Wiki 소급 감사 리포트 — 2026-08-29
대상: wiki/ 전체 270페이지 (verified 225, field-tested 45)
축: 범용성(transferability) · 의미적 중복(duplication) · 팩트(fact)
방법: 병렬 에이전트 8개 — 범용성 전수 1, 중복 전수 1, 팩트 체크 6(도메인 분할, 인용 URL 실제 fetch 후 원문 대조)
결과: blocker 33건(24페이지) · advisory 58건
총평
- 범용성: 사실상 클린. 270페이지 전수(트리거+지시 테이블 정독)에서 blocker 0. 사설 repo 언급 ~80건은 전부 Sources/필드 증거 안. advisory 1건뿐.
- 중복: 클린. 같은 지시를 주제로 한 중복 페이지 0. 상호 링크 누락 2쌍만 발견.
- 팩트: 실질적 문제 집중. 인용문 조작/불일치, 소스가 주장을 뒷받침하지 않는 귀속 오류, 죽은 링크,
verified 과대 표기, frontmatter 오형식이 24페이지에서 발견.
Blockers (33건, 24페이지)
A. 인용이 소스에 없음 / 소스가 주장과 무관 (15건)
| 페이지 |
문제 |
| testing/data/test-data-and-isolation.md |
abseil SWE-book "verbatim 인용"이 원문에 없음 (유사 문장만 존재) |
| testing/quality/differential-run-agreement.md |
arXiv 2410.21904는 mutation testing 논문 — differential testing 아님 |
| testing/quality/differential-run-agreement.md |
Stryker mutant-states 페이지에 RIP 모델 내용 없음 |
| testing/quality/history-dependent-checks-on-shallow-clones.md |
git-rev-parse 인용문이 실제 문구와 불일치 |
| testing/strategy/differential-testing.md |
arXiv 2102.07498의 실제 논문 제목이 인용 제목과 다름 (인용 제목의 논문 없음) |
| testing/async/async-testing.md |
readline question/'line' 상호작용 주장이 Node 문서에 없음 (필드 로그만 근거) |
| platforms/toolchains/compiler-sysroot-on-macos.md |
clang UsersManual에 SDKROOT 언급 자체가 없음 |
| infrastructure/agent-orchestration/usage-limit-paused-workers.md |
costs 문서 "인용문"이 실제 페이지에 없음 |
| backend/common/reliability/client-side-rate-limiting.md |
Auth0 문서에 /oauth/token 공표 한도 없음 (Okta 절반은 확인됨) |
| backend/java/kotlin/frameworks-and-jpa.md |
Kotlin 2.1/2.2 어노테이션 타깃 버전 분기가 인용 문서에 없음 |
| frontend/auth/token-handling-client-side.md |
OWASP JWT 치트시트는 서버측 검증만 다룸 — 인용된 주제(sidejacking/쿠키 저장) 없음 |
| frontend/performance/bundle-and-assets.md |
web.dev CLS 문서에 "reserve space" 지침 없음 |
| security/authn/session-vs-token.md |
RFC 8725에 "짧은 수명" 권고 없음 (알고리즘/issuer 고정은 확인) |
| security/input/validation-at-trust-boundaries.md |
OWASP Path_Traversal 페이지에 canonicalize-then-prefix-check 없음 |
| databases/schema-design/foreign-keys-and-referential-actions.md |
ddl-constraints.html이 오히려 해당 주제를 명시적으로 다른 곳으로 넘김 (deferrable FK 설명 없음) |
B. 수치/식별자 조작 의심 (5건)
| 페이지 |
문제 |
| frontend/agent-interfaces/agent-facing-tool-surfaces.md |
Chrome 150 deprecation — 4개 인용 소스 어디에도 없음 |
| frontend/agent-interfaces/agent-facing-tool-surfaces.md |
문자 예산(30/500/150/1.5K) — 인용 소스에 수치 없음 |
| frontend/agent-interfaces/agent-facing-tool-surfaces.md |
origin trial Chrome 156 종료 — 소스에 없음, 조작 의심 |
| frontend/design/html-in-canvas.md |
WICG 스펙 API명 오기: texElementImage2D→실제 texElementSubImage2D, copyElementImageToTexture→drawElementImageToTexture |
| databases/data-survey/surveying-live-data-for-a-rule.md |
인용된 "data-migration-expert" 에이전트/파일이 해당 repo에 없음 (실제: data-migration-reviewer, 다른 경로) |
C. 죽은 링크 (2건)
| 페이지 |
문제 |
| security/dependencies/agent-skill-supply-chain.md |
openwork skills-lock.json — 기본 브랜치가 dev라 main 링크 404, 현 브랜치에도 파일 없음 |
| security/dependencies/agent-skill-supply-chain.md |
book-to-skill SECURITY-NOTICE.md — 기본 브랜치 master라 main 링크 404 (내용 자체는 gh api로 실재 확인) |
D. verified 과대 표기 — 외부 검증 가능 소스 없음 (3건)
| 페이지 |
조치 |
| infrastructure/agent-orchestration/dispatching-after-a-completion-report.md |
소스가 전부 내부 Orca CLI → field-tested로 강등 (범용성 advisory도 이 페이지: 실행 가능한 행이 전부 orca 문법) |
| qa/exploratory/lowered-declaration-survival.md |
유일 소스가 내부 재현 로그 → field-tested로 강등 |
| security/data/masking-verification.md |
핵심 주장(채널별 검증)이 OWASP 로깅 치트시트 범위 밖, 실근거는 내부 재현 → field-tested로 강등 |
E. 낡은 내용 — 현행 문서와 모순 (2건)
| 페이지 |
문제 |
| infrastructure/agent-orchestration/usage-limit-paused-workers.md |
"자동 재개 미구현" 주장 ↔ 현행 costs 문서: v2.1.234+ autoContinueAtUsageLimit로 구현됨 |
| infrastructure/agent-orchestration/usage-limit-paused-workers.md |
"open feature request"로 인용한 issue #36320은 CLOSED(duplicate) |
F. frontmatter 오형식 — 계통적 결함 (4건)
| 페이지 |
문제 |
| platforms/processes/tool-diagnostics-without-a-failing-exit-code.md |
last_verified·related 키 중복 (YAML last-key-wins로 앞 값 유실) |
| infrastructure/agent-orchestration/session-completion-gates.md |
동일 + 소스 불릿 1개가 sources 리스트 밖에 고아로 존재 |
| infrastructure/agent-orchestration/worktree-isolated-workers.md |
동일 패턴 |
| qa/deliverables/generated-artifacts-as-deliverable-source.md |
related:에 리스트 리터럴 3개 중첩 — 파서가 앞 2개 리스트 유실 |
※ structure-checks.js가 중복 키를 안 잡는 사각지대 — 체커 보강 필요.
G. 페이지 내부 모순 (2건)
| 페이지 |
문제 |
| backend/common/change-impact/widening-a-closed-value-table.md |
grep "7군데" 주장 vs 페이지 자체 열거 합계 6 |
| qa/document-verification/spec-document-gates.md |
"8개 변조 사본" vs "32개 전부 실패" — 미조정 (advisory로 분류되었으나 수정 대상) |
Advisories (58건) — 대분류
- CI에서 fetch 불가 URL (~25건): gnu.org 429, Apple 문서 JS 렌더링, testing.googleblog JS 렌더링, openjdk/baeldung/uvicorn/psycopg 403, tanstack 500, dl.acm 403 등. 내용 오류 아님 — "unverifiable from CI" 기록. 게이트 프롬프트의 advisory 처리 원칙과 일치.
- non-verbatim 인용 (~8건): 의미는 맞지만 따옴표 처리된 문구가 원문과 다름 (예: databases/transactions/application-clock의 dml-returning 인용, testing/mocking/captured-call-arguments의 mock 문서 인용).
- frontmatter sources ↔ 본문 Sources 불일치 (~6건): tests-that-cannot-fail(3개 URL 누락), permissions-and-exec-bits, pane-delivery-confirmation, missing-container-metrics, resource-limits-and-probes, spec-artifact-checks(귀속 오류).
- 소스 과잉 귀속/과잉 일반화 (~8건): 인용 페이지가 아닌 하위/다른 페이지에 있는 내용을 귀속 (column-data-types의 MySQL TOC, path-resolution의 brew FAQ, background-services의 crontab LOGNAME, comparing-two-execution-plans의 v1.7 컬럼명 등).
- 중복 감사 cross-link 누락 2건: reproducible-reports ↔ reproduce-first, keys-ahead-of-their-consumer ↔ unenforced-declarations.
- 범용성 1건: dispatching-after-a-completion-report.md 실행 행 전부 orca 문법.
- spec-document-gates.md 중복 키 (advisory 분류지만 F와 동일 계통).
커버리지
- 페이지: 270/270 (스킵 0) — 범용성 전수, 팩트 전수
- URL fetch: 약 440회 (도메인별 ~85–95 + security 40)
- 판정 원칙: fetch 실패는 재시도 1회 후 advisory("unverifiable"), 추측 판정 금지 — 전 에이전트 준수
- 한계: pdf/JS-SPA/봇차단 소스는 검증 불가로 남음 (advisory 1분류). 필드 측정 로그 자체의 재현은 범위 밖.
권고 수정 순서
- F(frontmatter 4건) + 체커 보강: 기계 수정 + structure-checks.js에 duplicate-key 검사 추가 (재발 방지)
- C(죽은 링크 2) + E(낡은 내용 2): 링크를 실제 경로/커밋 고정 링크로 교체, usage-limit 페이지 현행화
- A·B(조작/무관 인용 20): 페이지별로 (i) 올바른 소스로 교체, (ii) 주장을 소스가 실제 말하는 수준으로 축소, (iii) 소스 없으면 해당 행 삭제 또는 field-tested 강등
- D(강등 3건): confidence를 field-tested로 변경
- G(내부 모순 2): 수치 재측정 또는 서술 정정
- advisory 2·3·4분류: non-verbatim 인용 정정, sources 리스트 동기화 — wiki-lint 정기 패스에서 처리 가능
Wiki 소급 감사 리포트 — 2026-08-29
대상: wiki/ 전체 270페이지 (verified 225, field-tested 45)
축: 범용성(transferability) · 의미적 중복(duplication) · 팩트(fact)
방법: 병렬 에이전트 8개 — 범용성 전수 1, 중복 전수 1, 팩트 체크 6(도메인 분할, 인용 URL 실제 fetch 후 원문 대조)
결과: blocker 33건(24페이지) · advisory 58건
총평
verified과대 표기, frontmatter 오형식이 24페이지에서 발견.Blockers (33건, 24페이지)
A. 인용이 소스에 없음 / 소스가 주장과 무관 (15건)
B. 수치/식별자 조작 의심 (5건)
C. 죽은 링크 (2건)
D.
verified과대 표기 — 외부 검증 가능 소스 없음 (3건)E. 낡은 내용 — 현행 문서와 모순 (2건)
F. frontmatter 오형식 — 계통적 결함 (4건)
※ structure-checks.js가 중복 키를 안 잡는 사각지대 — 체커 보강 필요.
G. 페이지 내부 모순 (2건)
Advisories (58건) — 대분류
커버리지
권고 수정 순서