leerness 1.36.59 → 1.36.61

package/CHANGELOG.md

@@ -1,5 +1,24 @@
 # Changelog
 
+## 1.36.61 — 2026-07-21 — .harness 정책·시드 문서군 en 완역 (F-05 3회차) — R-0001 검수 6회전
+
+en init 한글 잔존 53/61 → **23/61** (1·3회차 누적). 검수 계약 대조: 6개 대표 문서 규칙 전량 보존 확인.
+
+- **en 오버라이드 27종**: guideline/guardrails/anti-lazy/writeback/session-close/secret/encoding/consistency/testing/checklists/plan·brief·current-state·decisions·task-log·handoff 시드 등 — read-order 참조층 전체.
+- **자가 선검**: en brief placeholder 문구가 감지 regex 와 미스매치(작성됨 오판) → 문구 정합 (anchors status true/true 실측).
+- (검수 High) **정책문서 12종 managed 편입** — 언어 전환 시 구 언어로 방치되던 것 (사용자 상태 문서는 의도 제외). 실측: en→ko 전환 시 guardrails 한국어 재생성 + 커스텀 1회 + 영어 잔재 0.
+- (검수 Medium) **en 센티널 인식**: audit/lazy detect 의 `(자동)`, design-token `(실제 값으로 업데이트)` 만 알던 감시자들에 `(auto)`/`(update with real value)` 병기 — KO/EN audit 발견 동수(5=5) 회복, --fix 양 언어.
+- 검증: e2e en 케이스 10종 무한글로 확대, selftest 332, 검수 통과 프로브(파서 소비자 전수·integrity en 12종·전환 왕복).
+
+## 1.36.60 — 2026-07-21 — 언어 전환 현지화 병합 (F-05 2회차, 전 검수 High 해소) — R-0001 검수 5회전
+
+KO→EN(역방향 포함) 전환 시 지시 레이어가 새 언어로 재생성되고 사용자 커스텀만 정확히 1회 이월된다.
+
+- **차감 병합**: _managedMerge 에 altTemplate — 구 언어 canonical 라인은 커스텀이 아니므로 미이월 (종전: KO 템플릿 통째 Preserved 이월). session-workflow.md managed 편입(전환 시에도 낡은 언어로 방치되던 것). 보존 래퍼 문구 현지화(en 프로젝트에 한글 래퍼가 새 한글원이 되던 것).
+- **검수 2차 High 반영 — 전환-감지 게이트 + 출현-횟수 차감**: 무조건 차감은 동일-언어 재init 에서 "반대-언어 템플릿과 우연히 일치하는 정당한 커스텀"을 유실시켰다(실측) — 이전 저장 언어를 포착해 실제 전환일 때만 차감 + Set 대신 출현 1회분만 소비(복제 커스텀 생존).
+- (검수 #2) adapter 경로에 래퍼 언어 전달(전환-감지 없는 경로라 altTemplate 은 의도적 미적용). (검수 #3) 구버전 workflow 의 자동생성 이력 로그(MCP 도구-카운트/마일스톤 불릿)는 좁은 시그니처로 필터 — 낡은 모순 잔재 방지, 사용자 산문 보존.
+- 실측: 동일-언어 커스텀 생존 / 전환 차감 유지(잔존 한글=커스텀 1건) / 이력 필터+커스텀 보존. selftest 332(+1), 검수 통과 프로브(migrate/update 동일 경로·EN 래퍼 혼재 무중복·5종 바이트 멱등).
+
 ## 1.36.59 — 2026-07-21 — en 지시-레이어 완전 번역 (외부감사 F-05 시리즈 1회차) — R-0001 검수 4회전
 
 실측 베이스라인: en init 산출물 61개 중 한글 잔존 53개. 1회차 = en 에이전트가 "지시"로 읽는 최고 영향 5종.

package/README.md

@@ -122,7 +122,7 @@ MIT
 <!-- leerness:project-readme:start -->
 ## Leerness Project Harness
 
-이 프로젝트는 Leerness v1.36.59 하네스를 사용합니다. AI 에이전트는 작업 전 `leerness handoff`로 컨텍스트를 적재하고, 작업 후 `leerness check`/`leerness audit`/`leerness session close`를 수행해야 합니다.
+이 프로젝트는 Leerness v1.36.61 하네스를 사용합니다. AI 에이전트는 작업 전 `leerness handoff`로 컨텍스트를 적재하고, 작업 후 `leerness check`/`leerness audit`/`leerness session close`를 수행해야 합니다.
 
 ### 정체성 — AI 에이전트 운영 레이어 (UR-0030)
 
@@ -176,7 +176,7 @@ leerness memory restore decision <date|title>
 
 ### MCP server (외부 AI 통합)
 
-Leerness v1.36.59는 stdio JSON-RPC MCP server를 내장합니다 — Claude Code · Cursor · Codex CLI 등 외부 AI에 **86개 도구**를 노출:
+Leerness v1.36.61는 stdio JSON-RPC MCP server를 내장합니다 — Claude Code · Cursor · Codex CLI 등 외부 AI에 **86개 도구**를 노출:
 
 ```jsonc
 // 카테고리별
@@ -197,7 +197,7 @@ Leerness v1.36.59는 stdio JSON-RPC MCP server를 내장합니다 — Claude Cod
 `<<autonomous-loop-dynamic>>` 신호만 보내면 AI가:
 1) 다음 라운드 후보 선정 → 2) 코드 변경 → 3) 회귀 테스트 갱신 → 4) 전체 e2e 스위트 통과 → 5) npm publish + git tag → 6) main push → 7) session close → 8) 다음 라운드 예약.
 
-현재 누적: **v1.9.x → 1.36.59 릴리스 태그 이력** (수백 라운드) · _reports/는 비공개 보존.
+현재 누적: **v1.9.x → 1.36.61 릴리스 태그 이력** (수백 라운드) · _reports/는 비공개 보존.
 
 ### 성능 가이드
 
@@ -235,6 +235,6 @@ leerness release pack --close --auto-main-push
 - `.harness/session-handoff.md`: 다음 세션 인수인계 (자동 작성)
 - `.harness/lessons.md` / `decisions.md` / `rules.md`: 영구 메모리 (5 surface)
 
-Last synced by Leerness v1.36.59: 2026-07-22
+Last synced by Leerness v1.36.61: 2026-07-22
 <!-- leerness:project-readme:end -->
 

package/bin/leerness.js

@@ -34,7 +34,7 @@ const { CAPABILITY_SURFACE, POWERFUL_COMMANDS, ADAPTERS, REUSE_CATEGORIES, REUSE
 const { tokenizeForRank: _tokenizeForRank, expandQuery: _expandQuery, scoreHits: _scoreHits, suggestTerms: _suggestTerms } = require('../lib/search-core');  // 1.36.23: memory search 랭킹 코어(순수·0-deps)
 const { findCorruptedStateJson: _findCorruptedStateJson } = require('../lib/state-integrity');  // 1.36.1 (클린룸 리뷰 FN): .harness/*.json 상태 무결성 (audit/health/check 공유)
 
-const VERSION = '1.36.59';
+const VERSION = '1.36.61';
 
 // 1.9.290 (UR-0037, Codex gpt-5.5 #4 수렴): CLI 전용 부작용은 require 시 실행하지 않는다.
 //   이전: warning listener 제거 / NODE_OPTIONS 변경 / chcp IIFE 가 top-level 즉시 실행 → require('harness') 시 호스트 프로세스 오염.
@@ -667,6 +667,34 @@ leerness memory restore <surface> <target>   # archive → active 복귀 (DELETE
   if (lang === 'en') {
     _files['AGENTS.md'] = `${MARK}\n# Leerness Agent Instructions\n\n## ⭐ First action every session\n**Read \`.harness/session-workflow.md\` first and follow its 6-step workflow**: analyze request → plan → distribute → sub-agent work → integrated verification → close. Keep the same flow regardless of round length/complexity — that is what prevents drift.\n\n## Static vs dynamic — the leerness boundary\n**AGENTS.md = static project instructions** (coding rules, test commands, prohibitions, deploy steps — rarely change).\n**leerness = dynamic work state, memory, verification, handoff** (current goal, changed files, failed attempts, verification results, next-agent handoff — change every task).\n- Put rules/commands/prohibitions here in AGENTS.md.\n- Dynamic state (decisions/lessons/plan/progress/verification/handoff) is recorded by leerness in the **default workspace \`.harness/\`** (decisions.md / lessons.md / plan.md / progress-tracker.md / session-handoff.md) via \`leerness handoff\`, \`decision add\`, \`lesson save\`, etc.\n- (Optional) the JSON state substrate of \`leerness state show|start|record|verify|handoff\` (or MCP \`leerness_state_*\`) lives in \`.leerness/\` — the cross-agent handoff standard, separate from the main workspace.\n- leerness **complements** AGENTS.md, it does not replace it. Static instructions here, dynamic state in leerness.\n\n## Mandatory read order (session start)\n1. **.harness/session-workflow.md** (6-step workflow — highest priority)\n2. .harness/context-routing.md\n3. .harness/session-handoff.md\n4. .harness/current-state.md\n5. .harness/plan.md\n6. .harness/progress-tracker.md\n7. .harness/guideline.md\n8. .harness/protected-files.md\n9. .harness/writeback-policy.md\n10. .harness/anti-lazy-work-policy.md\n11. **.harness/rules.md** (user-defined standing rules — follow every session)\n\n## Required behavior\n- Run \`leerness handoff .\` at session start to load context (handoff prints active rules automatically).\n- **Ask-on-ambiguity duty**: if the user's request has parts open to interpretation (vague qualifiers, pronouns, multiple alternatives, unclear scope), **ask the user instead of guessing**. Ask the questions produced by \`leerness clarify "<request>"\` verbatim. Even without detected signals, when your own judgment is split, asking wins.\n- **Preview-approval duty (new features)**: when the user asks for a new feature, **do not write code first**. ① Register a preview with \`leerness preview add "<feature>" --design "<design/UX>" --features "<list>"\`, ② present it and ask for approval or revisions, ③ on approval run \`leerness preview approve <P-ID>\`; on revision \`leerness preview revise <P-ID> --note "..."\` and re-present. **Never write the feature's code before approve.**\n- Classify work with \`leerness route <task-type>\` (planning, feature, bugfix, refactor, research, consistency, release, migration, session-start, session-close, harness-maintenance).\n- Never delete protected files/managed sections — merge, archive, or mark deprecated instead.\n- After meaningful changes update progress-tracker, current-state, task-log, session-handoff.\n- Before claiming completion, self-verify with \`leerness check .\` or \`leerness lazy detect .\` and answer the \`leerness lens\` self-questions per domain (code: "would a senior developer find this needlessly complex?" / design: "would a senior designer and an ordinary user find this pretty and intuitive?").\n- Before changes run the guards: \`leerness scan secrets .\`, \`leerness encoding check .\`.\n- Before duplicating a capability check design-system.md, consistency-policy.md, reuse-map.md.\n- Close every session with \`leerness session close .\` — 9 categories (done/in-progress/incomplete/planned/waiting/on-hold/blocked/dropped/verification) + active-rule verification.\n- Updates: \`leerness update --check\` (detect) → \`leerness update --yes\` (auto-migrate).\n\n## Natural-language retro/insights/brainstorm\n| User phrase | Run immediately |\n|---|---|\n| "retrospective / look back / wrap up" | \`leerness retro\` |\n| "retro for last N days" | \`leerness retro --days N\` |\n| "stats / cumulative metrics / insights" | \`leerness insights\` |\n| "brainstorm about X / materials on X / review before starting X" | \`leerness brainstorm "X"\` |\n\nsession close prints a one-line summary automatically every time and runs a deep retrospective every 5 sessions; call the commands immediately when the user asks explicitly.\n\n## Natural-language standing rules\nWhen the user states a standing rule ("do Y every X"), register it immediately:\n\n| User phrase | Run immediately |\n|---|---|\n| "bump the version every update" | \`leerness rule add "bump version (patch)" --trigger every-update\` |\n| "add patch notes every commit" | \`leerness rule add "add patch notes" --trigger every-commit\` |\n| "deploy at session close" | \`leerness rule add "deploy (release publish)" --trigger session-close\` |\n| "pause/stop rule X" | \`leerness rule pause <ID>\` (find the ID with \`rule list\`) |\n| "remove rule X" | \`leerness rule remove <ID>\` |\n| "stop all rules" | \`leerness rule stop\` |\n| "resume rules" | \`leerness rule resume-all\` or \`leerness rule resume <ID>\` |\n\nAfter registering, report the result (ID + trigger + description) and apply it every session until the user explicitly says stop/remove.\n\n## Automatic rule verification\n- **every-update / version rules**: checks package.json version change.\n- **CHANGELOG / patch-note rules**: checks CHANGELOG.md mtime.\n- **test / verify rules**: checks today's verify traces in review-evidence.md.\n- **deploy / publish / push rules**: not auto-verifiable → guides \`leerness release publish\`.\n\nAutomate verifiable rules with \`leerness release bump\`, \`leerness release note "..."\`, and \`leerness release publish\`.\n`;
     _files['CLAUDE.md'] = `${MARK}\n# Claude Code Instructions\n\nFollow AGENTS.md. Always run \`leerness handoff .\` at the start and \`leerness session close .\` before ending a session.\n\n**⭐ First action every session**: follow the 6-step workflow in \`.harness/session-workflow.md\` (analyze → plan → distribute → sub-agent → verify → close). On drift critical, recover with \`leerness drift check --auto-fix\`.\n\nProtected files must not be deleted. Read .harness/anti-lazy-work-policy.md before claiming completion.\n\n## Natural-language standing rules\nWhen the user states "do Y every X", immediately run \`leerness rule add "Y" --trigger every-X\`. Registered rules are printed by every \`handoff\` and verified by \`session close\`. Only \`rule pause/remove\` when the user explicitly says stop/remove.\n\nSee the "Natural-language standing rules" table in AGENTS.md for the full mapping.\n`;
+    // 1.36.61 (F-05 3회차): .harness 정책·시드 문서군 en 완역 — 지시 5종(1.36.59)에 이어 read-order 가 가리키는 참조층.
+    _files['.harness/project-brief.md'] = fm('project-brief', ['confirming project purpose', 'judging new features', 'planning'], ['project purpose changes', 'users/scope change'], `# Project Brief\n\n## Project\n${project}\n\n## Purpose\n- Update this with the real purpose of the project.\n\n## Users\n-\n\n## Success Criteria\n-\n`, 'en');
+    _files['.harness/plan.md'] = fm('plan', ['before starting work', 'new request received', 'scope change', 'new project detected'], ['plan add/update/drop', 'milestone changes', 'goal changes'], `# Plan\n\n## Goal\n- Maintain the overall plan around the user's purpose.\n\n## Scope\n- Record what is in scope.\n\n## Out of Scope / Dropped\n| ID | Item | Reason | Date |\n|---|---|---|---|\n\n## Milestones\n\n### M-0001. Organize the project plan\nStatus: planned\nProgress: 0%\n\nTasks:\n- [ ] Fill project-brief.md with the real project purpose\n- [ ] Fill context-map.md with the real file structure\n`, 'en');
+    _files['.harness/guideline.md'] = fm('guideline', ['checking quality bar before implementing', 'checking plan-execution criteria'], ['dev standards change', 'verification routine changes'], `# Guideline\n\n## Operating Principle\n- Work against the goal and scope in plan.md.\n- Judge done/not-done by request status in progress-tracker.md.\n- Never record progress percentages here — plan.md/progress-tracker.md are the single source.\n\n## Quality Gate\n- Check the relevant route before changes (\`leerness route <task-type>\`).\n- After changes run \`leerness verify\`, \`leerness audit\`, \`leerness check\`.\n- Before claiming completion run \`leerness lazy detect\`.\n- Close every session with \`leerness session close\`.\n`, 'en');
+    _files['.harness/plan-progress-boundary.md'] = fm('plan-progress-boundary', ['when plan and progress overlap', 'work-tracking structure changes'], ['role-separation criteria change'], `# Plan / Progress Boundary\n\n## plan.md\n- Overall goal, milestones, in/out of scope, plan-change history.\n\n## progress-tracker.md\n- Per-request status, evidence, next action.\n- ID rule: T-0001 monotonically increasing; IDs from plan add are unique across plan/progress.\n\n## guideline.md\n- The execution standards to follow while doing plan/progress work.\n`, 'en');
+    _files['.harness/current-state.md'] = fm('current-state', ['session start', 'taking over work'], ['current state changes', 'next work changes'], `# Current State\n\nUpdated: ${today()}\n\n## Now\n-\n\n## Next\n-\n\n## Blockers\n-\n`, 'en');
+    _files['.harness/writeback-policy.md'] = fm('writeback-policy', ['before finishing work', 'deciding what to update'], ['record targets change'], `# Writeback Policy\n\n- plan.md: user purpose, milestones, scope additions/exclusions\n- progress-tracker.md: per-request status and evidence (updated in place)\n- current-state.md: current state and next work\n- task-log.md: execution history (auto-appended)\n- session-handoff.md: next-session handoff (written by \`session close\`)\n- decisions.md: hard-to-reverse decisions\n- design-system.md: UI/UX/component standards\n- feature-contracts.md: input/output/state/error contracts\n- review-evidence.md: verification results (auto-accumulated)\n`, 'en');
+    _files['.harness/task-type-map.md'] = fm('task-type-map', ['classifying user requests'], ['task types added'], `# Task Type Map\n\n| User Request | Task Type | Route |\n|---|---|---|\n| plan this / draft a roadmap | planning | leerness route planning |\n| implement / build a feature | feature | leerness route feature |\n| fix a bug | bugfix | leerness route bugfix |\n| refactor / clean up | refactor | leerness route refactor |\n| research / compare | research | leerness route research |\n| unify design / consistency | consistency | leerness route consistency |\n| deploy / release | release | leerness route release |\n| migration | migration | leerness route migration |\n| session start / resume | session-start | leerness route session-start |\n| session end | session-close | leerness route session-close |\n`, 'en');
+    _files['.harness/protected-files.md'] = fm('protected-files', ['before deleting/cleaning/migrating files'], ['protected targets change'], `# Protected Files\n\nAI agents must not delete or reset these files without explicit user approval.\n\n- .harness/\n- .harness/skills/\n- .harness/library/\n- AGENTS.md\n- CLAUDE.md\n- .cursor/rules/leerness.mdc\n- .github/copilot-instructions.md\n- .claude/commands/\n- .claude/skills/\n- README.md Leerness managed section\n\nUse merge, archive, or deprecated markers instead of deletion.\n`, 'en');
+    _files['.harness/guardrails.md'] = fm('guardrails', ['before all work', 'before security/permissions/refactoring'], ['prohibition rules change'], `# Guardrails\n\n- Never store tokens/keys/passwords — record only environment-variable names.\n- No unrequested large refactors (user pre-approval when changing 5+ files).\n- Record the impact of API/DB/env-var changes in task-log.\n- Never delete Leerness protected files/managed sections.\n- Keep Korean text as UTF-8 without BOM.\n- Destructive Git operations (\`git reset --hard\`, \`git push --force\`, …) only after explicit user approval.\n`, 'en');
+    _files['.harness/consistency-policy.md'] = fm('consistency-policy', ['before duplicating UI/features', 'judging reuse'], ['consistency policy changes'], `# Consistency Policy\n\nBefore building something that does the same job, find and reuse/extend/connect the existing implementation.\n\n## Recursive Reuse Rule\n1. Find existing elements with the same capability.\n2. Check whether self-reference/base rules/recursive flow is needed.\n3. Reuse or extend the existing element.\n4. If you must build new, record why in reuse-map.md.\n\n## Audit Trigger\n\`leerness audit\` checks:\n- duplicate design-guide files\n- undefined design-system.md tokens\n- empty reuse-map.md with 3+ components/utils found\n- plan vs progress alignment\n`, 'en');
+    _files['.harness/testing-strategy.md'] = fm('testing-strategy', ['before verification', 'before release'], ['test strategy changes'], `# Testing Strategy\n\n- Typecheck (\`tsc --noEmit\` or equivalent)\n- Lint (\`npm run lint\` etc.)\n- Unit/Integration/E2E\n- Manual smoke test\n- Browser/UI smoke (on frontend changes)\n\n## Evidence Format\nEach completed task must reference an evidence ID stored in .harness/review-evidence.md.\n`, 'en');
+    _files['.harness/review-checklist.md'] = fm('review-checklist', ['before PR/review'], ['review criteria change'], `# Review Checklist\n\n- [ ] Aligned with the plan\n- [ ] progress-tracker updated\n- [ ] No protected files deleted\n- [ ] Design/feature reuse checked\n- [ ] No secrets in code (\`leerness scan secrets\`)\n- [ ] Korean encoding OK (\`leerness encoding check\`)\n- [ ] Lazy-work check passed (\`leerness lazy detect\`)\n`, 'en');
+    _files['.harness/release-checklist.md'] = fm('release-checklist', ['before deploy'], ['deploy conditions/env vars/rollback change'], `# Release Checklist\n\n- [ ] \`leerness verify .\`\n- [ ] \`leerness audit .\`\n- [ ] \`leerness scan secrets .\`\n- [ ] \`leerness encoding check .\`\n- [ ] project typecheck/lint/test\n- [ ] env vars (.env.example) synced\n- [ ] rollback path confirmed\n- [ ] CHANGELOG updated\n`, 'en');
+    _files['.harness/session-close-policy.md'] = fm('session-close-policy', ['before session end'], ['session-close format changes'], `# Session Close Policy\n\nEvery session must list:\n- Completed\n- In progress\n- Incomplete\n- Planned\n- Waiting\n- On hold\n- Blocked\n- Dropped\n- Verification (commands run, results)\n- Recommended next direction\n- Next exact step\n\n\`leerness session close\` extracts the 9 categories automatically and writes the handoff block for the next session into session-handoff.md.\n`, 'en');
+    _files['.harness/anti-lazy-work-policy.md'] = fm('anti-lazy-work-policy', ['before claiming completion'], ['lazy-work prevention criteria change'], `# Anti Lazy Work Policy\n\n## Rules\n1. **No completion without evidence**: claiming "done" requires command output/test results/screenshot paths in progress-tracker's evidence column.\n2. **No empty handoff**: if Completed/In Progress/Next Exact Step are all empty at session end, close is marked "insufficient".\n3. **Honest partial implementation**: if not fully implemented, set status \`incomplete\` and write one line in Next Exact Step describing what remains.\n4. **Verification records**: accumulate typecheck/lint/test results in review-evidence.md.\n5. **TODO tracking**: any new \`TODO\`/\`FIXME\`/\`XXX\` in code gets a same-ID row in progress-tracker.\n6. **Automatic false-done detection**: \`leerness lazy detect\` checks done rows without evidence, empty handoffs, untracked new TODOs, and missing test-run traces.\n7. **Quality-lens self-questions**: before claiming done, answer \`leerness lens\` per domain — code: "would a senior developer find this needlessly complex?" / design: "would a senior designer and an ordinary user find it pretty, comfortable, intuitive?". If you cannot answer "yes (pass)", it is not done. When changing one domain, re-check the causally linked domains (↔ in \`lens\` output).\n`, 'en');
+    _files['.harness/session-handoff.md'] = fm('session-handoff', ['session start', 'taking over next work'], ['session end'], `# Session Handoff\n\nLast generated: (auto)\n\n## Completed\n-\n\n## In Progress\n-\n\n## Incomplete / Waiting / On Hold / Blocked\n-\n\n## Dropped\n-\n\n## Verification\n-\n\n## Recommended Direction\n-\n\n## Next Exact Step\n-\n`, 'en');
+    _files['.harness/leerness-maintenance.md'] = fm('leerness-maintenance', ['work start', 'before migration/release'], ['version policy changes'], `# Leerness Maintenance\n\nAI agents should check:\n\n\`\`\`bash\nleerness --version\nleerness self check .\nleerness update --check       # auto-detect with 24h cache\nleerness update --yes         # auto-migrate when a new version is found\ncat .harness/HARNESS_VERSION\nnpm view leerness version\n\`\`\`\n`, 'en');
+    _files['.harness/language-policy.md'] = fm('language-policy', ['before writing documents'], ['language changes'], `# Language Policy\n\nSelected language: ${lang}\n\nAll Leerness notes, skill notes, session reports, and task lists default to the language above (except when the user explicitly requests another language).\n`, 'en');
+    _files['.harness/secret-policy.md'] = fm('secret-policy', ['before skill/deploy/config changes'], ['sensitive-data policy changes'], `# Secret Policy\n\n## Rules\n- Record only environment-variable names; move values to .env.local, CI secrets, or a cloud secret manager.\n- Never leave tokens/passwords/cookies/IDs/card numbers in plaintext in code/docs/logs.\n- Run \`leerness scan secrets .\` before changes.\n\n## Patterns scanned\n- AWS Access Key (\`AKIA[0-9A-Z]{16}\`)\n- GitHub PAT (\`ghp_[A-Za-z0-9]{36}\`)\n- OpenAI key (\`sk-[A-Za-z0-9]{20,}\`)\n- Anthropic key (\`sk-ant-[A-Za-z0-9-]{20,}\`)\n- Google API key, Slack token, generic private key, hardcoded password\n`, 'en');
+    _files['.harness/encoding-policy.md'] = fm('encoding-policy', ['before creating files', 'garbled-Korean reports', 'before deploy'], ['encoding policy changes'], `# Encoding Policy\n\n## Rules\n- All text files are **UTF-8 without BOM**.\n- Windows .bat files start with \`chcp 65001 >nul\`.\n- PowerShell .ps1 files start with \`[Console]::OutputEncoding = [System.Text.Encoding]::UTF8\`.\n- Python files may need \`# -*- coding: utf-8 -*-\` (for Python 2 compat).\n- LF line endings recommended (unify via .gitattributes on Windows too).\n\n## Auto check\n\`leerness encoding check\` verifies BOM, NUL, chcp 65001 in .bat, and Korean round-trip.\n`, 'en');
+    _files['.harness/architecture.md'] = fm('architecture', ['feature work', 'refactoring', 'migration'], ['structure changes'], `# Architecture\n\n## Overview\n- Record the real structure.\n\n## Data Flow\n-\n\n## External Dependencies\n-\n`, 'en');
+    _files['.harness/context-map.md'] = fm('context-map', ['exploring related files', 'before implementing features'], ['file structure changes'], `# Context Map\n\n| Area | Files | Notes |\n|---|---|---|\n| App | src/** | update with real paths |\n| Tests | tests/** | verification paths |\n`, 'en');
+    _files['.harness/decisions.md'] = fm('decisions', ['checking design decisions'], ['important decisions made'], `# Decisions\n\n## Template (example — add real decisions below, outside the code block)\n\n\`\`\`md\n### ${today()} — Decision title\n- Decision:\n- Reason:\n- Alternatives:\n- Impact:\n\`\`\`\n`, 'en');
+    _files['.harness/task-log.md'] = fm('task-log', ['checking work history'], ['after every meaningful task'], `# Task Log\n\n## ${today()}\n- Leerness v${VERSION} initialized.\n`, 'en');
+    _files['.harness/design-system.md'] = fm('design-system', ['UI changes', 'adding components', 'merging design guides'], ['design standards change', 'reusable patterns found'], `# Design System\n\n## Canonical File\nThis file is the canonical design guide (over designguide.md, design-guide.md, etc.).\n\n## Tokens\n| Token | Value | Notes |\n|---|---|---|\n| color.primary | (update with real value) | |\n| color.surface | | |\n| spacing.unit | | |\n| typography.body | | |\n\n## Reusable Patterns\n| Pattern | Where | Reuse Rule |\n|---|---|---|\n`, 'en');
+    _files['.harness/reuse-map.md'] = fm('reuse-map', ['before creating new components/APIs/helpers', 'duplicate capability detected'], ['reusable elements added'], `# Reuse Map\n\n| Capability | Existing Element | Reuse Method | Notes |\n|---|---|---|---|\n`, 'en');
+    _files['.harness/feature-contracts.md'] = fm('feature-contracts', ['before implementing/changing features'], ['feature IO/state/error changes'], `# Feature Contracts\n\n## Template\n- Feature:\n- Input:\n- Output:\n- States:\n- Errors:\n- Related files:\n- Test evidence ID:\n`, 'en');
     _files['.cursor/rules/leerness.mdc'] = `${MARK}\n---\nalwaysApply: true\n---\nFollow AGENTS.md and .harness/context-routing.md.\nRun: \`leerness handoff .\` at session start.\nRun: \`leerness session close .\` at session end.\nPreserve Leerness protected files.\n`;
     _files['.github/copilot-instructions.md'] = `${MARK}\n# Copilot Instructions\n\nUse AGENTS.md and .harness/ as project memory.\nDo not remove protected Leerness files.\nBefore completion, ensure plan.md, progress-tracker.md, current-state.md, session-handoff.md are updated.\n`;
     _files['.harness/session-workflow.md'] = fm('session-workflow', ['session start', 'new user request arrives', 'before distributing complex work'], ['workflow step changes'], `# Session Workflow — 6 steps of AI harness engineering\n\n> **At the start of every session the main agent reads this document first and follows the 6 steps as written.**\n> Same flow regardless of round length/complexity — that is what prevents drift.\n\n## Step 1. Analyze the request + check the environment\n\`\`\`bash\nleerness handoff .            # load context + automatic drift warning\nleerness drift check .        # signals + 4-level verdict\n\`\`\`\n- Decompose the request (5W1H). If ambiguous, ask clarifying questions (except autonomous mode).\n- On drift critical, run \`leerness session close .\` or \`drift check --auto-fix\` first.\n\n## Step 2. Plan\n- 3+ steps → use TodoWrite or \`leerness plan add\`.\n- New capability → search existing assets first via \`leerness reuse-map\` / \`reuse find <query>\`.\n- Multiple modules → define an integration spec up front (e.g. TICK_SPEC.md).\n\n## Step 3. Distribute — sub-agent mapping\n\`\`\`bash\nleerness agents list                  # ready CLIs\nleerness agents quota                 # limits\nleerness agents dispatch "<task>" --to <id>\n\`\`\`\n- Best sub-agent by task type: text/translation/analysis → claude · deep code reasoning → codex · direct file edits → agy --yolo · security review → \`leerness review --persona security\`.\n- **Conflict-prevention rules (mandatory)**: give each sub-agent *the exact files only it may modify*; require mtime-verification reports (concurrent writes risk last-writer-wins); verify against the spec afterwards with \`leerness contract verify\`.\n\n## Step 4. Sub-agent work + individual self-verification\n- Each sub-agent reports only after passing its own module tests.\n- Report format: line counts, tests N/N PASS, issues found, mtime verification.\n\n## Step 5. Integrated verification\n\`\`\`bash\nleerness contract verify SPEC.md src/<mod>.js\nleerness verify-claim T-XXX --run-tests --strict-claims\nleerness review <file> --persona security,performance,ux\n\`\`\`\n- The main agent writes and runs its own integration scenario (independent verification).\n- Cross-check that sub-agent verification and main verification *agree*.\n\n## Step 6. Close + handoff + next-round suggestion\n\`\`\`bash\nleerness session close .             # --suggest on by default\nleerness session close . --no-suggest\n\n# or separately:\nleerness skill suggest .\nleerness drift check .\nleerness audit . --fix\n\`\`\`\n\n## 🧠 Memory CRUD Quick Reference\n\n| Surface | CREATE | READ | DELETE | RESTORE |\n|---|---|---|---|---|\n| **tasks** | task add | task list --json | task drop | task update |\n| **decisions** | decision add | decision list --json | decision drop | memory restore decisions |\n| **lessons** | lesson save | lesson list [--tag] | lesson drop | memory restore lessons |\n| **plan** | plan add | plan list --json | plan remove | memory restore plan |\n| **rules** | rule add | rule list --json | rule remove | (rule pause/resume) |\n\n\`\`\`bash\nleerness memory status [--json]\nleerness memory archive list [--surface s]\nleerness memory restore <surface> <target>\n\`\`\`\n\n**Recovering a wrongly saved item**:\n1. \`memory archive list\` — list restore candidates\n2. \`memory restore decisions "PostgreSQL"\` — archive → active\n3. handoff automatically notes archive activity within 24h every session\n\n## Auto-recovery & security\n- A skipped session close triggers drift critical at the next session start — recover with \`drift check --auto-fix\` (runs session close automatically on critical).\n- handoff auto-recalls past lessons (by current task keywords) and auto-suggests installed skills.\n- handoff prints a 1–2 line security summary (.env ↔ .env.example sync + .gitignore secret gaps); a missing .env in .gitignore is 🚨 CRITICAL — \`LEERNESS_AUTO_SECURITY_FIX=1\` runs \`audit --fix\` automatically.\n- \`leerness health\` gives a one-line combined check (drift + security + skills + usage + tasks).\n\n---\n\n## Quick checklist (before the session ends)\n- [ ] This round's tasks are registered in plan/progress-tracker (or task sync)\n- [ ] Every done item has evidence attached (verify-claim PASS)\n- [ ] contract verify PASS when sub-agents were used\n- [ ] drift score ≤ 30 — \`leerness drift check\`\n- [ ] session close was run\n- [ ] \`leerness health\` overall check\n- [ ] If .env is used: .gitignore secret patterns OK + .env.example synced\n- [ ] On security critical: \`LEERNESS_AUTO_SECURITY_FIX=1\` or \`audit --fix\`\n\n## Anti-patterns (drift signals)\n- ⚠ "work is done, just report and stop" → missed session close → drift critical next session\n- ⚠ "only TodoWrite updated, leerness unused" → \`task sync --from\` or \`task add\` is mandatory\n- ⚠ distributing to sub-agents without explicit file paths → concurrent-write conflicts\n- ⚠ self-reported "tests ran, PASS" only → verify-claim --run-tests not executed\n- ⚠ skipping contract verify → spec-mismatch bugs reach the user\n`, 'en');
@@ -738,9 +766,9 @@ function createBackup(root, reason, files, dry = false) {
 }
 
 // 1.9.1 P2 / 1.9.368 (UR-0025): MERGE_OVERWRITE_FILES → lib/catalogs, managedMerge 순수 코어 → lib/pure-utils. 얇은 래퍼.
-function managedMerge(file, next, previous, archiveDir) {
+function managedMerge(file, next, previous, archiveDir, mergeOpts = {}) {
   const archiveRel = archiveDir ? path.relative(process.cwd(), archiveDir).replace(/\\/g, '/') : '.harness/archive';
-  return _managedMerge(file, next, previous, archiveRel, MERGE_OVERWRITE_FILES);
+  return _managedMerge(file, next, previous, archiveRel, MERGE_OVERWRITE_FILES, mergeOpts);   // 1.36.60: altTemplate/lang 전달
 }
 
 function writeIfSafe(root, file, content, opts = {}) {
@@ -751,7 +779,7 @@ function writeIfSafe(root, file, content, opts = {}) {
   //   병합을 우회해 CLAUDE.md 커스텀 지시가 archive 에만 남았다(1.36.28 보존 재설계의 force 구멍).
   if (already && opts.mergeManaged) {
     const prev = read(p);
-    writeUtf8(p, managedMerge(file, content, prev, opts.archiveDir));
+    writeUtf8(p, managedMerge(file, content, prev, opts.archiveDir, { altTemplate: opts.altTemplate, lang: opts.lang }));
     return { action: 'merged', file };
   }
   if (already && !opts.force) return { action: 'preserved', file };
@@ -988,6 +1016,17 @@ async function install(root, opts = {}) {
   // 1.9.10: 스킬 카탈로그 출처 안내
   // 1.9.184 (사용자 명시): leerness-skillpack 미사용 정책 — 안내 메시지 제거. builtin catalog 만 사용.
   const files = coreFiles(root, lang, skills, { minimal: !!opts.minimal });
+  // 1.36.60 (F-05 2회차, 검수 #1 High): 반대-언어 canonical 템플릿 — 언어 전환 재init/마이그레이션 시
+  //   구 언어 템플릿 라인을 "사용자 커스텀"으로 오인해 통째로 Preserved 에 이월하던 것을 차감 병합으로 해소.
+  //   (검수 2차 High): 차감은 "실제 언어 전환일 때만" — 동일-언어 재init 에 무조건 적용하면 반대-언어 템플릿과
+  //   우연히 일치하는 정당한 커스텀이 유실된다. 이전 저장 언어를 쓰기 전에 포착해 전환 여부로 게이트.
+  let _prevLang = null;
+  try { _prevLang = read(path.join(root, '.harness', 'LANGUAGE')).trim().toLowerCase() || null; } catch {}
+  if (!_prevLang) { try { _prevLang = String(JSON.parse(read(path.join(root, '.harness', 'manifest.json'))).language || '').toLowerCase() || null; } catch {} }
+  const _langSwitched = !!(_prevLang && (_prevLang === 'ko' || _prevLang === 'en') && _prevLang !== lang);
+  const _altLang = lang === 'en' ? 'ko' : 'en';
+  let _altFiles = {};
+  if (_langSwitched) { try { _altFiles = coreFiles(root, _altLang, skills, { minimal: !!opts.minimal }); } catch {} }
   const backup = createBackup(root, opts.force ? 'force' : (opts.migration ? 'migration' : 'init'), files, opts.dry);
   if (opts.dry) {
     log(`Backup target: ${backup.archiveDir}`);
@@ -998,6 +1037,13 @@ async function install(root, opts = {}) {
   }
   const managedOverwrite = new Set([
     'AGENTS.md','CLAUDE.md','.cursor/rules/leerness.mdc','.github/copilot-instructions.md',
+    '.harness/session-workflow.md',   // 1.36.60 (검수 #1): managed 편입 — 종전엔 재init 시 구 언어/구 버전 그대로 방치
+    // 1.36.61 (검수 High): 순수 정책-문서 전체를 managed 병합 경로에 — 언어 전환 시 구 언어로 방치되던 나머지
+    //   (사용자 상태 문서 plan/brief/decisions 등은 제외 — 구조-인지 전략 필요, 데이터라 raw 교체 금지)
+    '.harness/guideline.md','.harness/plan-progress-boundary.md','.harness/guardrails.md',
+    '.harness/consistency-policy.md','.harness/testing-strategy.md','.harness/review-checklist.md',
+    '.harness/release-checklist.md','.harness/session-close-policy.md','.harness/anti-lazy-work-policy.md',
+    '.harness/secret-policy.md','.harness/encoding-policy.md','.harness/language-policy.md',
     '.harness/HARNESS_VERSION','.harness/manifest.json','.harness/LANGUAGE','.harness/skills-lock.json',
     '.harness/context-routing.md','.harness/writeback-policy.md','.harness/task-type-map.md',
     '.harness/leerness-maintenance.md','.harness/protected-files.md','.harness/AX_MIGRATION_GUIDE.md',
@@ -1032,7 +1078,7 @@ async function install(root, opts = {}) {
       actions.push({ file:f, action });
       continue;
     }
-    const r = writeIfSafe(root, f, c, { force: effForce, mergeManaged, archiveDir: backup.archiveDir });
+    const r = writeIfSafe(root, f, c, { force: effForce, mergeManaged, archiveDir: backup.archiveDir, altTemplate: _altFiles[f], lang });
     actions.push(r);
     _done++;
     drawProgress(_done, f);
@@ -5203,6 +5249,25 @@ function _selfTestCases() {
       const s = read(__filename);
       const wired = s.includes("_profile === 'core' ? _ALL_TOOLS.filter") && s.includes("Unknown tool: ${name} (core profile");
       return allExist && loop && cleanSer && wired;
+    } },
+    { name: '언어 전환 병합 (1.36.60, F-05 2회차 검수 High): altTemplate 차감 — 구 언어 템플릿을 커스텀 오인 이월 차단 + 커스텀 보존 + en 래퍼 — 순수 행위검사', run: () => {
+      const p = require('../lib/pure-utils');
+      const koT = '# 지시\n한국어 규칙 1\n한국어 규칙 2\n';
+      const enT = '# Instructions\nEnglish rule 1\nEnglish rule 2\n';
+      const prev = koT + 'MY CUSTOM LINE\n';
+      // 전환: KO(prev) → EN(next), altTemplate=koT — 한국어 템플릿 라인은 이월 안 되고 커스텀만
+      const m = p._managedMerge('AGENTS.md', enT, prev, '.harness/archive', null, { altTemplate: koT, lang: 'en' });
+      const noKoTemplate = !m.includes('한국어 규칙 1') && !m.includes('한국어 규칙 2');
+      const customKept = (m.match(/MY CUSTOM LINE/g) || []).length === 1;
+      const enWrapper = m.includes('carried over from the previous version') && !/[가-힣]/.test(m.replace('MY CUSTOM LINE', ''));
+      // altTemplate 없이(구 동작): 커스텀 오인 이월 — 회귀 방지용 대조군 (구 언어 라인이 이월됨을 확인)
+      const m2 = p._managedMerge('AGENTS.md', enT, prev, '.harness/archive', null, {});
+      const control = m2.includes('한국어 규칙 1');
+      // 멱등: 전환 결과에 같은 병합 재적용 시 커스텀 중복 없음
+      const m3 = p._managedMerge('AGENTS.md', enT, m, '.harness/archive', null, { altTemplate: koT, lang: 'en' });
+      const idem = (m3.match(/MY CUSTOM LINE/g) || []).length === 1;
+      const wired = read(__filename).includes('altTemplate: _altFiles[f]') && read(__filename).includes("'.harness/session-workflow.md',   // 1.36.60");
+      return noKoTemplate && customKept && enWrapper && control && idem && wired;
     } }
   ];
 }
@@ -10091,7 +10156,7 @@ function lazyDetect(root, opts = {}) {
   if (rows.length === 0) { issues++; _warn('progress-tracker is empty (no tasks tracked)',
     { kind: 'progress_empty', severity: 'warn' }); }
   const handoff = exists(handoffPath(root)) ? read(handoffPath(root)) : '';
-  if (!handoff.includes('Last generated:') || handoff.includes('Last generated: (자동)')) {
+  if (!handoff.includes('Last generated:') || handoff.includes('Last generated: (자동)') || handoff.includes('Last generated: (auto)')) {   // 1.36.61 (검수): en 센티널 병기
     issues++; _warn('session-handoff.md never auto-generated',
       { kind: 'handoff_never_generated', severity: 'warn' });
   }
@@ -17220,7 +17285,7 @@ function uiConsistency(root) {
     if (!m) continue;
     const key = m[1].trim();
     const val = m[2].trim();
-    if (key === 'Token' || /^-+$/.test(key) || val === 'Value' || /실제 값으로 업데이트/.test(val) || !val) continue;
+    if (key === 'Token' || /^-+$/.test(key) || val === 'Value' || /실제 값으로 업데이트|update with real value/i.test(val) || !val) continue;   // 1.36.61: en placeholder 병기
     tokens[key] = val;
   }
   const tokenSet = new Set(Object.values(tokens).map(v => v.toLowerCase()));
@@ -20367,7 +20432,7 @@ function adapterCmd(root, tool, opts = {}) {
   const results = [];
   for (const k of a.keys) {
     if (files[k] == null) continue;
-    results.push(writeIfSafe(root, k, files[k], { mergeManaged: managedOverwrite.has(k) }));
+    results.push(writeIfSafe(root, k, files[k], { mergeManaged: managedOverwrite.has(k), lang: lang === 'en' ? 'en' : 'ko' }));   // 1.36.60 (검수 2차 #2): 래퍼 언어 전달 (altTemplate 은 전환-감지 없는 이 경로에선 미적용 — High 회귀 방지)
   }
   if (a.mcp) results.push(_mergeMcpJson(root));
   if (json) { log(JSON.stringify({ adapter: tool, files: results }, null, 2)); return; }

package/lib/audit.js

@@ -55,7 +55,7 @@ function audit(root, opts = {}, deps = {}) {
   const naMarker = '<!-- leerness:na';
   const ds = exists(path.join(root,'.harness/design-system.md')) ? read(path.join(root,'.harness/design-system.md')) : '';
   if (ds.includes(naMarker)) ok('design-system.md marked NA (skipped)');
-  else if (!/\| color\.primary \|/.test(ds) || /\(실제 값으로 업데이트\)/.test(ds)) { warnings++; warn('design-system.md tokens not customized'); _finding('design_system_default', 'warn', 'design-system.md tokens not customized'); }
+  else if (!/\| color\.primary \|/.test(ds) || /\(실제 값으로 업데이트\)|\(update with real value\)/i.test(ds)) { warnings++; warn('design-system.md tokens not customized'); _finding('design_system_default', 'warn', 'design-system.md tokens not customized'); }   // 1.36.61: en placeholder 병기
   else ok('design-system tokens populated');
   const reuse = exists(path.join(root,'.harness/reuse-map.md')) ? read(path.join(root,'.harness/reuse-map.md')) : '';
   const reuseLines = reuse.split('\n').filter(l => l.startsWith('|') && !/Capability|---/.test(l)).length;
@@ -87,12 +87,12 @@ function audit(root, opts = {}, deps = {}) {
   }
   else if (milestoneIds.length) ok('all milestones linked in progress-tracker');
   const handoff = exists(handoffPath(root)) ? read(handoffPath(root)) : '';
-  if (handoff.includes('Last generated: (자동)')) {
+  if (handoff.includes('Last generated: (자동)') || handoff.includes('Last generated: (auto)')) {   // 1.36.61: en 센티널 병기
     warnings++; warn('session-handoff.md never auto-generated (run: leerness session close .)');
     _finding('handoff_not_generated', 'warn', 'session-handoff.md never auto-generated');
     // 1.9.35 #5: --fix → session-handoff.md 자동 생성 마커 갱신
     if (fix) {
-      const stamped = handoff.replace('Last generated: (자동)', `Last generated: ${today()} (leerness audit --fix)`);
+      const stamped = handoff.replace(/Last generated: \((자동|auto)\)/, `Last generated: ${today()} (leerness audit --fix)`);   // 1.36.61: 양 언어 --fix
       writeUtf8(handoffPath(root), stamped);
       ok('  ↳ fixed: session-handoff.md timestamp 갱신');
       fixed++;

package/lib/pure-utils.js

@@ -352,7 +352,7 @@ function _roadmapParseTokens(text) {
     const m = line.match(/^\|\s*([\w.\-]+)\s*\|\s*([^|]+?)\s*\|/);
     if (!m) continue;
     const key = m[1].trim(), val = m[2].trim();
-    if (!key || !val || key === 'Token' || /^-+$/.test(key) || val === 'Value' || /\(실제 값으로 업데이트\)/.test(val)) continue;
+    if (!key || !val || key === 'Token' || /^-+$/.test(key) || val === 'Value' || /\(실제 값으로 업데이트\)|\(update with real value\)/i.test(val)) continue;   // 1.36.61: en placeholder 병기
     if (val.length > 80) continue;
     tokens[key] = val;
   }
@@ -765,10 +765,16 @@ function _mergeReadmeSection(existing, block, START, END) {
 }
 // 1.9.368 (UR-0025): 관리 파일 마이그레이션 머지 순수 코어 — 이전 내용을 migration-preserved 블록으로 보존(데이터/인덱스 파일은 overwrite).
 //   archiveRel(사전 계산된 표시 경로) + overwriteSet 을 인자로 주입 → path/process/상수 비결합(순수).
-function _managedMerge(file, next, previous, archiveRel, overwriteSet) {
+function _managedMerge(file, next, previous, archiveRel, overwriteSet, opts = {}) {
   if (!previous || previous.trim() === next.trim()) return next;
   const tag = '<!-- leerness:migration-preserved -->';
   if (overwriteSet && overwriteSet.has(String(file).replace(/\\/g, '/'))) return next;
+  // 1.36.60 (F-05 2회차, 검수 High): 언어 전환 시 구 언어 canonical 템플릿(altTemplate)의 라인도 차감 —
+  //   종전엔 KO→EN 재init 에서 KO 템플릿 전체가 "커스텀"으로 오인돼 Preserved 에 통째 이월됐다.
+  //   (검수 2차 High): 차감은 "출현 횟수" 기반 — canonical 1회분만 소비해, 사용자가 의도적으로 템플릿 라인을
+  //   복제한 커스텀(2번째 출현)은 생존한다. Set 차감은 그 복제본까지 지웠다.
+  const altCounts = new Map();
+  for (const l of String(opts.altTemplate || '').split('\n').map(x => x.trim()).filter(Boolean)) altCounts.set(l, (altCounts.get(l) || 0) + 1);
   // 1.36.28 (사용자 보고 — 반복 마이그레이션 데이터 유실): 종전 <details> 중첩 설계는 previous 에 preserved 태그가
   //   있으면 통째로 next 만 반환 → 2번째 마이그레이션부터 사용자가 CLAUDE/AGENTS 에 직접 쓴 커스텀 지시가 조용히
   //   사라졌다(매 버전 AI 마이그레이션 흐름에서 상시 발화). 재설계: 라인-집합 diff 로 "새 템플릿에 없는 라인"만
@@ -780,17 +786,26 @@ function _managedMerge(file, next, previous, archiveRel, overwriteSet) {
   for (const raw of previous.split('\n')) {
     const l = raw.trim();
     if (!l || nextLines.has(l)) continue;
-    // preserved 래퍼 자체의 보일러플레이트는 이월 대상 아님 (구 <details> 형식 + 신 평문 형식 모두)
+    const _ac = altCounts.get(l);
+    if (_ac > 0) { altCounts.set(l, _ac - 1); continue; }   // canonical 출현분만 소비 (복제 커스텀 생존)
+    // 1.36.60 (검수 2차 #3): 구버전 workflow 의 자동생성 이력 로그(도구-카운트/마일스톤 불릿)는 커스텀이 아니라
+    //   릴리스마다 낡는 생성물 — 좁은 시그니처만 필터 (사용자 산문이 이 패턴과 겹칠 확률은 무시 가능)
+    if (String(file).includes('session-workflow') && (/^- .*MCP \*\*\d+\s*도구/.test(l) || /^- MCP \*\*\d+/.test(l) || /^- 🏆 마일스톤/.test(l) || /^- 🎉 \*\*\d+ 라운드/.test(l))) continue;
+    // preserved 래퍼 자체의 보일러플레이트는 이월 대상 아님 (구 <details> 형식 + 신 평문 형식 + en 래퍼 모두)
     if (l === tag || l === '---' || l === '<details>' || l === '</details>') continue;
     if (/^```/.test(l) || /^<summary>/.test(l)) continue;
-    if (/^## Preserved previous content|^Previous content was backed up|^`[^`]*archive[^`]*`$|^> 이전 버전에서 이어진/.test(l)) continue;
+    if (/^## Preserved previous content|^Previous content was backed up|^`[^`]*archive[^`]*`$|^> 이전 버전에서 이어진|^> Custom user\/project content carried over/.test(l)) continue;
     if (seen.has(l)) continue;
     seen.add(l);
     custom.push(raw.replace(/\\`\\`\\`/g, '```'));   // 구 형식에서 이스케이프된 펜스 복원
   }
   if (!custom.length) return next;
   const ar = archiveRel || '.harness/archive';
-  return next.trimEnd() + `\n\n---\n${tag}\n## Preserved previous content\n\n> 이전 버전에서 이어진 사용자/프로젝트 커스텀 내용 — 마이그레이션마다 자동 이월됩니다. 전체 원본 백업: \`${ar}\`\n\n` + custom.join('\n') + '\n';
+  // 1.36.60 (검수): 보존 래퍼 문구도 프로젝트 언어를 따른다 — en 프로젝트에 한글 래퍼가 새 한글원을 만들던 것 해소
+  const note = opts.lang === 'en'
+    ? `> Custom user/project content carried over from the previous version — re-carried automatically on every migration. Full original backup: \`${ar}\``
+    : `> 이전 버전에서 이어진 사용자/프로젝트 커스텀 내용 — 마이그레이션마다 자동 이월됩니다. 전체 원본 백업: \`${ar}\``;
+  return next.trimEnd() + `\n\n---\n${tag}\n## Preserved previous content\n\n${note}\n\n` + custom.join('\n') + '\n';
 }
 
 // 1.9.369 (UR-0025): --skills 값 파싱 순수 코어 — catalog 주입(harness skillCatalog 비결합). all/recommended/csv 처리 + catalog 필터.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "leerness",
-  "version": "1.36.59",
+  "version": "1.36.61",
   "description": "The AI-coding operations layer that makes \"done\" require evidence — persistent memory, evidence-gated completion checks, and clean handoffs for any AI agent (Claude Code, Codex, Cursor). State lives as plain files in your repo. CLI + MCP, 0 runtime dependencies.",
   "keywords": [
     "leerness",

package/scripts/e2e.js

@@ -7083,7 +7083,9 @@ total++;
     const H = /[가-힣]/;
     const de = fs.mkdtempSync(path.join(os.tmpdir(), 'leerness-en5-')); _dirs.push(de);
     const r1 = cp.spawnSync(process.execPath, [CLI, 'init', de, '--yes', '--language', 'en', '--no-env'], { encoding: 'utf8', timeout: 30000 });
-    const five = ['AGENTS.md', 'CLAUDE.md', '.harness/session-workflow.md', '.cursor/rules/leerness.mdc', '.github/copilot-instructions.md'];
+    // 1.36.61 (3회차): 정책·시드 문서군까지 확대 — 대표 10종 무한글
+    const five = ['AGENTS.md', 'CLAUDE.md', '.harness/session-workflow.md', '.cursor/rules/leerness.mdc', '.github/copilot-instructions.md',
+      '.harness/guideline.md', '.harness/guardrails.md', '.harness/anti-lazy-work-policy.md', '.harness/plan.md', '.harness/secret-policy.md'];
     const enOk = r1.status === 0 && five.every(p => { const t = fs.readFileSync(path.join(de, p), 'utf8'); return t.length > 100 && !H.test(t); });
     // en 지시 본문의 핵심 계약 존재(번역이 빈 껍데기가 아님) — 검수 #2 반영: lens 질문/체크리스트/안티패턴/자동화 지시까지
     const ag = fs.readFileSync(path.join(de, 'AGENTS.md'), 'utf8');