‚ด์šฉ์„ ๋ณต์‚ฌํ•ฉ๋‹ˆ๋‹ค
    
  2. AI ์–ด์‹œ์Šคํ„ดํŠธ ์„ค์ •

    • Cursor: User Rules ๋˜๋Š” Project Rules์— ์ถ”๊ฐ€
    • ChatGPT: Custom Instructions์— ์„ค์ •
    • Claude: System Prompt๋กœ ์‚ฌ์šฉ
  3. ์ฝ”๋”ฉ ์‹œ์ž‘

User: "์ธ์ฆ ์‹œ์Šคํ…œ ๋งŒ๋“ค์–ด์ค˜"
PCIP: [PM ๋ถ„์„] โ†’ [๋ณด์•ˆ ์ „๋ฌธ๊ฐ€ ๋ฐฐ์ •] โ†’ [์‹คํ–‰]

์‚ฌ์šฉ ์˜ˆ์‹œ

์˜ˆ์‹œ 1: ๋‹จ์ˆœ ์ž‘์—…(Silent Mode)

User: "๋ฒ„ํŠผ ์ƒ‰ ๋นจ๊ฐ„์ƒ‰์œผ๋กœ ๋ฐ”๊ฟ”์ค˜"
PCIP: [์ฆ‰์‹œ CSS ์—…๋ฐ์ดํŠธ]
์ฐธ๊ณ : ์ €์œ„ํ—˜ ์Šคํƒ€์ผ ๋ณ€๊ฒฝ ์™„๋ฃŒ

์˜ˆ์‹œ 2: ๋ณต์žก ์ž‘์—…(Explicit Mode)

User: "๊ฒฐ์ œ ์ฒ˜๋ฆฌ ์‹œ์Šคํ…œ ๋งŒ๋“ค์–ด์ค˜"
PCIP: 
- ์œ„ํ—˜๋„: Critical(Level 1)
- ์ „๋ฌธ๊ฐ€: ๊ฒฐ์ œ + ๋ณด์•ˆ + ๋ฐฑ์—”๋“œ
- ๋ถ„์„: PCI DSS ์ค€์ˆ˜ ํ•„์š”...
- ์ง„ํ–‰ํ• ๊นŒ์š”? Y/n

์˜ˆ์‹œ 3: ๋‹ค์ค‘ ์ „๋ฌธ๊ฐ€ ํ˜‘์—…

User: "๋ชจ๋ฐ”์ผ ์ปค๋จธ์Šค ์•ฑ(๋ณด์•ˆ ๊ฒฐ์ œ ํฌํ•จ) ๋งŒ๋“ค์–ด์ค˜"
PCIP: 
- PM: ๋†’์€ ๋ณต์žก๋„ ๊ฐ์ง€
- ์ „๋ฌธ๊ฐ€: ๋ชจ๋ฐ”์ผ + UX + ๊ฒฐ์ œ + ๋ณด์•ˆ
- ์ ‘๊ทผ: ํ† ํฐํ™” ๊ฒฐ์ œ๋ฅผ ๊ฐ–์ถ˜ PWA...

์˜ˆ์‹œ 4: Parent ์˜๊ฒฌ ์š”์ฒญ

User: "PCIP Parent ์˜๊ฒฌ์€?"
PCIP Parent(๋ณด์•ˆ ์—”์ง€๋‹ˆ์–ด): 
"ํ˜„์žฌ ์ธ์ฆ ๊ตฌํ˜„์„ ๊ธฐ์ค€์œผ๋กœ ๋กœ๊ทธ์ธ ์‹œ๋„ ๋ ˆ์ดํŠธ๋ฆฌ๋ฐ‹, ์•ˆ์ „ํ•œ ์„ธ์…˜ ํ† ํฐ ๋กœํ…Œ์ด์…˜์„ ๊ถŒ์žฅํ•ฉ๋‹ˆ๋‹ค..."

๊ตฌ์„ฑ ์˜ต์…˜

์œ„ํ—˜๋„ ์ปค์Šคํ„ฐ๋งˆ์ด์ง•

ํ”„๋กœ์ ํŠธ ์„ฑ๊ฒฉ์— ๋งž์ถฐ ์‹œ์Šคํ…œ ํ”„๋กฌํ”„ํŠธ์˜ ์ž„๊ณ„๊ฐ’์„ ์กฐ์ •ํ•˜์„ธ์š”.

  • ๋ณด์ˆ˜์ : Explicit Mode ์ž„๊ณ„๊ฐ’ ๋‚ฎ์ถค
  • ๊ณต๊ฒฉ์ : Silent Mode ์ž„๊ณ„๊ฐ’ ๋†’์ž„

์ „๋ฌธ๊ฐ€ ์„ ํƒ ์˜ค๋ฒ„๋ผ์ด๋“œ

์„ ํ˜ธ ์ „๋ฌธ๊ฐ€๋ฅผ ์ˆ˜๋™ ์ง€์ •ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

User: "๋กœ๊ทธ์ธ ์‹œ์Šคํ…œ(UX ์ค‘์‹ฌ)"
PCIP: UX ๋””์ž์ด๋„ˆ + ๋ฐฑ์—”๋“œ ์•„ํ‚คํ…ํŠธ ๋ฐฐ์ •

Contributing

  1. Fork the repository
  2. Create a feature branch
  3. Submit a pull request with detailed description

License

MIT License - see LICENSE file for details

Support

  • Issues: Report bugs and feature requests via GitHub Issues
  • Discussions: Share experiences and ask questions in GitHub Discussions
  • Documentation: Check SystemPrompt.md for detailed implementation

English

PCIP Framework

Parent-Child Instruction Processing Framework for AI Coding Assistants

License: MIT Version

Overview

PCIP (Parent-Child Instruction Processing) is a three-layer architecture that improves AI coding assistants' context understanding and code quality.

3-Line Summary

  • Your AI, Tiered: PCIP uses a PM โ†’ Parent (expert) โ†’ Child (worker) structure. The PM sizes up the job, then assigns the right Parent.
  • Need a Brain? (Explicit Mode): For gnarly, high-stakes tasks, the PM brings in a Parent to design a solid strategy.
  • Just Get It Done (Silent Mode): For basic, straightforward work, the Child handles it solo for maximum speed.

Problem Statement

Context Awareness Gaps in Current AI Assistants

  • Code generation that ignores overall architecture and domain requirements
  • Inconsistent with the existing codebase (security, performance, compliance)
  • High iteration overhead due to partial/incomplete implementations

Example (typical pitfalls when asked to โ€œBuild payment systemโ€):

  • Missing security (PCI DSS), error handling/rollback, fraud prevention, audit logging, refunds

Solution Architecture

Traditional vs PCIP

Traditional: User โ†’ AI โ†’ Code โ†’ Risky output
PCIP: User โ†’ PM (Analysis) โ†’ Parent (Guidance) โ†’ Child (Execution)

Layer Specifications

  • Meta Layer: PM
    • Conversational analysis, risk assessment (5 levels), expert team composition, external resource coordination
  • Parent Layer: Domain Expert
    • Security, Performance, UX, Payments, DevOps, Mobile, etc.; defines standards and guidance
  • Child Layer: Execution Engine
    • Implements precisely within guidance; ensures syntactic correctness and integration quality

Design Philosophy

The โ€œGood Parentโ€ Paradigm

A good parent anticipates short-term impacts, potential problems, and long-term consequences. If the task is within the childโ€™s capacity, they set safe boundaries so the child can act freely within them. Otherwise, they tighten guardrails, add tools, or expand context.

Good Parent Inference Loop

  • Observe โ†’ Hypothesize โ†’ Ask/Validate โ†’ Enrich with External Knowledge โ†’ Set Guardrails โ†’ Delegate โ†’ Monitor โ†’ Adjust

Pseudocode:

observe()
hypothesis = formHypothesis(context, risks)
unknowns = identifyUnknowns(hypothesis)
resolve(unknowns) via tests/docs/tools
guardrails = defineBoundaries(hypothesis, standards)
delegate(task, guardrails, successCriteria)
result = verify(outputs, tests, logs)
adjust(guardrails, plan) if driftDetected(result)

Short example (Login optimization):

  • Symptoms: Higher failure rate, DB load; Hypothesis: Missing rate-limit/session rotation + N+1 queries
  • External: OWASP ASVS; Guardrails: No plaintext passwords, attempt thresholds, token rotation, ORM eager loading
  • Delegate: Refactor + tests; Monitor: error/latency/query metrics; Adjust: thresholds + sliding window

Silent vs Explicit Mode

  • Silent (Levels 5โ€“4): Fast path for simple tasks; minimal interaction; brief summary
  • Explicit (Levels 1โ€“2, complex Level 3): Detailed analysis, plan, user approval, comprehensive documentation

Risk Assessment System

Level Risk Criteria Examples Mode
5 Very Low Single file, minimal impact CSS styling, console logs Silent
4 Low 2โ€“3 files, single domain Component updates, simple validation Silent
3 Moderate Multi-file or cross-domain New features, API changes Conditional
2 High Major system changes Database schema, payment integration Explicit
1 Critical System-wide overhaul Architecture migration, framework changes Explicit

Configuration

Dynamic Expert Selection (by conversational cues)

  • โ€œloginโ€, โ€œsecurityโ€, โ€œhardeningโ€ โ†’ Security Engineer + Backend Architect
  • โ€œslowโ€, โ€œoptimizeโ€, โ€œlatencyโ€ โ†’ Performance Engineer + Domain Expert
  • โ€œdesignโ€, โ€œUI/UXโ€, โ€œusabilityโ€ โ†’ UX Designer + Frontend Specialist
  • โ€œpaymentsโ€, โ€œcommerceโ€ โ†’ Payment Systems Expert + Security Engineer
  • โ€œmobileโ€, โ€œresponsiveโ€, โ€œappโ€ โ†’ Mobile Specialist + UX Designer

External Knowledge Integration

  • Confidence levels: High (direct) / Medium (verify with docs) / Low (consult specialized sources)
  • Sources: Technical docs, standards/compliance, real-time library versions, framework best practices

Quality Assurance (Triple Verification)

  1. Child: Syntactic correctness, baseline behavior
  2. Parent: Domain standards (security, performance, UX)
  3. PM: Architectural consistency and project-wide coherence

Performance Comparison

Phase Traditional AI PCIP Framework
Initial Request Quick implementation Analysis & Planning
Result Quality Partial/Incomplete Complete & Context-aware
Iteration Cycles Multiple fixes required Minimal revisions needed
Overall Time 1-hour task โ†’ 1 week Slightly slower start โ†’ Faster completion

Benefits: Reduced explanation overhead, expert perspectives, higher quality, fewer regressions, consistent architecture.
Limitations: Initial complexity for simple tasks (mitigated by Silent Mode), small learning curve, inherent AI constraints.


Changelog

  • 2.1: Updated SystemPrompt.md/SystemPromptEN.md (clearer instructions, terminology consistency, refined mode switching criteria). Updated README version badge.
  • 2.0: Organized bilingual README, standardized risk/mode tables, documented QA workflow.

Installation & Setup

Prerequisites: AI assistant supporting system prompts (Claude, ChatGPT, etc.), dev environment (Cursor/VS Code).

Quick Start:

  1. Copy contents of SystemPrompt.md
  2. Configure your AI assistant (Cursor: User/Project Rules; ChatGPT: Custom Instructions; Claude: System Prompt)
  3. Start coding
User: "Build a user authentication system"
PCIP: [PM analyzes] โ†’ [Security Parent assigned] โ†’ [Child implements]

Usage Examples:

  • Silent Mode:
User: "Change button color to red"
PCIP: Immediate CSS update; Note: low-risk styling change
  • Explicit Mode:
User: "Build payment processing system"
PCIP: Level 1 (Critical); Parents: Payments + Security + Backend; Plan with PCI DSS compliance โ†’ Proceed? Y/n
  • Ask Parent Opinion:
User: "What does the PCIP parent say?"
Parent (Security): "Add rate limiting and secure session rotation..."

Configuration Options

  • Risk threshold tuning (Conservative vs Aggressive)
  • Expert selection override (e.g., โ€œfocus on UXโ€ โ†’ UX Designer + Backend Architect)

Contributing / License / Support

  • Contributing: Fork โ†’ Feature branch โ†’ PR with details
  • License: MIT (see LICENSE)
  • Support: Issues / Discussions; docs in SystemPrompt.md

About

[System Prompt] Parent-Child Instruction Processing (PCIP) Framework with conversational learning, external knowledge integration, and dynamic expert leadership for optimal coding assistance.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published