docs: 보스 설계 문서와 Codex YAML 도구 추가

- 프로젝트 AGENTS 지침에 커밋 후 푸시 확인 규칙과 런타임 콘텐츠 에이전트 규칙을 보강
- 보스 설계 플레이북과 멀티플레이 보스 설계 철학 문서를 추가해 사람용 설계 기준을 정리
- Codex-Tools 디렉터리에 Markdown→YAML→검증 파이프라인 스크립트와 프롬프트, 스키마, 설정 파일을 추가
This commit is contained in:
2026-04-06 14:03:49 +09:00
parent c48d1bee52
commit c8edf838fd
12 changed files with 1201 additions and 1 deletions

View File

@@ -14,7 +14,7 @@ Multiplayer arena game built with **Unity 6000.3.10f1** and **Unity Netcode for
Design docs are maintained in Obsidian Vault: `/mnt/smb/Obsidian Vault/Colosseum` Design docs are maintained in Obsidian Vault: `/mnt/smb/Obsidian Vault/Colosseum`
- Always use the shared Obsidian Vault at `/mnt/smb/Obsidian Vault/Colosseum` for project documentation updates. - Always use the shared Obsidian Vault at `/mnt/smb/Obsidian Vault/Colosseum` for project documentation updates.
- **Obsidian Vault 파일은 반드시 파일 도구(read/write/edit)로만 접근한다.** cmd, powershell, bash 등은 UNC 경로를 정상적으로 처리하지 못하므로 사용하지 않는다. - 현재 환경에서 접근 가능한 셸(`bash`, `zsh` 등) 기준으로 Obsidian Vault 파일을 열고 수정해도 된다.
- Unless the user explicitly names another file, `체크리스트` means `/mnt/smb/Obsidian Vault/Colosseum/개발/프로토타입 체크리스트.md`. - Unless the user explicitly names another file, `체크리스트` means `/mnt/smb/Obsidian Vault/Colosseum/개발/프로토타입 체크리스트.md`.
- After completing work, always check whether the prototype checklist should be updated and reflect the new status when needed. - After completing work, always check whether the prototype checklist should be updated and reflect the new status when needed.
- If the work also changes a domain-specific checklist such as boss/player/combat, sync that checklist too. - If the work also changes a domain-specific checklist such as boss/player/combat, sync that checklist too.
@@ -337,6 +337,7 @@ public class NetworkedComponent : NetworkBehaviour
- After Unity-related edits, check the Unity console for errors before proceeding. - After Unity-related edits, check the Unity console for errors before proceeding.
- For networked play tests, prefer a temporary non-conflicting test port when needed and restore the default port after validation. - For networked play tests, prefer a temporary non-conflicting test port when needed and restore the default port after validation.
- The user has a strong project preference that play mode must be stopped before edits because network ports can remain occupied otherwise. - The user has a strong project preference that play mode must be stopped before edits because network ports can remain occupied otherwise.
- When the user requests a commit, create the commit and show the message, then **ask before pushing**. Only push after explicit confirmation.
- Commit messages should follow the recent project history style: use a type prefix such as `feat:`, `fix:`, or `chore:` and write the subject in Korean so the gameplay/UI/system change is clear from the log alone. - Commit messages should follow the recent project history style: use a type prefix such as `feat:`, `fix:`, or `chore:` and write the subject in Korean so the gameplay/UI/system change is clear from the log alone.
- When the change is substantial, include a blank line after the subject and add Korean bullet points that summarize the work by feature/purpose, similar to commit `0889bb0` (`feat: 드로그 집행 개시 패턴 및 낙인 디버프 추가`). - When the change is substantial, include a blank line after the subject and add Korean bullet points that summarize the work by feature/purpose, similar to commit `0889bb0` (`feat: 드로그 집행 개시 패턴 및 낙인 디버프 추가`).
- Prefer the format `type: 한글 요약` for the subject, then `- 변경 사항` bullet lines for the body. Use `feat:` for feature work, `fix:` for bug fixes, and `chore:` for non-feature maintenance such as scene/prefab cleanup, asset reorganization, or other miscellaneous upkeep. The body should mention key implementation scope, affected systems/assets, and validation or rollback notes when relevant. - Prefer the format `type: 한글 요약` for the subject, then `- 변경 사항` bullet lines for the body. Use `feat:` for feature work, `fix:` for bug fixes, and `chore:` for non-feature maintenance such as scene/prefab cleanup, asset reorganization, or other miscellaneous upkeep. The body should mention key implementation scope, affected systems/assets, and validation or rollback notes when relevant.
@@ -348,3 +349,19 @@ public class NetworkedComponent : NetworkBehaviour
- 프로젝트 용어로는 `필살기`보다 `고위력 기술`을 우선 사용한다. - 프로젝트 용어로는 `필살기`보다 `고위력 기술`을 우선 사용한다.
- 보호막은 단일 값이 아니라 `타입별 독립 인스턴스`로 취급한다. 같은 타입은 자기 자신만 갱신되고, 서로 다른 타입은 공존하며, 흡수 순서는 적용 순서를 따른다. - 보호막은 단일 값이 아니라 `타입별 독립 인스턴스`로 취급한다. 같은 타입은 자기 자신만 갱신되고, 서로 다른 타입은 공존하며, 흡수 순서는 적용 순서를 따른다.
- 보스 시그니처 전조는 가능한 한 정확한 진행 수치 UI보다 명확한 모션/VFX로 읽히게 한다. 차단 진행도나 정확한 누적 수치 노출은 명시 요청이 없으면 피한다. - 보스 시그니처 전조는 가능한 한 정확한 진행 수치 UI보다 명확한 모션/VFX로 읽히게 한다. 차단 진행도나 정확한 누적 수치 노출은 명시 요청이 없으면 피한다.
# Runtime Content Agent Rules
Purpose:
Keep boss YAML compatible with game runtime expectations.
Always:
- validate boss YAML shape before commit
- preserve `boss_id` stability
- keep runtime field names machine-readable and snake_case
- prefer additive changes over destructive changes
Never:
- change mechanic intent to satisfy schema formatting
- silently rename IDs
- auto-fill required combat values without a design source

309
Boss_Design_Playbook.md Normal file
View File

@@ -0,0 +1,309 @@
# Boss & Encounter Design Playbook
이 문서는 **4인 파티 협동 / 자유 역할 비율 / 논타겟 액션 / 보스 러쉬 / 점진 성장형 로그라이트** 게임을 위한 사람용 설계 문서다.
핵심 철학:
- 보스는 체력 덩어리가 아니라 **역할 수행과 패턴 해결을 시험하는 협동 퍼즐**이다.
- 전투 난이도는 운 좋은 고성능 빌드보다 **역할 분배, 포지셔닝, 기믹 처리, 개인 수행**을 우선 시험해야 한다.
- 시그니처 패턴은 형태보다 **문제와 해법이 명확해야** 한다.
- 긴 전투(약 10분)는 **3~4 페이즈 이상**과 **주기적 새 메커닉 도입**이 필요하다.
---
## 1. 게임 전제
### 파티 구조
- 4인 파티 고정 상정
- 자유 빌드형 역할 분배
- 큰 축: 탱커 / 딜러 / 지원
- 각 플레이어는 역할 비율로 기여 가능
- 예: Tank 0.8 / DPS 0.2 / Support 0.0
### 역할 의미
- **탱커**: 최고 위협 유지, 보스 방향/위치 안정화, 이동 후 재정렬
- **딜러**: 지속 DPS, 순간 DPS 체크, 기믹 처리 화력
- **지원**: 힐, 보호막, 버프, 자원 회복, 상태이상 해제, 부활
### 전투 구조
- 자유 이동, 3인칭, 논타겟 액션
- 공용 회피(구르기) + 약간의 무적 시간
- 블록은 완전 방어보다 피해 감소 성격
- 보스 타겟 선정은 기본적으로 Aggro/Threat 기반, 일부 패턴은 랜덤/거리 기반
- 작은 패턴의 지속 압박 + 큰 패턴의 시험 구조
- 전투 중 부활 존재, 부활은 긴 캐스팅을 요구
- 명시적 시간 제한은 지양하지만, 무한 전투 방지를 위해 보스는 점진 강화 가능
### 런 구조
- 기본 세팅 상점
- [보스 전투 → 랜덤/고정 보상 획득 → 상점/성장] * n
- 보스 수 3~4
- 보스 1회 전투 약 10분 목표
- 스테이지는 순차 증가
- 스테이지 1개 = 기본적으로 보스 전투 1회
- 한 번의 보스 전투는 단일 개체가 아닐 수 있음 (adds, 다중 보스, 군집형 가능)
### 성장 구조
- 고정 보상: 재화, 패시브 스킬 포인트
- 랜덤 보상: 장비 드랍, 상점 품목, 선택형 강화 등
- 빌드 변화 강도: Medium
- 역할 전환은 가능하나, 시작 역할에서 완전히 벗어나기는 어려움
- 파워 그래프는 점진적 증가
- 보스는 운빨이 아니라 **역할 분배와 수행 능력**을 검증해야 함
---
## 2. 설계 철학 요약
### 2.1 보스 = Encounter
이 게임에서 보스는 단일 개체가 아니라 **전투 상황 전체**다.
### 2.2 Role Pressure
좋은 보스는 역할을 이름으로 요구하지 않고, **각 역할이 해결해야 하는 문제를 지속적으로 만든다.**
- Tank Pressure: 고정 공격, 위협 유지, 위치 안정화
- Support Pressure: 광역 피해 복구, 상태이상 해제, 지속 유지력 소모 대응
- DPS Pressure: 지속 화력, 순간 화력, 기믹 대상 파괴
### 2.3 Single-Core Solve Pattern
시그니처 패턴은 여러 종류를 뒤섞기보다,
- 문제 1개
- 해법 1개
- 실패 이유 1개
처럼 명확해야 한다.
### 2.4 Sustain + Burst
좋은 보스는 둘 다 필요하다.
- **Sustain Pressure**: 쉬지 않는 작은 압박
- **Burst Check**: 큰 패턴, 순간 DPS, 즉사급 실패 위험
### 2.5 Revive is Risk
부활은 무료 복구가 아니라 위험 행동이어야 한다.
### 2.6 Soft Enrage
명시적 시간 제한 대신,
- 공격 속도 증가
- 피해 증가
- 패턴 밀도 증가
같은 점진 강화로 무한 지속을 방지한다.
---
## 3. Signature-first 설계 순서
보스 설계는 아래 순서로 진행한다.
1. **시그니처 패턴**
2. 역할 압박 정의
3. 페이즈 구조
4. 지속 압박과 순간 시험
5. 부활 상호작용
6. 시간 경과 강화
7. 개체 구성 (단일 / adds / 다중)
8. 난이도 목표
---
## 4. 설계 템플릿
```md
# BOSS DESIGN DOCUMENT
Boss Name:
Boss ID:
Stage:
Estimated Duration:
Entity Type:
- Single
- Boss + Adds
- Multi Boss
- Swarm
---
# 1. Signature Pattern
## Signature Name:
## Core Concept
이 패턴이 만드는 전투 상황.
## Player Problem
플레이어가 해결해야 하는 문제.
## Expected Solution
정답. 역할별 책임을 가능하면 명시.
## Failure Result
틀렸을 때 발생하는 결과.
## Why This Pattern Matters
이 패턴이 왜 이 보스의 정체성인지.
---
# 2. Role Pressure Design
## Tank Pressure
지속적으로 요구되는 행동:
Tank Failure Result:
## Support Pressure
지속적으로 요구되는 행동:
Support Failure Result:
## DPS Pressure
지속적으로 요구되는 행동:
DPS Failure Result:
---
# 3. Phase Structure
## Phase 1 — Learn
목표:
새 요소:
## Phase 2 — Pressure
목표:
새 요소:
## Phase 3 — Twist
목표:
새 요소:
## Phase 4 — Final
목표:
새 요소:
---
# 4. Sustained Pressure Design
## Passive Damage Sources
## Movement Pressure
---
# 5. Burst Pressure Design
## Major Mechanics
## DPS Check
Yes / No
---
# 6. Revive Interaction Design
Revive Trigger Reaction:
Revive Risk Level:
Low / Medium / High
---
# 7. Time Escalation (Soft Enrage)
Escalation Type:
Escalation Interval:
---
# 8. Multi-Entity Behavior
Entity Count:
Add Role:
---
# 9. Expected Party Role Demand
Required Tank Pressure:
Required DPS Pressure:
Required Support Pressure:
---
# 10. Difficulty Intent
Expected Attempts:
Primary Failure Cause:
Secondary Failure Cause:
```
---
## 5. 역할 압박을 만드는 방법
### 탱커 압박
- 최고 위협 대상에게 지속 고정 공격
- 보스 방향 유지 실패 시 파티 피해 증가
- 이동 패턴 후 재고정 요구
- 위협 유지 실패 시 후열 붕괴
### 지원 압박
- 탱커 유지력 소모
- 광역 피해 누적
- 상태이상 해제 필요
- 부활 시전 자체가 위험 유발
### 딜러 압박
- 탱커 외 대상에게도 위협이 가해짐
- 순간 DPS 요구 기믹
- 적절한 시간 내 처치 요구
- 지속 화력 부족 시 전투 피로도 급증
---
## 6. 긴 보스전 체크리스트
약 10분 전투라면 최소한 아래를 점검한다.
- 3개 이상 페이즈가 있는가
- 90~120초마다 새 메커닉이 들어오는가
- 시그니처 패턴이 후반부에 변형되는가
- 지속 압박과 순간 시험이 모두 있는가
- 탱커가 계속 필요한가
- 지원이 계속 필요한가
- 딜러가 지속/순간 DPS를 둘 다 요구받는가
- 부활이 안전하지 않은가
- 시간 경과 강화가 있는가
- adds나 다중 개체가 있다면 명확한 역할이 있는가
---
## 7. 나쁜 보스의 전형적 징후
- 모든 패턴을 같은 타이밍 Dodge로 해결 가능
- 탱커가 잠깐만 필요하고 대부분 무시 가능
- 지원이 없어도 큰 차이가 없음
- 딜러는 평딜만 하면 되고 순간 화력 체크가 없음
- 부활이 사실상 무료
- 시간만 끌면 결국 깨짐
- adds가 있지만 의미가 없음
- 시그니처 패턴이 모호해서 실패 이유가 불명확
---
## 8. 추천 실무 루프
1. 시그니처 패턴 1개를 먼저 적는다
2. 그 패턴이 어떤 역할을 필요하게 만드는지 적는다
3. 그 패턴이 어느 페이즈에서 어떻게 변형되는지 적는다
4. 작은 패턴으로 유지력 압박을 설계한다
5. 부활/시간경과 리스크를 넣는다
6. YAML로 변환한다
7. Codex로 검증한다
8. 보고서를 보고 문서를 다시 고친다

18
Codex-Tools/AGENTS.md Normal file
View File

@@ -0,0 +1,18 @@
# Codex Automation Agent Rules
Purpose:
Run the design-to-data pipeline for boss encounters.
Pipeline:
1. Read Markdown boss design doc from the design vault.
2. Convert it to boss YAML.
3. Validate the YAML using boss design rules.
4. Write a validation report next to the design doc.
5. Write runtime YAML into the game project.
Rules:
- Markdown is the source of truth.
- YAML is derived data.
- Validation should never modify the source Markdown.
- Do not invent mechanics.
- If a field is missing, record that as uncertainty or an issue.

22
Codex-Tools/README.md Normal file
View File

@@ -0,0 +1,22 @@
# Codex Boss Pipeline Tools
This directory contains helper scripts for a Markdown -> YAML -> Validation workflow.
## Files
- `skills/generate_boss_yaml.py`: generate runtime YAML from a Markdown boss design document.
- `skills/validate_boss_yaml.py`: validate a generated boss YAML and emit a Markdown report.
- `skills/run_boss_pipeline.py`: run both steps together.
- `config/paths.example.yaml`: copy and edit into `paths.yaml` for your environment.
## Recommended flow
1. Write `/mnt/smb/Obsidian Vault/Colosseum/보스/<boss>/<boss> 기획.md`
2. Run `run_boss_pipeline.py`
3. Review `<boss>_validation.md`
4. Edit the Markdown source, then rerun.
## Example
```bash
python skills/run_boss_pipeline.py \
--md "/mnt/smb/Obsidian Vault/Colosseum/보스/드로그/드로그 기획.md" \
--config config/paths.yaml
```

View File

@@ -0,0 +1,8 @@
# Fixed configuration keys for the boss pipeline.
# These names are now the canonical version used by all Python skills.
design_root: "/mnt/smb/Obsidian Vault/Colosseum"
game_project_root: "/home/dal4segno/Colosseum"
schema_path: "/home/dal4segno/Colosseum/tools/schemas/boss_schema_v1.yaml"
runtime_boss_output_dir: "/home/dal4segno/Colosseum/content/bosses"
validation_report_suffix: "_validation.md"

View File

@@ -0,0 +1,11 @@
Read the provided Markdown boss design document and convert it into YAML using the boss schema.
Rules:
- Do not invent mechanics.
- Preserve signature mechanic intent.
- Preserve tank / support / dps role pressure intent.
- If a field is missing, leave it empty or null rather than fabricating.
- Prefer machine-readable snake_case values.
- Preserve uncertainty in `source_notes` or `notes` fields.
Return only valid YAML.

View File

@@ -0,0 +1,8 @@
Validate the provided boss YAML against AGENTS.md and the project design philosophy.
Return a Boss Validation Report with:
- PASS / WARNING / FAIL
- scores
- detected issues
- suggested fixes
- explicit note of any missing source information

View File

@@ -0,0 +1,82 @@
# Boss YAML Schema v1
# Authoring target for Codex-readable encounter data.
boss:
metadata:
boss_id: "example_boss_id"
boss_name: "Example Boss"
stage: 1
expected_duration_sec: 600
entity_type: "single" # single | boss_plus_adds | multi_boss | swarm
signature:
name: "Example Signature"
concept: >
One-sentence summary of the signature encounter concept.
player_problem: >
What players must solve.
expected_solution:
tank: []
support: []
dps: []
failure_result:
type: "major_damage" # major_damage | wipe | mechanic_repeat | positioning_loss | other
description: "What happens on failure."
roles:
tank:
pressure_sources: []
failure_results: []
required_pressure: 1.0
support:
pressure_sources: []
failure_results: []
required_pressure: 1.0
dps:
pressure_sources: []
failure_results: []
required_pressure: 1.0
phases:
- phase_id: 1
name: "Learn"
duration_estimate_sec: 120
new_mechanics: []
role_pressure_multiplier:
tank: 1.0
support: 1.0
dps: 1.0
pressure:
sustained:
sources: []
burst:
mechanics: []
dodge_types_required: []
revive:
trigger_response:
enabled: true
reactions: []
revive_risk_level: "medium" # low | medium | high
escalation:
enabled: true
escalation_interval_sec: 60
escalation_types: []
entities:
total_entities: 1
composition:
main_boss: 1
add_behavior:
role: "none"
respawn: false
difficulty:
expected_attempts: 3
primary_failure_causes: []
secondary_failure_causes: []
design_intent: >
Short note on what the fight is trying to test.

View File

@@ -0,0 +1,371 @@
#!/usr/bin/env python3
"""Generate boss YAML from a Markdown design document.
Canonical assumptions:
- Markdown is the source of truth.
- YAML is derived runtime data.
- Missing data should remain empty rather than invented.
This parser is intentionally conservative and template-friendly.
"""
from __future__ import annotations
import argparse
import os
import re
from dataclasses import dataclass
from typing import Dict, List, Optional
import yaml
HEADING_RE = re.compile(r"^(#{1,6})\s+(.*?)\s*$", re.MULTILINE)
KV_LINE_RE = re.compile(r"^([A-Za-z0-9_가-힣 /()\-]+):\s*(.*?)\s*$")
SIGNATURE_SECTION_ALIASES = ["signature_pattern", "시그니처_패턴"]
CORE_CONCEPT_ALIASES = ["core_concept", "concept", "컨셉", "핵심_개념", "핵심_컨셉"]
PLAYER_PROBLEM_ALIASES = ["player_problem", "문제", "플레이어_문제", "플레이어_과제"]
FAILURE_RESULT_ALIASES = ["failure_result", "실패_결과"]
EXPECTED_SOLUTION_ALIASES = ["expected_solution", "해결_방법", "정답", "기대_해법", "예상_해법"]
ROLE_ALIASES = {
"tank": ["tank", "탱커"],
"support": ["support", "지원", "서포트"],
"dps": ["dps", "딜러"],
}
SUSTAINED_PRESSURE_SECTION_ALIASES = ["sustained_pressure_design", "지속_압박_설계"]
BURST_PRESSURE_SECTION_ALIASES = ["burst_pressure_design", "순간_압박_설계", "고점_압박_설계"]
TIME_ESCALATION_SECTION_ALIASES = ["time_escalation", "시간_압박", "시간_경과_강화"]
REVIVE_SECTION_ALIASES = ["revive_interaction_design", "부활_상호작용", "부활_상호작용_설계"]
DIFFICULTY_SECTION_ALIASES = ["difficulty_intent", "난이도_의도"]
EXPECTED_ATTEMPTS_LABELS = ["Expected Attempts", "예상 시도 횟수"]
PRIMARY_FAILURE_CAUSE_LABELS = ["Primary Failure Cause", "주 실패 원인"]
SECONDARY_FAILURE_CAUSE_LABELS = ["Secondary Failure Cause", "보조 실패 원인"]
@dataclass
class Section:
level: int
title: str
content: str
def slugify(text: str) -> str:
text = text.strip().lower()
text = re.sub(r"\s+", "_", text)
text = re.sub(r"[^a-z0-9_가-힣]", "", text)
return text or "boss"
def normalize_title(title: str) -> str:
t = title.strip().lower().replace("", "-")
t = re.sub(r"[^a-z0-9가-힣]+", "_", t)
return t.strip("_")
def maybe_int(text: Optional[str]) -> Optional[int]:
if not text:
return None
m = re.search(r"(\d+)", text)
return int(m.group(1)) if m else None
def parse_sections(markdown: str) -> List[Section]:
matches = list(HEADING_RE.finditer(markdown))
if not matches:
return [Section(level=1, title="document", content=markdown.strip())]
sections: List[Section] = []
for i, match in enumerate(matches):
start = match.end()
end = matches[i + 1].start() if i + 1 < len(matches) else len(markdown)
sections.append(
Section(
level=len(match.group(1)),
title=match.group(2).strip(),
content=markdown[start:end].strip(),
)
)
return sections
def section_map(sections: List[Section]) -> Dict[str, Section]:
mapping: Dict[str, Section] = {}
for section in sections:
key = normalize_title(section.title)
if key not in mapping:
mapping[key] = section
return mapping
def find_line_value(markdown: str, labels: List[str]) -> Optional[str]:
normalized = {normalize_title(label) for label in labels}
for raw in markdown.splitlines():
line = raw.strip().lstrip("- ")
if not line or ":" not in line:
continue
left, right = line.split(":", 1)
if normalize_title(left) in normalized:
value = right.strip()
return value or None
return None
def first_section(smap: Dict[str, Section], aliases: List[str]) -> Optional[Section]:
return next((smap[alias] for alias in aliases if alias in smap), None)
def first_content_value(section: Optional[Section]) -> Optional[str]:
if not section:
return None
for raw in section.content.splitlines():
line = raw.strip().lstrip("-* ").strip()
if line:
return line
return None
def collect_child_sections(sections: List[Section], parent_aliases: List[str]) -> List[Section]:
normalized_aliases = set(parent_aliases)
for index, section in enumerate(sections):
if normalize_title(section.title) not in normalized_aliases:
continue
children: List[Section] = []
parent_level = section.level
for child in sections[index + 1 :]:
if child.level <= parent_level:
break
children.append(child)
return children
return []
def parse_inline_list(section: Optional[Section]) -> List[str]:
if not section:
return []
items: List[str] = []
for raw in section.content.splitlines():
line = raw.strip()
if not line:
continue
if line.startswith(("-", "*")):
items.append(slugify(line.strip("-* "))[:64])
return items
def extract_role_pressure(smap: Dict[str, Section]) -> Dict[str, dict]:
config = {
"tank": ["tank_pressure", "탱커_압박", "tank"],
"support": ["support_pressure", "지원_압박", "support"],
"dps": ["dps_pressure", "딜러_압박", "dps"],
}
output: Dict[str, dict] = {}
for role_name, aliases in config.items():
section = next((smap[a] for a in aliases if a in smap), None)
output[role_name] = {
"required_pressure": None,
"pressure_sources": parse_inline_list(section),
"failure_results": [],
"notes": section.content.strip() if section else None,
}
return output
def extract_signature(markdown: str, sections: List[Section], smap: Dict[str, Section]) -> dict:
signature_section = first_section(smap, SIGNATURE_SECTION_ALIASES)
signature_name = find_line_value(markdown, ["Signature Name", "Signature Pattern", "시그니처 패턴", "시그니처 이름"])
core_concept = None
player_problem = None
failure_description = None
expected_solution: Dict[str, List[str]] = {}
if signature_section:
sub_sections = collect_child_sections(sections, SIGNATURE_SECTION_ALIASES)
sub_map = section_map(sub_sections)
signature_name = signature_name or first_content_value(
first_section(sub_map, ["signature_name", "시그니처_이름"])
)
core_concept = first_section(sub_map, CORE_CONCEPT_ALIASES)
player_problem = first_section(sub_map, PLAYER_PROBLEM_ALIASES)
failure_result = first_section(sub_map, FAILURE_RESULT_ALIASES)
expected = first_section(sub_map, EXPECTED_SOLUTION_ALIASES)
core_concept = core_concept.content.strip() if core_concept else None
player_problem = player_problem.content.strip() if player_problem else None
failure_description = failure_result.content.strip() if failure_result else None
if expected:
current_role: Optional[str] = None
role_items: Dict[str, List[str]] = {"tank": [], "support": [], "dps": []}
for raw in expected.content.splitlines():
line = raw.strip()
if not line:
continue
role_key = normalize_title(line.rstrip(":"))
for canonical_role, aliases in ROLE_ALIASES.items():
if role_key in {normalize_title(alias) for alias in aliases}:
current_role = canonical_role
break
else:
current_role = current_role
if current_role and role_key in {
normalize_title(alias)
for aliases in ROLE_ALIASES.values()
for alias in aliases
}:
continue
if current_role and line.startswith(("-", "*")):
role_items[current_role].append(line.strip("-* "))
elif current_role:
role_items[current_role].append(line)
expected_solution = {k: v for k, v in role_items.items() if v}
return {
"name": signature_name,
"concept": core_concept,
"player_problem": player_problem,
"expected_solution": expected_solution,
"failure_result": {
"type": None,
"description": failure_description,
},
}
def extract_phases(sections: List[Section]) -> List[dict]:
phases: List[dict] = []
for section in sections:
normalized = normalize_title(section.title)
if normalized.startswith("phase") or normalized.startswith("페이즈"):
phase_id = maybe_int(section.title) or len(phases) + 1
mechanics = parse_inline_list(Section(section.level, section.title, section.content))
phases.append(
{
"phase_id": phase_id,
"name": section.title,
"duration_estimate_sec": None,
"new_mechanics": mechanics,
"role_pressure_multiplier": {"tank": None, "support": None, "dps": None},
"notes": section.content.strip() or None,
}
)
return phases
def extract_escalation(markdown: str, smap: Dict[str, Section]) -> dict:
section = first_section(smap, TIME_ESCALATION_SECTION_ALIASES)
interval = maybe_int(find_line_value(markdown, ["Escalation Interval Sec", "Escalation Interval", "강화 간격", "시간 압박 간격"]))
types = parse_inline_list(section)
return {
"enabled": bool(section),
"escalation_interval_sec": interval,
"escalation_types": types,
"notes": section.content.strip() if section else None,
}
def extract_revive(markdown: str, smap: Dict[str, Section]) -> dict:
section = first_section(smap, REVIVE_SECTION_ALIASES)
level = find_line_value(markdown, ["Revive Risk Level", "부활 위험도"])
return {
"trigger_response": {
"enabled": bool(section),
"reactions": parse_inline_list(section),
},
"revive_risk_level": level,
"notes": section.content.strip() if section else None,
}
def generate_boss_yaml(markdown: str, md_path: str, boss_id_override: Optional[str] = None) -> dict:
sections = parse_sections(markdown)
smap = section_map(sections)
boss_name = find_line_value(markdown, ["Boss Name", "보스 이름"]) or os.path.splitext(os.path.basename(md_path))[0]
boss_id = boss_id_override or find_line_value(markdown, ["Boss ID", "boss_id", "보스 ID", "보스 아이디"]) or slugify(boss_name)
stage = maybe_int(find_line_value(markdown, ["Stage", "스테이지"]))
duration_sec = maybe_int(find_line_value(markdown, ["Estimated Duration Sec", "Estimated Duration", "전투 시간"]))
entity_type = find_line_value(markdown, ["Entity Type", "개체 타입"])
sustained_section = first_section(smap, SUSTAINED_PRESSURE_SECTION_ALIASES)
burst_section = first_section(smap, BURST_PRESSURE_SECTION_ALIASES)
difficulty_section = first_section(smap, DIFFICULTY_SECTION_ALIASES)
data = {
"boss": {
"metadata": {
"boss_id": boss_id,
"boss_name": boss_name,
"stage": stage,
"expected_duration_sec": duration_sec,
"entity_type": entity_type,
"source_document": md_path,
},
"signature": extract_signature(markdown, sections, smap),
"roles": extract_role_pressure(smap),
"phases": extract_phases(sections),
"pressure": {
"sustained": {
"sources": parse_inline_list(sustained_section),
"notes": sustained_section.content.strip() if sustained_section else None,
},
"burst": {
"mechanics": parse_inline_list(burst_section),
"notes": burst_section.content.strip() if burst_section else None,
},
"dodge_types_required": [],
},
"revive": extract_revive(markdown, smap),
"escalation": extract_escalation(markdown, smap),
"entities": {
"total_entities": None,
"composition": {},
"add_behavior": {"role": None, "respawn": None},
},
"difficulty": {
"expected_attempts": maybe_int(find_line_value(markdown, EXPECTED_ATTEMPTS_LABELS)),
"primary_failure_causes": [
value
for value in [find_line_value(markdown, PRIMARY_FAILURE_CAUSE_LABELS)]
if value
],
"secondary_failure_causes": [
value
for value in [find_line_value(markdown, SECONDARY_FAILURE_CAUSE_LABELS)]
if value
],
"notes": difficulty_section.content.strip() if difficulty_section else None,
},
"source_notes": {
"unparsed_sections": [
{"title": section.title, "content": section.content[:500]}
for section in sections
if section.level <= 3
]
},
}
}
return data
def main() -> None:
parser = argparse.ArgumentParser(description="Generate boss YAML from Markdown")
parser.add_argument("--md", required=True, help="Path to boss markdown design document")
parser.add_argument("--output", required=True, help="Path to output YAML")
parser.add_argument("--boss-id", default=None, help="Optional boss_id override")
args = parser.parse_args()
with open(args.md, "r", encoding="utf-8") as file:
markdown = file.read()
data = generate_boss_yaml(markdown, args.md, boss_id_override=args.boss_id)
os.makedirs(os.path.dirname(args.output), exist_ok=True)
with open(args.output, "w", encoding="utf-8") as file:
yaml.safe_dump(data, file, allow_unicode=True, sort_keys=False, width=100)
print(args.output)
if __name__ == "__main__":
main()

View File

@@ -0,0 +1,91 @@
#!/usr/bin/env python3
"""Run the full Markdown -> YAML -> Validation pipeline.
Canonical config keys used everywhere:
- design_root
- game_project_root
- schema_path
- runtime_boss_output_dir
- validation_report_suffix
"""
from __future__ import annotations
import argparse
import os
import subprocess
import sys
from pathlib import Path
from typing import Any, Dict
import yaml
HERE = Path(__file__).resolve().parent
REQUIRED_CONFIG_KEYS = [
"design_root",
"game_project_root",
"schema_path",
"runtime_boss_output_dir",
"validation_report_suffix",
]
def load_config(path: str) -> Dict[str, Any]:
with open(path, "r", encoding="utf-8") as file:
data = yaml.safe_load(file) or {}
missing = [key for key in REQUIRED_CONFIG_KEYS if key not in data]
if missing:
raise SystemExit(f"Missing config keys in {path}: {', '.join(missing)}")
return data
def infer_boss_id(md_path: str) -> str:
stem = Path(md_path).stem
stem = stem.replace(" 기획", "").replace("_기획", "")
stem = stem.strip()
return stem.lower().replace(" ", "_")
def main() -> None:
parser = argparse.ArgumentParser(description="Run boss pipeline")
parser.add_argument("--md", required=True, help="Path to boss Markdown document")
parser.add_argument("--config", required=True, help="Path to paths.yaml config")
parser.add_argument("--boss-id", default=None, help="Optional runtime boss_id / yaml filename override")
args = parser.parse_args()
config = load_config(args.config)
md_path = os.path.abspath(args.md)
if not os.path.exists(md_path):
raise SystemExit(f"Markdown file not found: {md_path}")
boss_id = args.boss_id or infer_boss_id(md_path)
yaml_output = os.path.join(config["runtime_boss_output_dir"], f"{boss_id}.yaml")
report_output = os.path.join(
os.path.dirname(md_path),
f"{boss_id}{config['validation_report_suffix']}",
)
schema_path = config["schema_path"]
if not os.path.exists(schema_path):
print(f"Warning: schema_path does not exist yet: {schema_path}")
gen_script = str(HERE / "generate_boss_yaml.py")
val_script = str(HERE / "validate_boss_yaml.py")
subprocess.run(
[sys.executable, gen_script, "--md", md_path, "--output", yaml_output, "--boss-id", boss_id],
check=True,
)
subprocess.run(
[sys.executable, val_script, "--yaml", yaml_output, "--output", report_output],
check=True,
)
print("Pipeline complete")
print(f"Markdown: {md_path}")
print(f"YAML: {yaml_output}")
print(f"Report: {report_output}")
if __name__ == "__main__":
main()

View File

@@ -0,0 +1,184 @@
#!/usr/bin/env python3
"""Validate boss YAML using the project's multiplayer boss rules."""
from __future__ import annotations
import argparse
import os
from typing import Any, Dict, List, Tuple
import yaml
def load_yaml(path: str) -> Dict[str, Any]:
with open(path, "r", encoding="utf-8") as file:
return yaml.safe_load(file) or {}
def score_to_label(issues: List[Tuple[str, str]]) -> str:
severities = [severity for severity, _ in issues]
if "ERROR" in severities:
return "FAIL"
if "WARNING" in severities:
return "WARNING"
return "PASS"
def validate(data: Dict[str, Any]) -> Tuple[Dict[str, int], List[Tuple[str, str]], List[str]]:
boss = data.get("boss", {})
issues: List[Tuple[str, str]] = []
fixes: List[str] = []
scores = {
"Role Integrity Score": 100,
"Pattern Diversity Score": 100,
"Execution Fairness Score": 100,
"Revive Risk Score": 100,
"Time Pressure Score": 100,
}
roles = boss.get("roles", {})
for role_name in ["tank", "support", "dps"]:
role = roles.get(role_name)
if not role:
issues.append(("ERROR", f"Missing role definition: {role_name}"))
scores["Role Integrity Score"] -= 25
fixes.append(f"Add `{role_name}` role pressure definition.")
continue
if not role.get("notes") and not role.get("pressure_sources"):
issues.append(("WARNING", f"No pressure details defined for {role_name}"))
scores["Role Integrity Score"] -= 12
fixes.append(f"Describe how the boss pressures {role_name} players.")
signature = boss.get("signature", {})
if not signature.get("name"):
issues.append(("ERROR", "Missing signature mechanic name"))
scores["Pattern Diversity Score"] -= 30
fixes.append("Add a named signature mechanic.")
if not signature.get("player_problem"):
issues.append(("WARNING", "Signature mechanic has no clear player problem"))
scores["Pattern Diversity Score"] -= 15
fixes.append("Describe the player problem the signature mechanic creates.")
if not signature.get("expected_solution"):
issues.append(("WARNING", "Signature mechanic has no clear expected solution"))
scores["Pattern Diversity Score"] -= 15
fixes.append("Describe the expected solution for the signature mechanic.")
if not (signature.get("failure_result") or {}).get("description"):
issues.append(("WARNING", "Signature mechanic has no failure result"))
scores["Execution Fairness Score"] -= 12
fixes.append("Define what happens when the signature mechanic is failed.")
phases = boss.get("phases") or []
if len(phases) < 3:
issues.append(("ERROR", f"Insufficient phases: {len(phases)} found, expected at least 3"))
scores["Pattern Diversity Score"] -= 30
fixes.append("Add at least 3 phases, ideally 4 for a ~10 minute fight.")
elif len(phases) < 4:
issues.append(("WARNING", "Only 3 phases defined; 4 is recommended for ~10 minute encounters"))
scores["Pattern Diversity Score"] -= 8
for phase in phases:
if not phase.get("new_mechanics") and not phase.get("notes"):
issues.append(("WARNING", f"Phase {phase.get('phase_id')} has no new mechanics listed"))
scores["Pattern Diversity Score"] -= 5
fixes.append(f"List at least one new mechanic for phase {phase.get('phase_id')}.")
pressure = boss.get("pressure", {})
dodge_types = pressure.get("dodge_types_required") or []
if dodge_types and len(dodge_types) < 3:
issues.append(("WARNING", "Fewer than 3 dodge response types defined"))
scores["Execution Fairness Score"] -= 10
fixes.append("List at least 3 dodge response types, such as timing, directional, and chained.")
revive = boss.get("revive", {})
revive_enabled = (revive.get("trigger_response") or {}).get("enabled")
if revive_enabled:
if not (revive.get("trigger_response") or {}).get("reactions"):
issues.append(("WARNING", "Revive response enabled but no reactions defined"))
scores["Revive Risk Score"] -= 20
fixes.append("Define at least one revive-triggered boss reaction.")
else:
issues.append(("WARNING", "No revive-triggered boss response defined"))
scores["Revive Risk Score"] -= 25
fixes.append("Add a risky revive interaction if revive exists in combat.")
escalation = boss.get("escalation", {})
if not escalation.get("enabled"):
issues.append(("ERROR", "No time escalation defined"))
scores["Time Pressure Score"] -= 35
fixes.append("Add soft-enrage or time-based escalation to prevent infinite sustain.")
else:
if not escalation.get("escalation_types"):
issues.append(("WARNING", "Escalation enabled but escalation types are empty"))
scores["Time Pressure Score"] -= 10
fixes.append("Specify how the boss escalates over time.")
if not escalation.get("escalation_interval_sec"):
issues.append(("WARNING", "Escalation interval is not defined"))
scores["Time Pressure Score"] -= 8
fixes.append("Specify an escalation interval in seconds.")
for score_name in list(scores.keys()):
scores[score_name] = max(0, scores[score_name])
return scores, issues, fixes
def render_report(yaml_path: str, data: Dict[str, Any], scores: Dict[str, int], issues: List[Tuple[str, str]], fixes: List[str]) -> str:
boss = data.get("boss", {})
metadata = boss.get("metadata", {})
result = score_to_label(issues)
lines = [
"# Boss Validation Report",
"",
f"Boss ID: {metadata.get('boss_id')}",
f"Boss Name: {metadata.get('boss_name')}",
f"Source YAML: `{yaml_path}`",
"",
f"Result: **{result}**",
"",
]
for score_name, value in scores.items():
lines.append(f"- {score_name}: {value}")
lines.extend(["", "## Detected Issues"])
if issues:
for severity, message in issues:
lines.append(f"- **{severity}**: {message}")
else:
lines.append("- None")
lines.extend(["", "## Suggested Fixes"])
if fixes:
seen = set()
for item in fixes:
if item not in seen:
lines.append(f"- {item}")
seen.add(item)
else:
lines.append("- No changes required.")
lines.extend([
"",
"## Notes",
"- Markdown source remains the source of truth.",
"- Regenerate YAML after editing the Markdown design document.",
"",
])
return "\n".join(lines)
def main() -> None:
parser = argparse.ArgumentParser(description="Validate boss YAML")
parser.add_argument("--yaml", required=True, help="Path to boss YAML")
parser.add_argument("--output", required=True, help="Path to output Markdown report")
args = parser.parse_args()
data = load_yaml(args.yaml)
scores, issues, fixes = validate(data)
report = render_report(args.yaml, data, scores, issues, fixes)
os.makedirs(os.path.dirname(args.output), exist_ok=True)
with open(args.output, "w", encoding="utf-8") as file:
file.write(report)
print(args.output)
if __name__ == "__main__":
main()

79
Docs/DESIGN_PHILOSOPHY.md Normal file
View File

@@ -0,0 +1,79 @@
# Multiplayer Boss Design Philosophy
This project focuses on 4-player cooperative boss encounters.
Bosses are not just enemies.
Bosses are cooperative execution challenges.
## Core Design Philosophy
Boss encounters must test:
- role execution
- positioning
- pattern recognition
- sustained coordination
Boss encounters should not primarily test:
- raw stat checks
- random power spikes
- build luck
## Role Model
Roles are soft ratios.
Players are not locked into hard classes.
Each player contributes a mixture of:
- Tank ratio
- DPS ratio
- Support ratio
Bosses should require total role pressure fulfillment, not specific class presence.
## Combat Identity
Combat is:
- non-target action
- directional dodge-based
- positioning-focused
Survival should depend more on execution than on numerical advantage.
## Signature Pattern Philosophy
Each boss must contain at least one defining signature mechanic.
That mechanic must:
- create a clear player problem
- have a clear expected solution
- produce a clear failure result
Ambiguous solutions are discouraged.
## Revive Philosophy
Revive is allowed, but revive must be risky.
Bosses should react to revive attempts.
Revive should increase encounter tension, not reduce it.
## Sustain Philosophy
There is no hard time limit by default.
Instead, bosses should escalate over time.
Encounters must become harder if players fail to execute properly.
## Build Philosophy
Build strength should support execution.
It should not replace execution.
Players should adapt builds, not brute-force encounters.
## Encounter Identity
Boss equals encounter.
Bosses may include:
- multiple entities
- supporting units
- positional hazards
Adds must serve meaningful gameplay roles, never filler.