bookwriting-evidence-link-master
Skill4 starsChanged 3 months agoMITView on GitHub
저자 책 ⑤ 단계 마스터 — EXPANDED 시드의 각 fact·claim에 수집된 근거를 결정론 연결. M3 AuthorityGrader + M13 R4MultiLayerVerifier + M2 HallucinationDetector 통합. CITE-NNN ID 발급 + forecast-registry P-NNN R1-R5 검증. LINKED→EXPANDED→EVIDENCED 게이트.
In this repository15 other files it gives its agents.
- bookwriting-absorb-masterSkill
- bookwriting-content-expand-masterSkill
- bookwriting-cross-chapter-relink-masterSkill
- bookwriting-evidence-collect-masterSkill
- bookwriting-fact-anchor-masterSkill
- bookwriting-insight-synthesis-masterSkill
- bookwriting-line-edit-masterSkill
- bookwriting-masterSkill
- bookwriting-narrative-craft-masterSkill
- bookwriting-notebooklm-bridgeSkill
- bookwriting-sentence-edit-masterSkill
- bookwriting-structure-review-masterSkill
- bookwriting-systemlink-masterSkill
- citation-gateSkill
- voice-gateSkill
The file
Files
bookwriting-evidence-link-master (⑤ 단계 마스터)
TLDR
EXPANDED 시드의 모든 claim에 ④에서 수집한 근거를 1:1 매핑·검증. R4 통과 + R1-R5 정합 + HallucinationDetector 0건 → EVIDENCED 진입.
Triggers
/bookwriting cycle <target>Step 5 자동/bookwriting-evidence-link <seed_id>— 저자 직접/bookwriting-evidence-link verify-r1-r5 <prediction_id>— R1-R5만/bookwriting-evidence-link cite-issue <claim_id>— CITE-NNN 발급만
Detailed Methodology
1. 결정론 chain (LLM 추론 영구 금지)
import paths
from r4_multi_layer_verifier import R4MultiLayerVerifier
from forecast_registry_syncer import ForecastRegistrySyncer
from citation_detection import HallucinationDetector
from authority_grader import AuthorityGrader
from id_generator import IdGenerator
from lifecycle_state_machine import LifecycleStateMachine, LifecycleState
from pathlib import Path
# Step 1: ④ 수집 결과 로드 (claim_id → [citations])
collected = load_evidence_collection_output(seed_id)
# Step 2: HallucinationDetector — 권위 5 signals 검증
detector = HallucinationDetector()
for claim_id, citations in collected.items():
signals = detector.detect_signals(text=claim_text, citations=citations)
if signals["any_critical"]:
return {"action": "fail_hallucination", "claim_id": claim_id, "signals": signals}
# Step 3: R4 다층 인용 강제 (≥2층)
r4 = R4MultiLayerVerifier()
results = r4.verify_all_claims(collected)
pass_rate = r4.count_pass_rate(results)
if pass_rate < 1.0: # 100% 통과 강제
failed = [r for r in results if not r["r4_pass"]]
return {"action": "fail_r4", "failed_claims": failed}
# Step 4: forecast-registry P-NNN R1-R5 검증
syncer = ForecastRegistrySyncer(paths.forecast_registry_path())
violations = syncer.verify_all()
if any(violations.values()):
return {"action": "fail_r1_r5", "violations": violations}
# Step 5: CITE-NNN ID 발급 (M7 atomic fcntl)
gen = IdGenerator(citation_registry_path)
for claim_id, citations in collected.items():
for c in citations:
c["cite_id"] = gen.issue("citation", current_max_num=...)
# Step 6: lifecycle EXPANDED → EVIDENCED 결정론 전환 (M11)
fsm = LifecycleStateMachine()
new_state = fsm.transition(LifecycleState.EXPANDED, LifecycleState.EVIDENCED)2. R1-R5 R4 ⊇ 게이트 매트릭스
| Rule | 검증 모듈 | 통과 조건 |
| R1 quantitative_anchor | M12 | 모든 P-NNN에 정량 앵커 ≥1 |
| R2 chapter_role | M12 | 1층_콜드오픈·2층_분석·3층_함의 enum |
| R3 system_loops | M12 | L1-L7 enum |
| R4 multi_layer | M13 | unique grades ≥2 |
| R5 verification_signal | M12 | 5년·10년 falsifies_if 명시 |
3. R4 fail 시 저자 escalate
if not r4_result["r4_pass"]:
return {
"action": "escalate",
"message": (
f"R4 위반 — claim_id={claim_id} 단일 권위만 인용. "
f"현재 layers={r4_result['layers']}. "
"추가 권위 1층 이상 수집 후 재시도 필요."
),
}4. citation-authority-db 누락 권위 — 자동 등록 영구 금지
저자 명시 확인 없이 citation-authority-db.yaml에 신규 권위 등록 영구 금지. 누락 권위 발견 시 저자 escalate, 저자 명시 후에만 DB 갱신.
5. lifecycle 전환
# EXPANDED → EVIDENCED (Round 18 STAGE_TO_TRANSITION 매핑)
fsm = LifecycleStateMachine()
new_state = fsm.transition(LifecycleState.EXPANDED, LifecycleState.EVIDENCED)저자 진북 정합
| 진북 | 검증 |
| #1 할루시네이션 0 | ✅ HallucinationDetector + R4 + R1-R5 결정론 통합 |
| #3 출처 명시 | ✅ R4 다층 강제 + CITE-NNN 1:1 매핑 |
| #5 보안 법적 위험 | ✅ 신규 권위 자동 등록 영구 금지 |
| #7 결정론 환원 | ✅ M3·M13·M2·M12·M7·M11 pure Python |
출처
SPEC: specs/bookwriting-evidence-link-master.md
실행 모듈: lib/r4_multi_layer_verifier.py + lib/forecast_registry_syncer.py + lib/citation_detection.py + lib/authority_grader.py + lib/id_generator.py + lib/lifecycle_state_machine.py
Sub-skill Orchestration Trace ⭐ (저자 7번째 절대 protocol)
본 마스터 호출 시 7개 sub-skill을 자동 orchestration.
| 순서 | Sub-skill | 호출 시점 | 페르소나 | 결정론 모듈 |
| 1 | bookwriting-evidence-link-graph-builder | Step 2 | Graph Builder | yaml.safe_load + dataclass |
| 2 | bookwriting-evidence-link-edge-detector | Step 3 | Edge Detector | 3 type 정규식 (logical·temporal·causal) |
| 3 | bookwriting-evidence-link-authority-mapper | Step 4 | Authority Mapper | M3 AuthorityGrader |
| 4 | bookwriting-evidence-link-r4-verifier ⭐ | Step 5 | R4 Verifier | M13 R4MultiLayerVerifier + 저자 묶음 4 옵션 |
| 5 | bookwriting-evidence-link-loop-integrator | Step 6 | Loop Integrator | yaml + L1-L7 dict |
| 6 | bookwriting-evidence-link-isolation-detector | Step 6 | Isolation Detector | degree dict + 저자 묶음 3 옵션 |
| 7 | bookwriting-evidence-link-graph-visualizer | Step 9 | Graph Visualizer | str 조립 markdown table + ASCII |
자동 호출 흐름
사용자 명령: /bookwriting-evidence-link cycle <target>
↓
Step 2: bookwriting-evidence-link-graph-builder (evidence-track → 노드)
↓
Step 3: bookwriting-evidence-link-edge-detector (논리·시간·인과 엣지)
↓
Step 4: bookwriting-evidence-link-authority-mapper (Grade A-E 100% 매핑)
↓
Step 5: bookwriting-evidence-link-r4-verifier ⭐ (R4 다층 강제 + 저자 묶음 4 옵션)
↓
Step 6 병렬:
├─ bookwriting-evidence-link-loop-integrator (L1-L7 매핑 80%+)
└─ bookwriting-evidence-link-isolation-detector (degree 0·1 + 저자 묶음)
↓
Step 9: bookwriting-evidence-link-graph-visualizer (markdown + ASCII)
↓
Master Synthesis: chapters/<ch>/evidence-track.yaml 그래프 필드 갱신저자 결정 묶음 강제
- R4 Verifier 단일 권위 주장 발견 시 저자 묶음 4 옵션 (외부 manual·Layer 1 유지·풀 제외·defer)
- Isolation Detector degree 0 발견 시 저자 묶음 3 옵션 (풀 제외·연결 추가·미해결 보관)
OrchestratorTracer (M18) 자동 기록
deterministic_modules_used: M3·M13 + pure Python dict 그래프 (networkx 의존 없음).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

