master-skill 0.12.9 → 0.12.11

package/.claude-plugin/marketplace.json

@@ -9,7 +9,7 @@
     {
       "name": "master-skill",
       "description": "FoJin-powered Buddhist AI persona framework — source-grounded, boundary-aware, fidelity-tested, runtime-ready. 15 prebuilt masters across 印度/汉传/藏传/南传 plus compare, debate, and curriculum meta-skills.",
-      "version": "0.12.9",
+      "version": "0.12.11",
       "source": "./",
       "author": {
         "name": "xr843",

package/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "master-skill",
   "description": "FoJin-powered Buddhist AI persona framework — source-grounded, boundary-aware, fidelity-tested, runtime-ready. 15 prebuilt masters across 印度/汉传/藏传/南传 plus compare, debate, and curriculum meta-skills.",
-  "version": "0.12.9",
+  "version": "0.12.11",
   "author": {
     "name": "xr843",
     "email": "xr843@users.noreply.github.com"

package/.cursor-plugin/plugin.json

@@ -2,7 +2,7 @@
   "name": "master-skill",
   "displayName": "Master Skill",
   "description": "FoJin-powered Buddhist AI persona framework — source-grounded, boundary-aware, fidelity-tested, runtime-ready. 15 prebuilt masters across 印度/汉传/藏传/南传.",
-  "version": "0.12.9",
+  "version": "0.12.11",
   "author": {
     "name": "xr843",
     "email": "xr843@users.noreply.github.com"

package/gemini-extension.json

@@ -1,6 +1,6 @@
 {
   "name": "master-skill",
   "description": "FoJin-powered Buddhist AI persona framework — source-grounded, boundary-aware, fidelity-tested, runtime-ready. 15 prebuilt masters across 印度/汉传/藏传/南传.",
-  "version": "0.12.9",
+  "version": "0.12.11",
   "contextFileName": "GEMINI.md"
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "master-skill",
-  "version": "0.12.9",
+  "version": "0.12.11",
   "type": "module",
   "description": "FoJin-powered Buddhist AI persona framework — source-grounded, boundary-aware, fidelity-tested, runtime-ready. 15 pre-built masters across 印度 / 汉传 / 藏传 / 南传, plus /compare-masters, /master-debate, and /master-curriculum.",
   "bin": {
@@ -16,7 +16,7 @@
     "validate:versions": "python3 scripts/check-manifest-versions.py",
     "test:hook": "bash hooks/tests/test_session_start.sh && bash hooks/tests/test_run_hook.sh && bash hooks/tests/test_run_hook_cmd.sh",
     "test:cli": "node --test tests/cli.test.mjs",
-    "test": "python3 scripts/check-gate-liveness.py && python3 scripts/validate.py --strict && python3 scripts/validate-citation-references.py && python3 scripts/validate-citation-templates.py && python3 scripts/validate-self-audit-sources.py && python3 scripts/validate-fidelity.py && python3 scripts/validate-persona-fidelity.py && python3 scripts/check-manifest-versions.py && python3 scripts/validate-routing.py && python3 scripts/validate-fixture-terms.py && python3 scripts/verify-adjudication.py && python3 scripts/test-fidelity.py --all --dry-run && node --test tests/cli.test.mjs && python3 -m pytest tests/ scripts/tests/ -q",
+    "test": "python3 scripts/check-gate-liveness.py && python3 scripts/validate.py --strict && python3 scripts/validate-citation-references.py && python3 scripts/validate-citation-templates.py && python3 scripts/validate-self-audit-sources.py && python3 scripts/validate-citation-contract.py && python3 scripts/validate-cross-critique.py && python3 scripts/validate-lore-triggers-content.py --strict && python3 scripts/validate-quote-attribution.py && python3 scripts/validate-promptfoo-configs.py && python3 scripts/validate-fidelity.py && python3 scripts/validate-persona-fidelity.py && python3 scripts/check-manifest-versions.py && python3 scripts/validate-routing.py && python3 scripts/validate-fixture-terms.py && python3 scripts/verify-adjudication.py && python3 scripts/test-fidelity.py --all --dry-run && node --test tests/cli.test.mjs && python3 -m pytest tests/ scripts/tests/ -q",
     "test:smoke": "python3 scripts/test-fidelity.py --master yinguang --max-tests 1",
     "prepack": "node bin/cli.mjs list"
   },

package/prebuilt/master-buddhaghosa/references/teaching.md

@@ -45,7 +45,7 @@
 
 > "诸比丘，何为三增上学？增上戒学、增上心学、增上慧学。"
 
-> 出处：【SC: AN 3.88 / Sikkhā Sutta】（SuttaCentral）
+> 出处：【SC: AN 3.88 / Tatiyasikkhā Sutta】（SuttaCentral）
 
 ### 2. 四十种业处（kammaṭṭhāna） — 培育止禅
 

package/prebuilt/master-huineng/references/teaching.md

@@ -90,7 +90,7 @@
 
 ### 菩提本无树偈
 
-神秀偈："身是菩提树，心如明镜台，时时勤拂拭，勿使惹尘埃。"慧能偈："菩提本无树，明镜亦非台，本来无一物，何处惹尘埃。"神秀从有入手，着于渐修；慧能从空入手，直契本性。五祖因此密传衣钵。
+神秀偈："身是菩提树，心如明镜台，时时勤拂拭，勿使惹尘埃。"慧能偈："菩提本无树，明镜亦非台，本来无一物，何处惹尘埃。"（两偈并见《六祖大师法宝坛经·行由品》）神秀从有入手，着于渐修；慧能从空入手，直契本性。五祖因此密传衣钵。
 
 **慧能大师的运用：** 以此说明自性本空，不可执著于"修"与"不修"的对立，直下承当。
 
@@ -104,7 +104,7 @@
 
 ### 一花开五叶
 
-达摩祖师偈："吾本来兹土，传法救迷情。一花开五叶，结果自然成。"后禅宗果然至慧能后开出五家（临济、曹洞、沩仰、云门、法眼），此为禅宗传法的重要预言与象征。
+达摩祖师偈："吾本来兹土，传法救迷情。一花开五叶，结果自然成。"（《六祖大师法宝坛经·付嘱品》引达摩付法偈）后禅宗果然至慧能后开出五家（临济、曹洞、沩仰、云门、法眼），此为禅宗传法的重要预言与象征。
 
 ## 关键术语表
 

package/prebuilt/master-huineng/references/voice.md

@@ -26,9 +26,9 @@
 慧能大师说法直接了当，不立文字但善用譬喻。语言口语化、平实朴素（本人不识字，以白话说法），同时极具穿透力。善用机锋——以反问、棒喝、出人意料的回答打破学人的思维惯性和概念执著。不做长篇大论的理论铺陈，而是直指要害，一语中的。
 
 **示例句：**
-1. "不是风动，不是幡动，仁者心动。"
-2. "何期自性本自清净，何期自性本不生灭，何期自性本自具足，何期自性本无动摇，何期自性能生万法。"
-3. "迷时师度，悟了自度。"
+1. "不是风动，不是幡动，仁者心动。"（《六祖大师法宝坛经·行由品》）
+2. "何期自性本自清净，何期自性本不生灭，何期自性本自具足，何期自性本无动摇，何期自性能生万法。"（《六祖大师法宝坛经·行由品》）
+3. "迷时师度，悟了自度。"（《六祖大师法宝坛经·行由品》）
 
 ### 常用比喻
 

package/prebuilt/master-milarepa/SKILL.md

@@ -12,6 +12,8 @@ sources:
   - title: 密勒日巴尊者传
     bdrc_id: W1GS56158
     tibetan_title: rNam thar
+  - title: 木纳记（尊者传汉译）
+    cbeta_id: B11n0073
 citation_format: "【《{title}》{section}】（BDRC: {bdrc_id}）"
 verified_by: xr843
 verified_at: 2026-05-02

package/prebuilt/master-milarepa/meta.json

@@ -7,7 +7,7 @@
     "version": 1,
     "claim_policy": "declared_sources_only",
     "required_for": ["doctrinal_claim", "practice_guidance", "text_interpretation"],
-    "allowed_source_types": ["tibetan_canon"],
+    "allowed_source_types": ["cbeta", "tibetan_canon"],
     "minimum_claim_coverage": 0.9,
     "live_retrieval_allowed": true
   },
@@ -30,6 +30,12 @@
       "type": "tibetan_canon",
       "id": "BDRC:W1GS56158",
       "title": "密勒日巴尊者传（rNam thar）"
+    },
+    {
+      "type": "cbeta",
+      "id": "B11n0073",
+      "title": "木纳记（尊者传汉译，惹穹多杰札把著、张心若笔录）",
+      "note": "CBETA 大藏经补编第 11 册，29 卷；《密勒日巴尊者传》的 1930 年代汉译，含所录道歌。FoJin 未收"
     }
   ],
   "version": "1.0.0",

package/prebuilt/master-milarepa/references/teaching.md

@@ -20,15 +20,17 @@
 
 米拉日巴反复在道歌中强调四共加行——暇满难得、人身无常、业果不虚、轮回是苦。这是噶举派一切修持的起点，没有这四者的觉受，谈大手印都是空中楼阁。
 
-> "汝今得此暇满身，胜过摩尼如意宝，若不修法令空过，何异盲人入宝山。"
+尊者在戒学歌里说暇满难得：
 
-> 出处：《道歌集·与亲妹白达之歌》（BDRC W1KG1252）
+> "人身难得无暇我亦知，思无常罪苦轮回自輭。"
+
+> 出处：《木纳记》卷十一（B11n0073）；《道歌集》相关诸歌见 BDRC W1KG1252
 
 ### 2. 出离心与精进闭关
 
 噶举派以"实修"为命脉，米拉日巴以亲身雪山闭关树立典范。教导弟子：闻思之后必须实修，实修之根本在于具格上师指导下的长期闭关，不在文字辩论。
 
-> "无人之处一座岩窟，胜过千座有名寺院；荨麻一握果腹之食，胜过百味轮回之筵。"
+山居胜于名寺、粗食胜于丰筵，是《道歌集》闭关诸歌反复申说的意思。（转述，非原文；张澄基汉译本有版权，本仓库无法逐字核对。）
 
 > 出处：《道歌集·闭关诸歌》主旨（BDRC W1KG1252）
 
@@ -36,9 +38,11 @@
 
 噶举派被称为"口耳传承"（snyan brgyud），其根本依止上师亲口传授。米拉日巴反复教导：对上师的信心是一切证悟的根本，没有信心则一切教法皆不入心。玛尔巴对其的"折磨"（修建石塔）非懲罚，而是净除黑魔术业障的方便。
 
-> "上师即三宝总集，信心即解脱根本；离上师别求佛者，如离日月寻光明。"
+尊者负母骨离乡时长歌：
 
-> 出处：《道歌集·上师瑜伽诸歌》主旨（BDRC W1KG1252）
+> "上师不动慈悲口，细译麻把教令勅。故乡有若牢狱鬼，无常如幻上师说，上师恩德最无上。"
+
+> 出处：《木纳记》卷十四（B11n0073）
 
 ### 4. 那洛六法（Naro Chodruk）
 
@@ -54,9 +58,11 @@
 
 米拉日巴的大手印不重经院辩证，而以道歌直指：
 
-> "心性本如虚空，无染亦无不染；念起如云聚散，本不离虚空体。"
+尊者为贡汤施主唱"六个力量"之歌：
+
+> "此心犹如虚空遍，照境法身清净明。……无来无去无所住，三世平等一定住。心中无有生死因，本来清净如虚空。白红云堆自开散，四大不断如足迹。"
 
-> 出处：《道歌集·见地诸歌》主旨（BDRC W1KG1252）
+> 出处：《木纳记》卷二十三（B11n0073）
 
 ### 6. 觉受 (nyams) 与证悟 (rtogs pa) 的区分
 

package/prebuilt/master-milarepa/references/voice.md

@@ -27,11 +27,11 @@
 米拉日巴尊者以**道歌（mGur）**说法，不立繁复经院术语。语言朴实、带浓厚山野气，善以**自然界譬喻**：雪山喻坚定、风喻无常、岩石喻法身、河流喻心相续。常以**亲身经历**为教材——早年咒杀仇家的罪业、玛尔巴上师的折磨、雪山中以荨麻为食的闭关。语气坦率、不矫饰，对修行人的怠惰直接呵斥，但对真诚求法者极为慈悲。
 
 **示例句（皆有典籍出处）：**
-1. "暇满人身难得如优昙，今既得之莫令空过。"（《道歌集》出离主题多处）
-2. "见乃明空双运，行乃自任无作，果即三身圆满。"（大手印见地之歌主旨）
-3. "无人闭关之山洞中，此身一座，便是大圆满座。"（《尊者传·闭关篇》主旨）
+1. "人身难得无暇我亦知，思无常罪苦轮回自輭。"（《木纳记》卷十一）
+2. "上师不动慈悲口，细译麻把教令勅。……上师恩德最无上。"（《木纳记》卷十四）
+3. "此心犹如虚空遍，照境法身清净明。……本来清净如虚空。"（《木纳记》卷二十三）
 
-> **重要**：上述为风格示意，引用时必须查具体卷次／章节。
+> **重要**：以上三句皆为《木纳记》原文（简体转写，省略处用「……」）。大手印见修行果的四分说是后世噶举的归纳，不要写成尊者原话。
 
 ### 常用比喻
 

package/prebuilt/master-milarepa/sources/INDEX.md

@@ -17,7 +17,7 @@
 ## 说明
 
 - 所有片段节选自 BDRC（Buddhist Digital Resource Center）公开元数据所对应的藏文典籍
-- 汉译参考：张澄基《密勒日巴大师全集》（慧炬出版社）
+- 汉译参考：张澄基《密勒日巴大师全集》（慧炬出版社）；CBETA 另收《木纳记》B11n0073（张心若笔录的尊者传汉译，29 卷，含所录道歌，FoJin 未收），人设所引原文据此
 - 英译参考：Garma C. C. Chang, *The Hundred Thousand Songs of Milarepa*; Andrew Quintman, *The Life of Milarepa* (Penguin Classics)
 - 用于教学引用，不代表完整经义；深入研究请查阅 BDRC、84000、Lotsawa House 全本
 - **密法相关章节本目录刻意不收录**（拙火、那洛六法、生起圆满次第具体修法步骤），密法须依止具格上师亲传

package/prebuilt/master-yinguang/references/voice.md

@@ -26,9 +26,9 @@
 印光大师以书信体弘法为主，文风恳切直接，文言白话兼用，严厉而不失慈悲。常以长者对晚辈的口吻说法，语重心长，不务虚华。
 
 **示例句：**
-1. "汝既发心念佛，当须发愿，愿离娑婆，愿生极乐。"
-2. "念佛之人，当吃长素。如或不能，当持六斋或十斋。"
-3. "学佛之人，先须克己慎独，事事皆从心地中做出。"
+1. "愿离娑婆，如狱囚之冀出牢狱。愿生极乐，如穷子之思归故乡。"（正编卷一·与徐福贤女士书）
+2. "念佛之人，当吃长素。如或不能，当持六斋，或十斋。"（续编卷上·答曾怡芝居士四问）
+3. "欲学佛法，先须克己慎独，事事皆从心地中真实做出。"（正编卷一·与丁福保居士书）
 
 ### 常用比喻
 

package/prebuilt/master-yinguang/sources/INDEX.md

@@ -7,7 +7,7 @@
 | 文件 | 来源经典 | CBETA | FoJin | 覆盖主题 |
 |---|---|---|---|---|
 | `wenchao-excerpts.md` | 《印光法師文鈔》 | 未收录 | — | 信愿行三资粮、摄耳谛听念佛、带业往生横超 |
-| `yihanbianfu-excerpts.md` | 《一函遍復》 | 未收录 | — | 敦伦尽分、闲邪存诚、因果教育 |
+| `yihanbianfu-excerpts.md` | 《印光法師文鈔續編》（一函遍復、复唐能诚居士书） | 未收录 | — | 敦伦尽分、闲邪存诚、因果教育 |
 | `jingtu-sanjing-excerpts.md` | 净土三经 | T12n0366/T12n0360 | [20](https://fojin.app/texts/20) / [18](https://fojin.app/texts/18) | 持名念佛、第十八愿、一心不乱 |
 
 ## 引用规范

package/prebuilt/master-yinguang/sources/yihanbianfu-excerpts.md

@@ -1,13 +1,14 @@
-# 《一函遍復》关键片段
+# 《印光法師文鈔續編》关键片段（一函遍復、复唐能诚居士书）
 
 > 印光大师著，出自《印光法師文鈔續編》。CBETA 与 FoJin 均未收录文钞，故无经号、无在线链接。
-> 本文件为教学引用用。节选文字尚未与印本逐字核对。
+> 本文件为教学引用用。以下引文已于 2026-09-16 与殆知阁所藏《文鈔續編》全文逐字核对；
+> 此前两段是真语拼接而成的改写，并非原文，已按原书改正。
 
 ## 敦伦尽分、闲邪存诚
 
 原典（节选）：
 
-> 敦伦尽分，闲邪存诚，诸恶莫作，众善奉行。正心诚意，以立人道之本。然后以此回向净土，求生西方。在家人第一须尽各人本分。如父慈子孝，兄友弟恭，夫和妇顺。
+> 又须父慈，子孝，兄友，弟恭，夫和，妇顺，主仁，仆忠，恪尽己分。不计他对我之尽分与否，我总要尽我之分。能于家庭，及与社会，尽谊尽分，是名善人。……念佛最要紧，是敦伦尽分，闲邪存诚，诸恶莫作，众善奉行。存好心，说好话，行好事。
 
 **引用格式：**【《印光法師文鈔續編》·一函遍復】
 
@@ -23,9 +24,9 @@
 
 原典（节选）：
 
-> 因果者，圣人治天下、佛度众生之大权也。若人人深信因果，则人心自善，风俗自淳，唐虞之治不难复见。
+> 吾常曰，因果者，圣人治天下，佛度众生之大权也。
 
-**引用格式：**【《印光法師文鈔續編》·一函遍復】
+**引用格式：**【《印光法師文鈔續編》·复唐能诚居士书】
 
 **教义要点：**
 - 因果为佛法入门之基础

package/requirements.txt

@@ -4,3 +4,5 @@ requests>=2.32.5,<2.33; python_version < "3.10"
 requests>=2.34.2; python_version >= "3.10"
 pypinyin>=0.55.0
 pyyaml>=6.0.3
+# 3h 把人设引文转成繁体再查 CBETA：检索只认繁体，简体一律 0 条。
+opencc-python-reimplemented>=0.1.7

package/scripts/check-gate-liveness.py

@@ -25,6 +25,7 @@ Usage:
 from __future__ import annotations
 
 import argparse
+import ast
 import functools
 import json
 import re
@@ -168,6 +169,75 @@ ADVISORY_GATES = {
     ),
 }
 
+# A gate that runs nowhere on a pull request has never guarded a change. This repo
+# has shipped that twice: `validate-curriculum-sources.py` was "wired into no
+# workflow, no npm script and no sub-check — only its own unit tests" (see the
+# sub-check in validate.py that now runs it), and on 2026-09-16
+# `validate-citation-templates.py` and `validate-self-audit-sources.py` were found
+# to live only inside `npm test`, which only npm-publish.yml runs, on a published
+# release. Both passed — the defect was latent, which is exactly why nothing
+# surfaced it.
+#
+# Every entry script under scripts/ must therefore be reachable from a workflow
+# that triggers on `pull_request`, or be declared here with the reason it is not.
+# `check_every_gate_runs_on_a_pr` keeps this true in both directions.
+NOT_A_PR_GATE = {
+    "cite.py": (
+        "a reader-facing offline lookup tool, documented in README.md and in the "
+        "personas' own SKILL.md; not a gate over repository content"
+    ),
+    "query.py": (
+        "a reader-facing offline search tool, documented alongside cite.py; not a "
+        "gate over repository content"
+    ),
+    "check-pe-subsystem.py": (
+        "inspects a built Windows executable, which exists only after the desktop "
+        "release build — release-desktop.yml is the only place it can run"
+    ),
+    "reaudit-report.py": (
+        "re-audits a committed eval run's stored answers; run by hand after a paid "
+        "sweep, against a report that does not exist on a PR"
+    ),
+    "regrade-report.py": (
+        "re-grades a committed eval run against the current judge; same as "
+        "reaudit-report.py — it needs a report a PR does not produce"
+    ),
+}
+
+
+# `npm test` is what CONTRIBUTING tells a contributor to run before touching
+# scripts/, in its own words 「避免在 CI 才发现」. A command that exists to pre-empt CI
+# has to cover what CI checks. It has fallen behind twice: pytest was missing from it
+# until 2026-09-03, and on 2026-09-16 four content gates the PR job runs —
+# validate-citation-contract, validate-cross-critique, validate-lore-triggers-content
+# and validate-quote-attribution — were absent, so a contributor could go green
+# locally and still be failed by CI.
+#
+# Anything the per-PR job runs must therefore appear in `npm test` too, or be
+# declared here. `check_npm_test_covers_pr_gates` keeps this true in both directions.
+NOT_IN_NPM_TEST = {
+    "check-eval-sdk-surface.py": (
+        "asserts the pinned eval SDKs still expose what test-fidelity.py calls — it "
+        "needs requirements-eval.txt installed, which a content contributor has no "
+        "reason to have"
+    ),
+    "smoke-eval-sdk.py": (
+        "stands up a local server for a keyless end-to-end SDK smoke; same eval-only "
+        "dependency, and far slower than the content gates around it"
+    ),
+    "select-fidelity-smoke.py": (
+        "picks which persona the CI smoke grades from job metadata — a CI scheduling "
+        "helper, not a check over repository content"
+    ),
+    "check-audit-ignores.py": (
+        "takes the cargo-audit JSON as an argument — security-scan.yml runs "
+        "`cargo audit --file desktop/Cargo.lock --json > audit.json` first. Without a "
+        "Rust toolchain and the advisory database there is nothing for it to read; "
+        "run bare it exits 2 on argparse usage"
+    ),
+}
+
+
 # The shape of a silent skip: a step that exits 0 because a secret is missing.
 _SKIP_ON_MISSING_SECRET = re.compile(r'\[\s+-z\s+"\$\{[A-Z_]+:-\}"\s+\]')
 
@@ -251,6 +321,161 @@ def check_declared_gates_still_exist(workflow_docs: dict[str, dict]) -> list[str
 # ---------------------------------------------------------------------------
 
 
+def _script_references(source: str, scripts: set[str]) -> set[str]:
+    """Which other scripts this source actually runs — imports and loaded filenames.
+
+    Both spellings are in use: `validate.py` loads five siblings through
+    `spec_from_file_location(..., "validate-curriculum-sources.py")`, which puts the
+    literal filename in the source, while `verify_citations.py` is pulled in as
+    `from verify_citations import …`. Counting only workflow text would report both
+    as unreachable and invent a defect where there is none.
+
+    Read through `ast`, not as text. The first version matched filenames anywhere in
+    the source and reported five scripts as reachable on the strength of *comments*:
+    `verify_citations.py` mentions "scripts/query.py" in a comment about a shared
+    guard, and verify_citations is imported by a job the PR runs, so query.py came
+    out "reachable". Comments do not survive parsing, and an exact-match on string
+    constants keeps a docstring that merely names a path from counting as a call.
+    """
+    try:
+        tree = ast.parse(source)
+    except SyntaxError:  # pragma: no cover — a syntactically broken script
+        return set()
+
+    wanted = {name: {name, f"scripts/{name}"} for name in scripts}
+    modules = {name[:-3]: name for name in scripts if "-" not in name[:-3]}
+    hit: set[str] = set()
+    for node in ast.walk(tree):
+        if isinstance(node, ast.ImportFrom) and node.module in modules:
+            hit.add(modules[node.module])
+        elif isinstance(node, ast.Import):
+            for alias in node.names:
+                if alias.name in modules:
+                    hit.add(modules[alias.name])
+        elif isinstance(node, ast.Constant) and isinstance(node.value, str):
+            for name, spellings in wanted.items():
+                if node.value in spellings:
+                    hit.add(name)
+    return hit
+
+
+def pr_reachable_scripts(root: Path, workflow_docs: dict[str, dict]) -> set[str]:
+    """Scripts a pull request actually executes, following indirect calls."""
+    scripts_dir = root / "scripts"
+    names = {p.name for p in scripts_dir.glob("*.py")}
+    # This file names five scripts in NOT_A_PR_GATE, and this file runs on every PR.
+    # Counting its own source as a caller made each declared script "reachable" and
+    # then reported the declaration as stale — the checker proving its own entries
+    # wrong. Declaring a script is not calling it.
+    sources = {
+        p.name: p.read_text(encoding="utf-8")
+        for p in scripts_dir.glob("*.py")
+        if p.name != Path(__file__).name
+    }
+
+    reachable: set[str] = set()
+    for path, doc in workflow_docs.items():
+        triggers = doc.get("on", doc.get(True))
+        keys = set(triggers) if isinstance(triggers, (dict, list)) else set()
+        if "pull_request" not in keys:
+            continue
+        reachable |= {name for name in names if name in (root / path).read_text(encoding="utf-8")}
+
+    # Fixpoint: a script the PR runs may load or import others.
+    while True:
+        grown = set(reachable)
+        for name in list(reachable):
+            grown |= _script_references(sources.get(name, ""), names)
+        if grown == reachable:
+            return reachable
+        reachable = grown
+
+
+def check_every_gate_runs_on_a_pr(root: Path, workflow_docs: dict[str, dict]) -> list[str]:
+    """An entry script must run on a pull request, or say why it does not."""
+    scripts_dir = root / "scripts"
+    if not scripts_dir.is_dir():
+        return []
+    entries = {
+        p.name
+        for p in scripts_dir.glob("*.py")
+        if "def main(" in p.read_text(encoding="utf-8")
+    }
+    reachable = pr_reachable_scripts(root, workflow_docs)
+
+    problems = [
+        f"scripts/{name} runs nowhere on a pull request and is not in NOT_A_PR_GATE "
+        "— a gate whose first real execution is the release has guarded nothing"
+        for name in sorted(entries - reachable)
+        if name not in NOT_A_PR_GATE
+    ]
+    problems += [
+        f"NOT_A_PR_GATE declares {name!r}, but no such script exists — stale entry"
+        for name in sorted(NOT_A_PR_GATE)
+        if name not in entries
+    ]
+    problems += [
+        f"NOT_A_PR_GATE declares {name!r}, but a pull request does run it now "
+        "— drop the entry rather than leave a false caveat standing"
+        for name in sorted(NOT_A_PR_GATE)
+        if name in reachable
+    ]
+    return problems
+
+
+def npm_test_scripts(root: Path) -> set[str]:
+    """The scripts the documented pre-push command actually runs."""
+    package = root / "package.json"
+    if not package.exists():
+        return set()
+    data = json.loads(package.read_text(encoding="utf-8"))
+    command = str((data.get("scripts") or {}).get("test") or "")
+    return set(re.findall(r"scripts/([a-z0-9_-]+\.py)", command))
+
+
+def pr_workflow_scripts(root: Path, workflow_docs: dict[str, dict]) -> set[str]:
+    """Scripts named outright by a workflow that triggers on `pull_request`.
+
+    Direct mentions only, unlike `pr_reachable_scripts`: `npm test` runs commands, so
+    what it has to match is the commands CI runs, not everything those import.
+    """
+    named: set[str] = set()
+    for path, doc in workflow_docs.items():
+        triggers = doc.get("on", doc.get(True))
+        keys = set(triggers) if isinstance(triggers, (dict, list)) else set()
+        if "pull_request" not in keys:
+            continue
+        named |= set(re.findall(r"scripts/([a-z0-9_-]+\.py)", (root / path).read_text(encoding="utf-8")))
+    return named
+
+
+def check_npm_test_covers_pr_gates(root: Path, workflow_docs: dict[str, dict]) -> list[str]:
+    """What CI runs on a PR, `npm test` must run too — or say why it does not."""
+    if not (root / "package.json").exists():
+        return []
+    in_ci = pr_workflow_scripts(root, workflow_docs)
+    in_npm = npm_test_scripts(root)
+
+    problems = [
+        f"scripts/{name} runs on every PR in CI but is not in `npm test` and not in "
+        "NOT_IN_NPM_TEST — the command that exists to pre-empt CI does not cover it"
+        for name in sorted(in_ci - in_npm)
+        if name not in NOT_IN_NPM_TEST
+    ]
+    problems += [
+        f"NOT_IN_NPM_TEST declares {name!r}, but `npm test` runs it now — drop the "
+        "entry rather than leave a false caveat standing"
+        for name in sorted(NOT_IN_NPM_TEST)
+        if name in in_npm
+    ]
+    problems += [
+        f"NOT_IN_NPM_TEST declares {name!r}, but no PR workflow runs it — stale entry"
+        for name in sorted(NOT_IN_NPM_TEST)
+        if name not in in_ci
+    ]
+    return problems
+
+
 def discover_test_files(root: Path) -> list[str]:
     return sorted(
         str(p.relative_to(root))
@@ -356,6 +581,8 @@ def run_all(root: Path, fidelity_report: Path | None = None) -> list[str]:
     workflows = read_workflows(root)
     problems += check_advisory_gates_declared(workflows)
     problems += check_declared_gates_still_exist(workflows)
+    problems += check_every_gate_runs_on_a_pr(root, workflows)
+    problems += check_npm_test_covers_pr_gates(root, workflows)
 
     # check_graded_suites_graded_something shipped fully written and unit-tested
     # but unreferenced by run_all — the anti-fake-green script had a check that

package/scripts/validate-quote-attribution.py

@@ -0,0 +1,126 @@
+#!/usr/bin/env python3
+"""Gate: every line a persona presents as a quotation must name where it is from.
+
+Steps 3h and 3i of the weekly check ask whether a quoted line exists — in CBETA,
+or in the compiled teachings CBETA does not hold. Neither asks whether the reader
+is told *which book* it came from, and that gap is where a misattribution
+survives both: master-nagarjuna's 「宁起我见积若须弥」 is real text, findable in
+CBETA, and not his — it is in 《大宝积经》. "Found in the canon" and "correctly
+attributed" are different questions, and only the first was being asked.
+
+This one is offline, so it runs on every PR instead of once a week.
+
+Attribution counts when the source is named:
+
+  - on the quoted line itself — 《书名》, 【…】, `Toh 3861`, `SC: SN 22.59`, 卷N;
+  - on a `> 出处：…` line a few lines below (the excerpt files put it there);
+  - on a sibling item of the same numbered list (one 出处 covers the block, which
+    is how master-fazang's and master-kumarajiva's voice samples are written);
+  - in the nearest section heading above (master-ajahn-chah's sutta excerpts name
+    the sutta in the `###` heading).
+
+It reuses `tools/verify_sources.collect_persona_quotes`, so the set it checks is
+exactly the set 3h and 3i judge — if the collector learns a new quotation shape,
+this gate covers it the same day.
+"""
+from __future__ import annotations
+
+import os
+import re
+import sys
+from pathlib import Path
+
+_TOOLS = os.path.join(os.path.dirname(os.path.abspath(__file__)), os.pardir, "tools")
+if _TOOLS not in sys.path:
+    sys.path.insert(0, _TOOLS)
+
+try:
+    from verify_sources import PREBUILT_DIR, collect_persona_quotes
+except ImportError as exc:  # pragma: no cover - the gate must not pass silently
+    print(f"FAIL: cannot import the quote collector from tools/ ({exc}).")
+    print("A gate that cannot load what it checks has not checked anything.")
+    sys.exit(1)
+
+# 什么算「写明了出处」。SC:/Toh/PTS 是南传与藏传的写法，卷N 与《…》是汉传的。
+_SOURCE_NAMED = re.compile(
+    r"《[^》]{1,40}》|【|Toh[:\s]\d|SC[:：]|\b[SMAD]N \d|Vism|PTS|卷[一二三四五六七八九十百千\d]|[Ss]utta"
+)
+_NUMBERED_ITEM = re.compile(r"^\s*\d+\.\s")
+_LOOK = 8
+
+
+def attribution(lines: list[str], number: int) -> str | None:
+    """这条引文的出处写在哪儿；哪儿都没写就返回 None。"""
+    i = number - 1
+    if i < 0 or i >= len(lines):
+        return None
+    if _SOURCE_NAMED.search(lines[i]):
+        return "on the line"
+
+    # 摘录文件把「出处」写在引文下方几行。
+    for j in range(i + 1, min(len(lines), i + 1 + _LOOK)):
+        if "出处" in lines[j]:
+            return f"出处 line {j - i} below" if _SOURCE_NAMED.search(lines[j]) else None
+
+    # 编号示例句：整块共用一个出处，挂在其中一条上。
+    if _NUMBERED_ITEM.match(lines[i]):
+        low = i
+        while low - 1 >= 0 and _NUMBERED_ITEM.match(lines[low - 1]):
+            low -= 1
+        high = i
+        while high + 1 < len(lines) and _NUMBERED_ITEM.match(lines[high + 1]):
+            high += 1
+        for j in range(low, high + 1):
+            if j != i and _SOURCE_NAMED.search(lines[j]):
+                return f"sibling list item {j + 1 - low}"
+
+    # 最近的小节标题。
+    for j in range(i - 1, max(-1, i - 1 - _LOOK), -1):
+        if lines[j].startswith("#"):
+            return f"heading {i - j} above" if _SOURCE_NAMED.search(lines[j]) else None
+    return None
+
+
+def unattributed(quotes: list[tuple[str, str, str]], read) -> list[tuple[str, str]]:
+    """[(位置, 引文)]：读者无从知道出自哪部书的引文。"""
+    cache: dict[str, list[str]] = {}
+    missing: list[tuple[str, str]] = []
+    for where, _master, quote in quotes:
+        rel, _, number = where.rpartition(":")
+        if not number.isdigit():
+            continue  # meta.json 的 lore_triggers 之类，不是文档行
+        if rel not in cache:
+            cache[rel] = read(rel)
+        if attribution(cache[rel], int(number)) is None:
+            missing.append((where, quote))
+    return missing
+
+
+def main() -> int:
+    def read(rel: str) -> list[str]:
+        return Path(PREBUILT_DIR, rel).read_text(encoding="utf-8").splitlines()
+
+    quotes = collect_persona_quotes()
+    if not quotes:
+        print("FAIL: the quote collector returned nothing — this gate examined an empty set.")
+        return 1
+
+    missing = unattributed(quotes, read)
+    if missing:
+        print(f"FAIL: {len(missing)} quoted line(s) name no source:")
+        for where, quote in missing:
+            print(f"  {where}\n      「{quote[:56]}」")
+        print(
+            "\nA reader cannot check a quotation whose book is never named, and a line\n"
+            "that happens to exist somewhere in the canon passes the weekly checks while\n"
+            "still being the wrong master's words. Name the work on the line, in the\n"
+            "`> 出处：` line below it, on a sibling numbered item, or in the heading above."
+        )
+        return 1
+
+    print(f"OK: all {len(quotes)} quoted lines name the work they come from.")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

package/tools/compiled-teaching-sources.json

@@ -0,0 +1,89 @@
+{
+  "_comment": "CBETA 不收、但有免费全文可取的编集语录。周检 3i 用它把人设里当原话引的句子拿到原书里逐字找。3h 只能查 CBETA，查不到时对这些祖师一律记『未判定』——2026-09-15 修掉的那批拼接引文（虚云、印光）正是长在这个盲区里。coverage 决定这一步能不能判错：complete 表示该祖师声明的编集语录全都在下面取得到，找不到即伪造；partial 表示还有声明了却取不到全文的，找不到只记未判定，绝不报错。宁可少判，不可错判。",
+  "corpora": [
+    {
+      "master": "master-yinguang",
+      "corpus_title": "《印光法师文钞》",
+      "coverage": "complete",
+      "coverage_reason": "meta.json 声明的编集语录是正编、续编、三编，外加一个只作总名解析用的『Yinguang:Wenchao』（正续三编的合称，非另一部书）。这三编就是《文钞》的全部，且都在殆知阁语料库里。2026-09-16 核验：三个文件在 commit e9e11f1 下分别为 1285781 / 1410227 / 1973347 字节，且 voice.md 现有三条引文逐字命中（正编 2 条、续编 1 条）。故查不到即可判错。",
+      "verified_on": "2026-09-16",
+      "not_a_separate_book": [
+        "Yinguang:Wenchao"
+      ],
+      "not_a_separate_book_reason": "『Yinguang:Wenchao』是正编、续编、三编的合称，meta.json 里留着它只为解析只写总名的引用，并非第四部书。列在这里，coverage=complete 才是可核验的声明而不是一句断言——测试会要求每个声明过的编集来源要么有全文、要么在这里说明为什么不需要。",
+      "texts": [
+        {
+          "id": "Yinguang:WenchaoZhengbian",
+          "title": "印光法师文钞正编",
+          "url": "https://raw.githubusercontent.com/daizhige-org/daizhigev20/e9e11f19b7e6bd9c1284bbdab71d2a9cb94d637c/佛藏/藏外/印光法师文钞.md",
+          "encoding": "utf-8"
+        },
+        {
+          "id": "Yinguang:WenchaoXubian",
+          "title": "印光法师文钞续编",
+          "url": "https://raw.githubusercontent.com/daizhige-org/daizhigev20/e9e11f19b7e6bd9c1284bbdab71d2a9cb94d637c/佛藏/藏外/印光法师文钞续编.md",
+          "encoding": "utf-8"
+        },
+        {
+          "id": "Yinguang:WenchaoSanbian",
+          "title": "印光法师文钞三编",
+          "url": "https://raw.githubusercontent.com/daizhige-org/daizhigev20/e9e11f19b7e6bd9c1284bbdab71d2a9cb94d637c/佛藏/藏外/印光法师文钞三编.md",
+          "encoding": "utf-8"
+        }
+      ],
+      "source_note": "殆知阁古代文献 v2.0（github.com/daizhige-org/daizhigev20），默认分支是 data 不是 master——用 master 取会 404，2026-09-16 就这样踩过一次。地址钉在 commit e9e11f1（2026-09-12）上，分支移动不会悄悄换掉被核对的底本；哪天该 commit 取不到，3i 记未判定而不是判错。url 一律写字面字符、不要预先百分号编码：编码由 fetch_compiled_text 统一做，清单里再编一次就成了 %25 开头的双重编码，一取就 404 —— 2026-09-16 首次实跑正是这样让印光三部全部『取不到』的。"
+    },
+    {
+      "master": "master-xuyun",
+      "corpus_title": "《虚云和尚法汇》《虚云老和尚年谱》",
+      "coverage": "partial",
+      "coverage_reason": "meta.json 声明了开示录、法汇、年谱三种，BFNN 上只有岑学吕编的法汇六部与年谱，没有《虚云老和尚开示录》——后者是净慧所编，《虚云和尚全集》相对岑本新增约六十余万字，仅开示就多出 110 余则。也就是说虚云的真引文完全可能出自这里取不到的那部分，所以这部语料只能用来确认，不能用来判错。哪天开示录有了可取的免费全文，再改成 complete。",
+      "verified_on": "2026-09-16",
+      "texts": [
+        {
+          "id": "Xuyun:Nianpu",
+          "title": "虚云和尚年谱",
+          "url": "http://bookgb.bfnn.org/books2/1184.htm",
+          "encoding": "gb18030"
+        },
+        {
+          "id": "Xuyun:Fahui",
+          "title": "虚云和尚法汇—法语",
+          "url": "http://bookgb.bfnn.org/books2/1185.htm",
+          "encoding": "gb18030"
+        },
+        {
+          "id": "Xuyun:Fahui",
+          "title": "虚云和尚法汇—开示",
+          "url": "http://bookgb.bfnn.org/books2/1186.htm",
+          "encoding": "gb18030"
+        },
+        {
+          "id": "Xuyun:Fahui",
+          "title": "虚云和尚法汇—书问",
+          "url": "http://bookgb.bfnn.org/books2/1187.htm",
+          "encoding": "gb18030"
+        },
+        {
+          "id": "Xuyun:Fahui",
+          "title": "虚云和尚法汇—文记",
+          "url": "http://bookgb.bfnn.org/books2/1188.htm",
+          "encoding": "gb18030"
+        },
+        {
+          "id": "Xuyun:Fahui",
+          "title": "虚云和尚法汇—规约",
+          "url": "http://bookgb.bfnn.org/books2/1189.htm",
+          "encoding": "gb18030"
+        },
+        {
+          "id": "Xuyun:Fahui",
+          "title": "虚云和尚法汇—诗歌偈赞",
+          "url": "http://bookgb.bfnn.org/books2/1190.htm",
+          "encoding": "gb18030"
+        }
+      ],
+      "source_note": "BFNN（bookgb.bfnn.org）转录岑学吕编本，gb18030 编码，默认 Python-urllib 即可取（不像 FoJin 会挡 UA）。2026-09-16 核验：voice.md 现有三条引文逐字命中 1186《法汇—开示》。"
+    }
+  ]
+}

package/tools/fojin-known-absent.json

@@ -66,6 +66,17 @@
       "reason": "大藏经补编（B）与《佛教大藏经》（G）所收的法尊译本不在 FoJin 收录范围内。2026-09-15 以 tools/fojin_bridge.py 直接查证：lookup-cbeta 对 B0067、B0048、B0068、B0029、G2518 以及完整号 B10n0067、G148n2518 均无结果，而同一次请求里的 T0235 正常解析为 text_id=7 —— 是缺这些书，不是接口不通。",
       "verified_absent_on": "2026-09-15",
       "consequence": "master-atisha 引用《菩提道灯论》时拿不到 fojin.app 活链接。离线审计不受影响；周检 3f 的原文取自 CBETA API，照常核对「原典」块。"
+    },
+    {
+      "cbeta_id": "B11n0073",
+      "short_cbeta_id": "B0073",
+      "title": "木纳记（尊者传汉译）",
+      "used_by": [
+        "master-milarepa"
+      ],
+      "reason": "大藏经补编（B）不在 FoJin 收录范围内。2026-09-16 以 tools/fojin_bridge.py 查证：lookup-cbeta 对 'B0073' 与 'B11n0073' 均无结果，而同一次请求里的 'T0235' 正常解析为 text_id=7 —— 是缺这部书，不是接口不通。",
+      "verified_absent_on": "2026-09-16",
+      "consequence": "master-milarepa 引用《木纳记》时拿不到 fojin.app 活链接。离线审计不受影响；周检 3f 的原文取自 CBETA API，照常核对引文。"
     }
   ]
 }

package/tools/verify_sources.py

@@ -673,6 +673,80 @@ def collect_excerpt_quotes() -> list[tuple[str, str, str, int | None]]:
     return quotes
 
 
+def collect_compiled_excerpt_blocks() -> list[tuple[str, str, str]]:
+    """(位置, 引文, 所引篇名)：「原典」块中引用格式指向 CBETA 之外编集语录的那些。
+
+    `collect_excerpt_quotes` 只收引用格式带 CBETA 经号的块，而《文钞》没有经号，
+    于是 master-yinguang 的五个「原典」块对 3f 不可见；它们的 `>` 行又是裸行文、
+    不带引号，`collect_persona_quotes` 同样收不到。2026-09-16 核出其中两块是用
+    真语拼接的改写，却一直以「原典」示人 —— 没有任何一步检查看得见它们。
+    """
+    base = Path(PREBUILT_DIR)
+    blocks: list[tuple[str, str, str]] = []
+    for path in sorted(base.glob("*/sources/*-excerpts.md")):
+        where = path.relative_to(base).as_posix()
+        label_line, lines = 0, []
+        for number, line in enumerate(path.read_text(encoding="utf-8").splitlines(), 1):
+            if line.startswith("原典"):
+                label_line, lines = number, []
+            elif not label_line:
+                continue
+            elif line.startswith(">"):
+                lines.append(line[1:].strip())
+            elif line.startswith("#"):
+                label_line, lines = 0, []
+            elif "引用格式" in line:
+                citation = _DOC_CITATION.search(line)
+                text = citation.group(1) if citation else ""
+                if text and not _DOC_CBETA_ID.search(text) and any(lines):
+                    blocks.append((f"{where}:{label_line}", "\n".join(lines), text))
+                label_line, lines = 0, []
+    return blocks
+
+
+def classify_compiled_excerpt_blocks(
+    blocks: list[tuple[str, str, str]],
+    corpora: dict[str, dict],
+    fetch,
+) -> tuple[list[tuple[str, str, str]], list[tuple[str, str]], list[tuple[str, str]]]:
+    """「原典」块是不是所引编集语录的原文；省略号分段，每段都要在原书里找得到。
+
+    与 3i 判引文行同理：原书取不到一律记未判定，接口不通不是证据。语料不全
+    （coverage=partial）也只能确认、不能定罪 —— 判「原书没有」需要读得到全部。
+    """
+    mismatched: list[tuple[str, str, str]] = []
+    verified: list[tuple[str, str]] = []
+    unknown: list[tuple[str, str]] = []
+    bodies: dict[str, str | None] = {}
+    for where, quote, citation in blocks:
+        master = where.split("/", 1)[0]
+        corpus = corpora.get(master)
+        if not corpus:
+            unknown.append((where, f"{master} declares no fetchable corpus"))
+            continue
+        texts = corpus.get("texts") or []
+        for text in texts:
+            url = str(text.get("url"))
+            if url not in bodies:
+                bodies[url] = fetch(url, text.get("encoding") or "utf-8")
+        readable = [bodies.get(str(t.get("url"))) for t in texts]
+        if not any(body for body in readable):
+            unknown.append((where, "could not read the declared full texts"))
+            continue
+        segments = [s for s in re.split(r"…+", quote) if len(_han_only(s)) >= EXCERPT_MIN_CLAUSE * 2]
+        if not segments:
+            unknown.append((where, "no segment long enough to search"))
+            continue
+        absent = [s for s in segments if not any(b and _han_only(s) in b for b in readable)]
+        if not absent:
+            verified.append((where, citation))
+        elif corpus.get("coverage") == "complete":
+            mismatched.append((where, absent[0].strip(), citation))
+        else:
+            unknown.append((where, f"{master}'s free full texts do not cover every declared compilation"))
+    return mismatched, verified, unknown
+
+
 def cbeta_juan_plain_text(html: str) -> str:
     """`/stable/juans` 返回的 HTML → 正文。
 
@@ -955,6 +1029,281 @@ def classify_bdrc_records(
     return mismatched, unknown
 
 
+CBETA_SEARCH_URL = "https://cbdata.dila.edu.tw/stable/search"
+
+# 「当原话呈现」的三种写法：voice.md 的编号示例句、teaching.md 的引用块、
+# 行内带书名号的「云/曰」。模板句（含「……」或「/」选项）、统一拒答话术、
+# 以及人设自己标了「转述/非原文/主旨」的行都不是引文，不收。
+_QUOTE_SAMPLE = re.compile(r'^\s*\d+\.\s*[“"「『]([^”"」』\n]{8,200})')
+_QUOTE_BLOCK = re.compile(r'^\s*>\s*[“"「『]([^”"」』\n]{8,200})')
+_QUOTE_SAID = re.compile(r'(?:云|曰|偈云|经云|论云)\s*[：:]?\s*[“"「『]([^”"」』\n]{8,200})')
+# 具名引出 + **冒号**：「佛说：""」「神秀偈：""」「慧能曰：""」「达摩祖师偈：""」。
+# 冒号是把「引原典」与「人设自己的话」分开的判别式 —— 后者写作「常说"看看那个想要
+# 解决问题的心"」「先问"为什么想读？"」，一律没有冒号。2026-09-16 量过：这一条能收进
+# 慧能的风幡、神秀与达摩的偈、阿姜查所引三段巴利经文，而不碰任何一句话术示例。
+_QUOTE_ATTRIBUTED = re.compile(
+    r"(?:佛|世尊|[㐀-鿿]{2,6}(?:祖师|大师|尊者|菩萨|长老|禅师|居士)?)"
+    r'\s*(?:偈曰|偈云|偈|曰|说)\s*[：:]\s*[“"「『]([^”"」』\n]{8,200})'
+)
+# 《书名》同行引文，不需要动词：「《金刚经》"一切有为法…"」「闻《金刚经》至"应无所住
+# 而生其心"」。原先只认「云/曰」，这类引文一条都进不来。
+_QUOTE_TITLED = re.compile(r'《[^》\n]{2,30}》[^“"「『\n]{0,10}[“"「『]([^”"」』\n]{8,200})')
+_QUOTE_BOILER = re.compile(
+    r"具格上师|亲近善知识|不可由文字|网络传授|须依止|本平台|不得对个体|面对面访谈"
+    r"|如需深入学习|SuttaCentral|BDRC|fojin"
+    # 书单与指引句：「汉译可参《菩提道灯论》（任杰译）」「《清净道论》汉译：叶均居士
+    # 译本」「…可在 ajahnchah.org 免费下载」。它们写在引号里，却不是谁说过的话，
+    # 送去全文检索只会变成查无此句。2026-09-16 量出 22 条这样的行。
+    r"|可参|可详参|可阅|可查|查阅|译本|出版社|下载|开示全集|不可不读|逐句观照"
+)
+_QUOTE_PARAPHRASE = re.compile(r"转述|非原文|主旨|整理|概括|要旨|讲解")
+# 讲「这句话该不该引、该怎么标」的行，本身不是引文：纠错说明（「常被当作龙树的话
+# 引用，但《大智度论》中没有」）、禁用示例（「不可用宗喀巴的精确分判作为阿底峡立场」）。
+# 收了它们，周检会对一条文档已经查明并改正的记录拉响假警报。
+# 标记必须是关于**引用行为**的成句短语：试过「勿」「不可用」这类泛词，会误伤《坛经》
+# 「勿使惹尘埃」、罗什「且勿急」、虚云「不可用意识思量卜度」这些真引文（2026-09-16 实测）。
+# 「中没有」同样要紧跟书名号，否则撞上「心中没有」之类的寻常行文。
+_QUOTE_META = re.compile(
+    r"常被当作|误传|讹传|应保守表述|不要把|不得加引号|引用规范|disclaimer|》中没有|》中查无"
+)
+_QUOTE_HAN = re.compile(r"[\u3400-\u9fff]")
+
+
+def collect_persona_quotes() -> list[tuple[str, str, str]]:
+    """(位置, 祖师目录, 引文)：人设 references / sources 里当原话呈现的句子。
+
+    3f 只看摘录里的「原典」块，而编造的语录恰恰长在别处 —— 2026-09-15 一次手工
+    核查在 voice.md 的「示例句」里查出玄奘「因明立量，非为诤胜」、智顗「功在渐次，
+    证在圆融」等五条查无出处，还有三条是灌顶、澄观、彭际清的话挂在祖师名下。
+    """
+    base = Path(PREBUILT_DIR)
+    quotes: list[tuple[str, str, str]] = []
+    for meta_path in sorted(base.glob("*/meta.json")):
+        master = meta_path.parent.name
+        for path in sorted((meta_path.parent / "references").glob("*.md")) + sorted(
+            (meta_path.parent / "sources").glob("*.md")
+        ):
+            where_base = path.relative_to(base).as_posix()
+            for number, line in enumerate(path.read_text(encoding="utf-8").splitlines(), 1):
+                if (
+                    "出处" in line
+                    or "引用格式" in line
+                    or _QUOTE_PARAPHRASE.search(line)
+                    or _QUOTE_META.search(line)
+                ):
+                    continue
+                for pattern, needs_title in (
+                    (_QUOTE_SAMPLE, False),
+                    (_QUOTE_BLOCK, False),
+                    (_QUOTE_SAID, True),
+                    (_QUOTE_ATTRIBUTED, False),
+                    (_QUOTE_TITLED, False),
+                ):
+                    match = pattern.search(line)
+                    if not match:
+                        continue
+                    if needs_title and "《" not in line:
+                        break
+                    quote = match.group(1).split("——", 1)[0]
+                    if (
+                        len(_QUOTE_HAN.findall(quote)) >= 8
+                        and "……" not in quote
+                        and "/" not in quote
+                        and not _QUOTE_BOILER.search(quote)
+                    ):
+                        quotes.append((f"{where_base}:{number}", master, quote))
+                    break
+    return quotes
+
+
+def persona_source_families() -> dict[str, set[str]]:
+    """{祖师目录: 声明来源的家族集合}。只声明 CBETA 的人设，引文必须在 CBETA 里。"""
+    families: dict[str, set[str]] = {}
+    for meta_path in sorted(Path(PREBUILT_DIR).glob("*/meta.json")):
+        meta = json.loads(meta_path.read_text(encoding="utf-8"))
+        kinds = {str(src.get("type")) for src in meta.get("sources") or [] if src.get("type")}
+        if kinds:
+            families[meta_path.parent.name] = kinds
+    return families
+
+
+def declared_cbeta_works() -> dict[str, list[str]]:
+    """{祖师目录: [CBETA API 的 work 参数]}，取自各 meta.json 声明的 cbeta 来源。"""
+    works: dict[str, list[str]] = {}
+    for meta_path in sorted(Path(PREBUILT_DIR).glob("*/meta.json")):
+        meta = json.loads(meta_path.read_text(encoding="utf-8"))
+        found = {
+            _cbeta_api_work(str(src.get("id")))
+            for src in meta.get("sources") or []
+            if src.get("type") == "cbeta" and _cbeta_api_work(str(src.get("id")))
+        }
+        if found:
+            works[meta_path.parent.name] = sorted(found)
+    return works
+
+
+# CBETA 全文检索只认繁体：简体「应无所住而生其心」查 0 条，繁体 343 条（2026-09-16
+# 实测，另试过 lang/variants/simplified 等七种参数，都不会放宽）。opencc 的 `s2t`
+# 会出「爲」「衆」这类异体，CBETA 用「為」「眾」，照样查不到，所以用 `s2tw` 再补一层
+# 归一 —— 少了这一层，《坛经》《中论》的真引文都会被报成查无此句。
+_TRADITIONAL_FIX = str.maketrans({"爲": "為", "衆": "眾", "眞": "真", "僞": "偽"})
+
+
+def to_traditional(text: str) -> str:
+    import opencc
+
+    for config in ("s2tw", "s2tw.json"):
+        try:
+            return opencc.OpenCC(config).convert(text).translate(_TRADITIONAL_FIX)
+        except Exception:  # noqa: BLE001 — 配置名在不同发行包里写法不同
+            continue
+    raise RuntimeError("opencc has no s2tw config")
+
+
+def cbeta_search_hits(clause: str, work: str | None = None) -> int | None:
+    """CBETA 全文检索命中数；`work` 限定在一部书里。接口出错返回 None（未知）。"""
+    import urllib.error
+    import urllib.request
+
+    params = {"q": clause, "rows": 1}
+    if work:
+        params["work"] = work
+    url = f"{CBETA_SEARCH_URL}?{urllib.parse.urlencode(params)}"
+    try:
+        with urllib.request.urlopen(url, timeout=30) as response:
+            return json.loads(response.read().decode("utf-8")).get("num_found") or 0
+    except (urllib.error.URLError, OSError, ValueError, TypeError):
+        return None
+
+
+def classify_persona_quotes(
+    quotes: list[tuple[str, str, str]],
+    families: dict[str, set[str]],
+    works: dict[str, list[str]],
+    search,
+) -> tuple[list[tuple[str, str, str]], list[tuple[str, str]]]:
+    """把人设里当原话引的句子分成「CBETA 没有这句」与「比不了」两类。
+
+    判「没有」只对**声明来源全是 CBETA** 的人设成立。master-xuyun 的语录出自
+    《虚云和尚法汇》、master-yinguang 的出自《文钞》，两部都不在 CBETA —— 对它们，
+    查不到只说明这一步够不着，不是伪造。查得到却不在声明作品里的（玄奘引窥基所记
+    的唯识比量、蕅益《要解》在净土十要本），同样记为未知：行文里往往已注明他书。
+    """
+    mismatched: list[tuple[str, str, str]] = []
+    unknown: list[tuple[str, str]] = []
+    for where, master, quote in quotes:
+        clauses = sorted((c for c in quote_clauses(quote) if len(c) >= 4), key=len, reverse=True)
+        if not clauses:
+            unknown.append((where, "no clause long enough to search"))
+            continue
+        clause = to_traditional(clauses[0])
+        anywhere = search(clause, None)
+        if anywhere is None:
+            unknown.append((where, "CBETA did not answer"))
+        elif not anywhere:
+            if families.get(master) == {"cbeta"}:
+                mismatched.append((where, quote, clause))
+            else:
+                unknown.append((where, f"{master} also declares non-CBETA sources"))
+        elif not any(search(clause, work) for work in works.get(master, [])):
+            unknown.append((where, "only in works this persona does not declare"))
+    return mismatched, unknown
+
+
+COMPILED_SOURCES_FILE = Path(__file__).parent / "compiled-teaching-sources.json"
+
+
+def compiled_teaching_corpora() -> dict[str, dict]:
+    """{祖师目录: 该祖师在 CBETA 之外、有免费全文可取的编集语录}。
+
+    3h 只能查 CBETA，所以虚云的《法汇》、印光的《文钞》一律落进「未判定」——
+    2026-09-15 修掉的那批拼接引文正是长在这个盲区里。清单把原书地址登记下来，
+    3i 就能真的进原书逐字找。
+    """
+    if not COMPILED_SOURCES_FILE.exists():
+        return {}
+    data = json.loads(COMPILED_SOURCES_FILE.read_text(encoding="utf-8"))
+    return {str(entry["master"]): entry for entry in data.get("corpora") or []}
+
+
+def _han_only(text: str) -> str:
+    """只留汉字。两边都这么归一，标点和空白的写法差异就不会造成假的对不上。"""
+    return "".join(_QUOTE_HAN.findall(re.sub(r"<[^>]+>", "\n", text)))
+
+
+def fetch_compiled_text(url: str, encoding: str = "utf-8") -> str | None:
+    """取一部编集语录的全文，归一成纯汉字；取不到返回 None（未知，不是「原书没有」）。"""
+    import urllib.error
+    import urllib.parse
+    import urllib.request
+
+    # urllib 不像 curl 会自己处理非 ASCII 路径，原样传中文路径会抛 UnicodeEncodeError。
+    split = urllib.parse.urlsplit(url)
+    safe = urllib.parse.urlunsplit(split._replace(path=urllib.parse.quote(split.path)))
+    try:
+        with urllib.request.urlopen(safe, timeout=90) as response:
+            raw = response.read()
+    except (urllib.error.URLError, OSError, ValueError):
+        return None
+    return _han_only(raw.decode(encoding, errors="replace"))
+
+
+def classify_compiled_teaching_quotes(
+    quotes: list[tuple[str, str, str]],
+    corpora: dict[str, dict],
+    fetch,
+) -> tuple[list[tuple[str, str, str]], list[tuple[str, str]], list[tuple[str, str]], list[str]]:
+    """到编集语录原书里逐字找人设当原话引的句子。
+
+    判「原书没有这句」只对 coverage 标 `complete` 的语料成立：正编、续编、三编就是
+    《文钞》的全部，都取得到，找不到即伪造。虚云标 `partial` —— 净慧编的《开示录》
+    比岑学吕的《法汇》多出六十余万字，BFNN 上没有，找不到只说明这一步够不着。
+    任何一部取不到，也一律记未判定：接口不通不是证据。
+
+    第四个返回值是「整部语料一篇都没取到」的祖师 —— 那说明这一步对他什么也没检查。
+    不把它单独报出来，一个取数早就坏掉的 3i 会年复一年地绿着，跟没有这道门禁一样。
+    """
+    mismatched: list[tuple[str, str, str]] = []
+    verified: list[tuple[str, str]] = []
+    unknown: list[tuple[str, str]] = []
+    bodies: dict[str, str | None] = {}
+    touched: set[str] = set()
+    for where, master, quote in quotes:
+        corpus = corpora.get(master)
+        if not corpus:
+            continue
+        touched.add(master)
+        wanted = _han_only(quote)
+        if len(wanted) < 8:
+            unknown.append((where, "quote too short to search"))
+            continue
+        found_in, unreachable = None, []
+        for text in corpus.get("texts") or []:
+            url = str(text.get("url"))
+            if url not in bodies:
+                bodies[url] = fetch(url, text.get("encoding") or "utf-8")
+            body = bodies[url]
+            if body is None:
+                unreachable.append(str(text.get("title")))
+            elif wanted in body:
+                found_in = str(text.get("title"))
+                break
+        if found_in:
+            verified.append((where, found_in))
+        elif unreachable:
+            unknown.append((where, f"could not read {', '.join(sorted(set(unreachable)))}"))
+        elif corpus.get("coverage") == "complete":
+            mismatched.append((where, quote, str(corpus.get("corpus_title") or master)))
+        else:
+            unknown.append((where, f"{master}'s free full texts do not cover every declared compilation"))
+    unreadable = sorted(
+        str(corpora[master].get("corpus_title") or master)
+        for master in touched
+        if all(bodies.get(str(t.get("url"))) is None for t in corpora[master].get("texts") or [])
+    )
+    return mismatched, verified, unknown, unreadable
+
+
 def verify_ids(bridge, cbeta_map: dict[str, list[str]], titles: dict[str, str]) -> dict[str, dict]:
     """Verify all CBETA IDs and return {full_cbeta_id: {text_id, short_id, title, ...}}.
 
@@ -1274,6 +1623,58 @@ def _run_legacy_link_verification(*, fix: bool) -> int:
     if not bdrc_mismatched and not bdrc_unknown:
         print(f"  All {len(bdrc_sources)} declared BDRC work ids resolve to a record with the declared title")
 
+    # Step 3h: 人设里当原话引的句子，CBETA 里有没有（见 classify_persona_quotes）。
+    print("\n[3h/4] Checking quoted lines in persona docs against CBETA...")
+    persona_quotes = collect_persona_quotes()
+    quote_line_mismatched, quote_line_unknown = classify_persona_quotes(
+        persona_quotes, persona_source_families(), declared_cbeta_works(), cbeta_search_hits
+    )
+    for where, quote, clause in quote_line_mismatched:
+        print(f"    [WRONG] {where}: CBETA has no 「{clause}」 — 「{quote[:40]}」")
+    if quote_line_unknown:
+        print(f"  Could not check {len(quote_line_unknown)} quoted line(s) — unknown, not wrong:")
+        for where, reason in quote_line_unknown:
+            print(f"    {where}: {reason}")
+    if not quote_line_mismatched and not quote_line_unknown:
+        print(f"  All {len(persona_quotes)} quoted lines are in CBETA, in a work the persona declares")
+
+    # Step 3i: 3h 够不着的那些 —— 祖师自己的语录本就在 CBETA 之外，进原书逐字找
+    # （见 classify_compiled_teaching_quotes）。
+    print("\n[3i/4] Checking quoted lines against compiled teachings CBETA does not hold...")
+    compiled_corpora = compiled_teaching_corpora()
+    compiled_mismatched, compiled_verified, compiled_unknown, compiled_unreadable = (
+        classify_compiled_teaching_quotes(persona_quotes, compiled_corpora, fetch_compiled_text)
+    )
+    for where, quote, corpus_title in compiled_mismatched:
+        print(f"    [WRONG] {where}: {corpus_title} has no 「{quote[:40]}」")
+    for title in compiled_unreadable:
+        print(f"    [BROKEN] {title}: not one declared full text loaded — this step checked nothing")
+    if compiled_verified:
+        print(f"  Verified {len(compiled_verified)} quoted line(s) in the compiled teachings:")
+        for where, title in compiled_verified:
+            print(f"    {where}: {title}")
+    if compiled_unknown:
+        print(f"  Could not check {len(compiled_unknown)} quoted line(s) — unknown, not wrong:")
+        for where, reason in compiled_unknown:
+            print(f"    {where}: {reason}")
+    if not compiled_corpora:
+        print("  No compiled-teaching corpora are declared")
+
+    # 同一步里的第二类：「原典」块本身。3f 只认带 CBETA 经号的引用格式，《文钞》
+    # 没有经号，这些块此前对每一步检查都不可见（见 collect_compiled_excerpt_blocks）。
+    compiled_blocks = collect_compiled_excerpt_blocks()
+    block_mismatched, block_verified, block_unknown = classify_compiled_excerpt_blocks(
+        compiled_blocks, compiled_corpora, fetch_compiled_text
+    )
+    for where, segment, citation in block_mismatched:
+        print(f"    [WRONG] {where}: {citation} has no 「{segment[:40]}」")
+    if block_verified:
+        print(f"  Verified {len(block_verified)} 「原典」 block(s) word for word in the compiled teachings")
+    if block_unknown:
+        print(f"  Could not check {len(block_unknown)} 「原典」 block(s) — unknown, not wrong:")
+        for where, reason in block_unknown:
+            print(f"    {where}: {reason}")
+
     # Step 4: Update URLs
     # Build replacement map: full_cbeta_id -> str(internal_text_id)
     id_replacement_map: dict[str, str] = {}
@@ -1321,6 +1722,10 @@ def _run_legacy_link_verification(*, fix: bool) -> int:
     print(f"  Doc citation links to another work: {len(doc_mismatched)}")
     print(f"  Excerpt quotes not in the cited text: {len(quote_mismatched)}")
     print(f"  BDRC records that do not match: {len(bdrc_mismatched)}")
+    print(f"  Quoted lines CBETA does not have: {len(quote_line_mismatched)}")
+    print(f"  Quoted lines the compiled teachings do not have: {len(compiled_mismatched)}")
+    print(f"  Compiled teaching corpora that could not be read: {len(compiled_unreadable)}")
+    print(f"  Excerpt blocks the compiled teachings do not have: {len(block_mismatched)}")
     if unknown_to_cbeta:
         print(f"  CBETA unreachable for:     {len(unknown_to_cbeta)} (not counted as wrong)")
     if dry_run and all_changes:

package/tools/version_manager.py

@@ -67,7 +67,14 @@ def rollback(teacher_dir: str, target_version: str) -> bool:
 
 
 def cleanup_old_versions(teacher_dir: str) -> int:
-    """Remove old versions beyond MAX_VERSIONS limit."""
+    """Remove archived versions beyond MAX_VERSIONS, oldest first.
+
+    Prunes only what `list_versions` reports — directories named `v…` — and never a
+    `_before_rollback` copy. The first version selected *every* directory under
+    `versions/`, so anything else a maintainer kept there was deleted once the
+    archive passed the limit, and the backup `rollback` writes so that a bad
+    rollback can be undone was evicted by age like an ordinary version.
+    """
     versions_dir = os.path.join(teacher_dir, "versions")
     if not os.path.exists(versions_dir):
         return 0
@@ -75,8 +82,11 @@ def cleanup_old_versions(teacher_dir: str) -> int:
     entries = []
     for entry in os.listdir(versions_dir):
         entry_path = os.path.join(versions_dir, entry)
-        if os.path.isdir(entry_path):
-            entries.append((entry_path, os.path.getmtime(entry_path)))
+        if not os.path.isdir(entry_path):
+            continue
+        if not entry.startswith("v") or entry.endswith("_before_rollback"):
+            continue
+        entries.append((entry_path, os.path.getmtime(entry_path)))
 
     entries.sort(key=lambda x: x[1], reverse=True)