agentleFS
idoforgodidoforgod/cys-bookwriting-skills/skills/citation-gate/SKILL.md

citation-gate

Skill4 starsChanged 3 months agoMITView on GitHub

저자 책 학술 신뢰성 cross-cutting 가드. M2 HallucinationDetector + M3 AuthorityGrader + M4 QuantitativeAnchorExtractor 결정론 호출. 5 신호 자동 감지 — Critical 시 즉시 차단·재작성 안 함. 저자 진북 #1 (할루시네이션 0) + #3 (출처 명시) + #5 (법적 위험) backbone.

The file

Files
SKILL.md

citation-gate (cross-cutting 메타 가드)

TLDR

저자 책의 모든 인용·주장·근거가 할루시네이션 없는 실재 출처에 의해 뒷받침되도록 시스템적 강제. lib/citation_detection.HallucinationDetector + lib/authority_grader.AuthorityGrader 결정론 호출.

Triggers

  • 모든 단계 마스터 출력 직후 자동 발동 (cross-cutting)
  • /citation-gate verify <text> — 5 신호 즉시 검사
  • /citation-gate audit <chapter> — 챕터 전체 인용 일괄 감사
  • /citation-gate authority-map <chapter> — 권위 7층 분포 보고

Detailed Methodology

1. 결정론 호출 (LLM 추론 영구 금지)

import paths
from citation_detection import HallucinationDetector, CitationMetadata
from authority_grader import AuthorityGrader, Grade
from pathlib import Path

det = HallucinationDetector()
grader = AuthorityGrader(paths.authority_db_path())

citation = CitationMetadata(
    author="Dario Amodei",
    date="2024-10-11",
    publication="Anthropic",
    source_file=paths.inbox_dir() / "2024-amodei.md",
)
signals = det.detect_all(text, citation)
grade = grader.assign_grade(author=citation.author)

if signals.is_critical():
    # 즉시 차단 — 재작성 영구 금지
    escalate_to_user(signals, citation)

2. 5 할루시네이션 신호 (Round 4·9 누적)

  1. source_file_missing — Path.exists() 결정론
  2. source_format_vague — "한 보고서에 따르면" 등 정규식
  3. metadata_incomplete — author·date·publication 누락
  4. url_doi_invalid — URL·DOI 정규식
  5. stat_no_source — 통계 + source_file 부재

3. 권위 7층 (A·A_preprint·B·C·D·E·F·G)

GradeLayer예시
A1창업자 증언·peer-reviewed (Amodei·Hassabis·WHO·IPCC)
A_preprint1arxiv·biorxiv (peer-review 전)
B2저자 본인 (미래학자)
C3인접 학문 (김대식·유현준·Zuboff·Crawford)
D4사상가·역사가 (Harari·Turkle·Haraway)
E5역사 평행 (대공황·러다이트)
F6신학 권위 (Calvin·Bavinck·Augustine)
G7토착 지혜 (이부영·서대석)

4. R4 다층 인용 강제

result = grader.verify_r4_multi_layer([Grade.A, Grade.D])
# {"layer_count": 2, "r4_pass": True, "action": "approve"}

5. 위반 시 처리 (Round 4 #Q33)

  • Critical (할루시네이션 의심) → 즉시 차단·재작성 안 함·저자 묶음
  • High (메타 누락·R4 단일 권위) → 저자 묶음 escalate
  • Medium (Grade C·D만) → warning + 진행

6. trauma_bearing_witness_category (Round 17·18)

생존자·유가족·정치 박해 피해자 인용 시 자동 발동:

  • consent_status: required_before_quote
  • safe_messaging_invoke: true
  • hotline_required: 한국 1393·109 명시 강제

저자 진북 정합

진북검증
#1 할루시네이션 0✅ M2 5 신호 결정론 + pytest 통과
#3 출처 명시✅ citation-authority-db 57 권위 + 7 grade
#5 법적 위험✅ trauma_bearing_witness_category protocol

출처

SPEC: specs/citation-gate.md (17 항목 풀) 실행 모듈: lib/citation_detection.py + lib/authority_grader.py 권위 DB: expert_pool/citation-authority-db.yaml (57 권위 entry)

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.