integrate the skills

split the review skills into self- and other-
the self- variant makes informational comments
the other- variant makes nice comments

Google docs is a failure

skills/self-review/SKILL.md

@@ -0,0 +1,248 @@
+---
+name: self-review
+description: Review your own CN→EN translations — three-pass review (terminology → mechanical → flow). Edit commented.dj with patch. For reviewing someone else's work, load other-review.
+---
+
+{% Serves AGENTS.md Workflow B1 (Self-Review) %}
+
+# Self-Review（自审）
+
+Review YOUR OWN translations. You created the English — you can edit freely.
+For reviewing someone else's work, load `other-review` instead.
+
+Two input formats: CSV/XLSX batch, or `.dj` comparison file.
+
+`bilingual.dj` is script-generated ground truth — **never edit it.**
+Copy to `commented.dj` for all review work. Always use `patch` (mode='replace')
+for edits — not regex-based string replacement in `execute_code`.
+
+## Workflow A: CSV/XLSX batch review
+
+Use when input is a CSV/XLSX with `Chinese`/`English` columns. Produces an `edit-suggestions.dj` file.
+
+### 1. Get the data into CSV
+
+If XLSX, export to CSV (or use `openpyxl`). CSV is faster.
+
+### 2. Read the full file
+
+Use `read_file` with offsets for complete coverage. Don't sample.
+
+### 3. Write a systematic analysis script
+
+Write to `/tmp/script.py`, run with `python3 /tmp/script.py`. No heredocs or `-c`.
+The script should:
+- Parse CSV with `csv.DictReader`
+- Apply detection rules per category
+- Collect issues: row number, CN text, EN text, problem, suggested fix
+- Group/deduplicate identical issues
+
+Common detection categories:
+- **Buddhist terminology**: 正念→mindfulness (not "righteous thoughts"), 布施→generosity (not "alms")
+- **Identity terms**: 学士/修士/胜士/智士 are practice stages, not titles
+- **Literal machine translations**: "hard drive" for 硬盘 (endurance)
+- **四摄法 terms**: 同事→"acting in harmony", 爱语→"kind speech"
+- **Grammar**: subject-verb agreement, unbalanced quotes
+- **Typos/formatting**: Chinese punctuation in English, "IOS"→"iOS"
+- **Inconsistency**: same CN term translated differently across rows
+
+### 4. Write edit-suggestions.dj
+
+Format:
+```
+# 1
+
+original: <Chinese text or key term>
+translated: <current English>
+
+<Explanation and suggested fix.>
+
+# 2
+...
+```
+
+One entry per problem category, not per row. Mention affected row numbers.
+
+## Workflow B: .dj file review (self-review)
+
+Use when you translated the text and want to review your own work.
+
+### File rules
+
+1. `bilingual.dj` — extracted from script. **Never edit.**
+2. `cp bilingual.dj commented.dj` — all edits go here
+3. `{% ... %}` comments document non-obvious translation choices
+
+### 1. Read the full file
+
+Use `read_file` with `offset` and `limit` for full coverage of large files
+(this session had 322 lines, 93KB). For files >200 lines, paginate
+explicitly rather than reading the whole thing at once.
+
+If you have already read part of the file with `read_file` earlier in the
+session, use `terminal: cat` (or `sed -n 'A,Bp'`) to get an un-deduped view
+of the rest. `read_file` deduplicates within a session.
+
+### 2. Scan for problems — three passes, in order
+
+The review is best done in three distinct passes, each catching a different
+category of error. Don't try to catch everything in one scan.
+
+**Pass 1 — terminology + consistency + line count** (fast, mechanical):
+- **Line count**: source and target must match exactly. Mismatch means paragraphs were dropped, merged, or split.
+- Same CN term translated differently across the file (e.g. 人生佛教/人间佛教 conflation, 恨→resentment, 修行/修学)
+- Buddhist terminology against the MPI terms DB (see terms-db-alignment below)
+- Mistranslation of key terms, wrong proper names, garbled text
+- Mid-paragraph truncation: CN covers 3–5 clauses but EN stops after 1–2 sentences. Signal: CN has quoted speech, poems, or a rhetorical climax absent from EN. Flag as "Missing Content" not "Incomplete."
+- Use `references/common-issues-taxonomy.md` as a structured checklist for accuracy issues
+
+**Pass 2 — mechanical/formatting** (also mechanical, but easy to skip):
+- TOC format: AGENTS.md says TOC must be plain bullet list, no link targets. Strip `[I. Heading](#...)` markdown links if present.
+- Double words, double punctuation, capitalisation typos, processing artifacts, stray spacing in Chinese text
+- Numbering mismatches between CN and EN headings
+- Redundant English calques: when the target mirrors a Chinese grammar pattern literally, it can read as a typo (e.g. "mind of death-mindfulness" for 念死之心 — should be "mindfulness of death"). Clunky idioms: 一念之差 → "a single thought of difference" is unidiomatic. Standard renderings exist (e.g. "a single errant thought," "a moment's carelessness," or rephrase as "a single thought can make all the difference").
+
+**Pass 3 — flow/tonal/calques** (read the whole English as prose):
+- Re-read the full English target. Does it hang together as prose, or does it read as "translationese"?
+- Dramatic verbs that are calques of Chinese: "draw forth," "into full play," "shoulder," "look to with hope." See `references/translation-pitfalls.md` for the full calque checklist.
+- Subject-shift calques: English substitutes a concrete agent (practitioners, people) for an abstract system noun (Buddhism, religion) — see pitfalls.
+- Factual inconsistencies across paired descriptions of the same person/place/thing.
+- Tonal coherence inside parallel lists: verb choice should be identical across First/Second/Third items.
+- Intensifier drift: same intensifier ("profoundly important") used 3+ times in one section reads as over-translation.
+- The user may explicitly request this pass ("are the words together nicely?", "does it read well?"). Treat such prompts as a signal to do the full re-read, not just spot-check.
+
+**Terms database drift** (cross-cutting — apply during Pass 1):
+- Cross-reference glossary terms against the MPI terms database
+- CLI preferred: `python3 $MPI_PROJECT_ROOT/terms-search/search.py <query>`. For a review, batch many queries in one `execute_code` script (subprocess loop) — one terminal call per term is slow and noisy.
+- Source priority: DoT定稿 > 内部特色词 > 佛教术语 > 经论名
+- Fix both glossary comments AND body text
+- See `references/terms-db-alignment.md` for batch-lookup patterns
+
+### 3. Dump findings
+
+Write to `translation-findings.dj` for issues that don't fit as inline fixes:
+
+```
+Finding N — Title (line numbers)
+  Chinese: ...
+  English: ...
+  Issue: description
+```
+
+### 4. Apply fixes with `patch`
+
+Surgical string replacement in `commented.dj` with `patch` (mode='replace').
+Never use regex-based string replacement in `execute_code` for .dj edits.
+`patch` is safer, surfaces conflicts, and produces a reviewable diff.
+Verify with `cat` — never rely on `read_file` (session dedup).
+
+### 5. Add inline edit suggestions
+
+Copy `bilingual.dj` → `commented.dj`, then apply inline corrections + `{% %}` comments.
+
+### 6. Final sweep
+
+Run `python3 scripts/sweep.py <source.dj> <target.dj> [--stale term1,term2] [--new term1,term2]`. This runs all mechanical checks in one call: line parity, heading parity, Unicode em/en-dashes, Markdown bold, Chinese punctuation, TOC link artifacts, unbalanced quotes, and stale/new term assertions. Run even when no content patches were needed — it serves as final validation.
+
+## Buddhist terminology reference
+
+See `references/buddhist-terminology.md` for Chinese-English term mappings and common pitfalls.
+
+## Workflow C: Typeset proofread (DOCX manuscript vs PDF layout)
+
+Use when the user gives a manuscript DOCX and a typeset PDF and asks to proofread.
+Goal: catch typesetting errors (missing text, typos, wrong special characters, bad line
+breaks), not translation quality.
+
+### 0. Clarify scope FIRST
+
+Before any extraction: ask what they want checked. "Proofread" can mean:
+- Text accuracy (missing/doubled words, typos introduced by typesetter)
+- Special characters (quotes, dashes, ellipses)
+- Formatting (page numbers, headers, TOC layout)
+- All of the above
+
+Do not run extraction pipelines until scope is clear.
+
+### 1. Extract text
+
+- DOCX → plain: `pandoc file.docx -f docx -t plain --wrap=none`
+- PDF → plain: `pdftotext -layout file.pdf` (preserves positional info)
+
+### 2. Clean PDF artifacts
+
+- Strip InDesign slug lines, page headers, page numbers
+- Join hyphenated line breaks (line ending `-` + next line starting lowercase)
+- Fix drop-cap artifacts (e.g. `L iving` → `Living`)
+
+### 3. Compare
+
+- Extract English paragraphs from DOCX (skip Chinese lines, match blank-line pattern)
+- Check each DOCX paragraph exists as substring in PDF body text
+- Flag paragraphs not found; investigate each (may be heading renumbering, not missing)
+
+### Pitfalls specific to this workflow
+
+- **PDF paragraph joining is lossy** — page breaks split paragraphs. Don't expect
+  perfect paragraph matching; check content coverage, not paragraph identity.
+- **Heading numbering differs** — DOCX has `1.`, `(1)`; PDF has `I`, `1)`. Ignore
+  heading-only differences.
+- **InDesign PDFs insert extra spaces** around drop caps and special characters.
+  Normalize multi-space to single space before comparison.
+
+### Pitfalls
+
+- **Never edit `bilingual.dj`** — it's script-generated ground truth. Copy to `commented.dj` first.
+- **Use `patch`, not regex** — for all `.dj` edits. `patch` surfaces conflicts and produces diffs.
+- **`commented.dj` comments must not split paragraphs** — always place `{% %}` after the FULL EN paragraph, not mid-sentence. Scan for merged comments after insertions and split them. Collapse triple+ blank lines created by comment insertions.
+- **Clarify scope before diving into extraction pipelines** — if the user says
+  "proofread this" or "校对这篇文章", ask what specifically they want checked
+  before running pandoc/pdftotext. Getting interrupted mid-pipeline wastes
+  context.
+- **Don't use heredocs or `-c`** — write to `/tmp/script.py` first
+- **Deduplicate aggressively** — group by problem type, not per-row
+- **Buddhist terminology is technical** — don't guess. When uncertain, flag for review
+- **Never delete .dj comparison files** — intentional work artifacts
+- **Verify patches with `cat`** — `read_file` dedup makes it unreliable
+- **Re-read before fixing** — user may have made interim edits
+- **Batch terminology lookups** — when checking many terms against the terms DB, run them in one `execute_code` script that loops over a query list and calls `search.py` via `subprocess.run`. One terminal call per term floods the context with repetitive output.
+
+## Human Review Protocol (审议)
+
+When giving feedback to human translators — whether in a review team or as an AI
+assistant flagging issues — follow the Oriental Translation Workshop protocol.
+See `references/deliberation-protocol.md` for full guidance.
+
+Key points:
+- **Rejoice first** (随喜): affirm what works before flagging issues
+- **Three-tier issues**: Level 1 (spelling/grammar/format) — fix directly. Level 2
+  (omission/mistranslation/wordiness) — suggest or fix with tracked changes. Level 3
+  (citation versions / marginal wording) — discuss with translator
+- **Tone**: questions, not commands; collaborative inquiry, not correction
+- **Address translator as 菩萨** (Bodhisattva) — respectful peer
+
+## Common Issues Taxonomy
+
+Use `references/common-issues-taxonomy.md` as a structured checklist when reviewing.
+Categories:
+
+- **Accuracy**: omission, mistranslation (over-free, over-literal, misunderstanding,
+  wrong word choice), overtranslation, terminology errors
+- **Readability**: redundancy (long sentences, passive voice, nominalization),
+  poor structure (top-heavy sentences), wrong register, weak transitions
+
+## References
+
+- `references/buddhist-terminology.md` — Chinese-English Buddhist term mappings and pitfalls
+- `references/terms-db-alignment.md` — Batch-aligning glossary terms against the MPI terms database
+- `references/translation-pitfalls.md` — Recurring CN→EN mistranslation patterns (关爱→compassion, 生生增上, etc.)
+- `references/proofreading-patterns.md` — DOCX/PDF extraction techniques, block-based pairing, common manuscript issues
+- `references/docx-md-extraction.md` — Extracting `.docx.md` (pandoc markdown) to bilingual, TOC guards, CN/EN boundary regex
+- `references/edit-suggestions-in-bilingual.md` — Inline edit suggestions + `{% %}` comments, two-file comparison (commented.dj / bilingual.dj)
+- `references/deliberation-protocol.md` — Oriental Translation Workshop review protocol: tiers, rejoicing, tone
+- `references/common-issues-taxonomy.md` — Structured taxonomy of accuracy and readability issues with examples
+
+## Scripts
+
+- `scripts/sweep.py` — Mechanical validation sweep for completed reviews
+- `scripts/review_csv.py` — Batch CSV/XLSX translation review

skills/self-review/references/buddhist-terminology.md

@@ -0,0 +1,47 @@
+# Buddhist Text Translation — Terminology
+
+Terms encountered in Chinese-English translation of Dharma study materials. These may vary by translator/context; document actual usage per-project.
+
+## Section headers (common triad)
+
+| Chinese | English options seen | Notes |
+|---------|---------------------|-------|
+| 法义 | Understanding, Dharma Teachings | |
+| 思考 | Contemplation, Reflection | Consistency within a document matters more than which word |
+| 练习 | Practice, Application, Exercises | "Application" seen as section header; "Practice"/"Exercise" in running text |
+
+## Key Buddhist terms
+
+| Chinese | English | Pitfalls |
+|---------|---------|----------|
+| 慈经 | Metta Sutta (Karaniya Metta Sutta) | NOT "Mettavihari Sutta" |
+| 回向 | Dedication (of merit) | |
+| 因缘之网 | Web of Causes and Conditions | Also "Network of Causes and Conditions" |
+| 感恩 | Gratitude | |
+| 众生 | sentient beings | Consistent throughout |
+| 使人内心调柔 | makes one's heart gentle | 使人 = makes ONE(self), never "makes others" |
+| 利益思维 | benefit-oriented thinking | NOT "mindset of benefiting others" — it's about considering benefits TO oneself |
+| 恩田 | field of gratitude / gratitude as a field of merit | |
+| 观照 | attend to the mind / mindful observation | Contemplative practice, not intellectual study. NOT "observe" (passive) or "study" (analytical). |
+| 闻思修 | hearing, contemplating, cultivating | 修 = broad cultivation/practice, not specifically 禅 (meditation). Distinct from 禅修 (meditative cultivation). |
+| 八步三禅 | Eight Steps and Three Meditations | Community-specific structured contemplative method |
+| 传帮带 | transmit, help, guide (three-part mentoring) | Core community methodology |
+| 分灯 | lamp-dividing (decentralization) | Deliberate decentralization of authority across nodes |
+| 自觉 | self-awareness, voluntary commitment | First of the "Three Spirits" |
+| 法治 | rule-based governance | Second of the "Three Spirits" |
+| 无我利他 | selfless service to others | Third of the "Three Spirits" |
+| 凡夫心 | ordinary mind | Mind governed by afflictions, contrasted with awakened mind |
+| 贪嗔痴 | greed, anger, ignorance (三毒) | The three root poisons: rāga, dveṣa, moha |
+| 愿心 | mind of vows, bodhicitta aspiration | Plural "vows" in English |
+| 重要感、优越感、主宰欲 | sense of importance, superiority, desire to control | Three ego-driven motivations |
+
+## Structural patterns
+
+- Section numbering: Chinese uses 一、二、三... English should pick one style (Part One/Two, First/Second, I/II) and stick with it.
+- Poetry/prayer blocks: Chinese uses parallel structures (感恩... 感恩...; 愿... 愿...). English must match the parallelism.
+- 愿 (yuàn) at sentence start = optative "May..." — not "we hope that", not "we should".
+
+## Formatting artifacts
+
+- Stray spaces between Chinese characters (eg. `感 恩 研 究`) are justified-text paste artifacts from the source document — remove them.
+- `<!-- === Progress : Below are unprocessed === -->` is a processing marker — remove from final file.

skills/self-review/references/common-issues-taxonomy.md

@@ -0,0 +1,174 @@
+# Common Translation Issues Taxonomy
+
+From 译文常见问题与案例示范 (Common Issues in Translation, 2026-03-30).
+Categorized by accuracy and readability.
+
+## Accuracy Issues
+
+Standard: convey meaning at the content level, not word-by-word correspondence.
+"Translation is not simply word conversion."
+"You must understand what thought it is actually trying to express."
+
+### 1. Omission (漏翻)
+
+Content present in source but absent from target.
+
+Example:
+> 随着黑白牌的运用，象征这部分已能为大家提供系统的学习和训练资料。
+>
+> With the help of the Black-and-White Cards, this approach now offers a structured
+> system of study.
+>
+> *Missing:* 学习和训练 (learning and training) — "study" only covers half.
+> Also: 象征这部分 (symbolizing this aspect) is dropped.
+
+### 2. Mistranslation (错翻)
+
+#### a. Over-free translation, weakening the original
+
+> "道德"一词，人们耳熟能详。即使在道德日渐边缘化的今天，人们依然会用"这人
+> 很有道德""这么做太缺德了"来评价周遭人事。
+>
+> "Morality" is a familiar term, and even in today's world, where it has become
+> increasingly marginalized, expressions like "immorality" are still commonly used
+> to describe certain behaviors and individuals.
+>
+> *Issue:* The source gives two concrete colloquial expressions ("这人很有道德" /
+> "这么做太缺德了") — the target collapses them into a single abstract "expressions
+> like 'immorality'." The vividness and specificity are lost.
+
+#### b. Over-literal translation, harming comprehension
+
+> 作为素菜馆，应该尽量提供绿色食品。
+>
+> A vegetarian restaurant should strive to offer green food.
+>
+> *Issue:* 绿色食品 = "safe and healthy food" (Chinese idiom), not
+> "green food" (color of food in English). → `safe and healthy options`
+
+#### c. Misunderstanding the source
+
+> 政府现在倡导文化自信，习主席提出的"讲仁爱，重民本，守诚信，崇正义，尚和合，求大同"。
+>
+> President Xi Jinping has called for a confident embrace of Chinese culture,
+> highlighting virtues such as...
+>
+> *Issue:* Two different subjects: the government advocates cultural confidence;
+> Xi proposed the six virtues. The target merges both into Xi alone.
+
+#### d. Wrong word choice causing misunderstanding
+
+> 提供一些通俗易懂的法宝
+>
+> with easy Buddhist materials
+>
+> *Issue:* 通俗易懂 means "accessible to a general audience," not "easy/simple."
+> The Dharma can be accessible but is never "simple." → `accessible`
+
+### 3. Overtranslation (多翻)
+
+Adding parentheticals or expansions not in the source.
+
+> 出家人不仅要对三宝、师长、道友建立没有染污的情感
+>
+> Monastics should cultivate pure and untainted emotions not only toward the Three
+> Jewels (the Buddha, the Dharma, the Sangha), their teachers (The Head Monk and
+> Abbot), and fellow practitioners...
+>
+> *Issue:* Parenthetical expansions like "(the Buddha, the Dharma, the Sangha)"
+> and "(The Head Monk and Abbot)" are not in the source and should be removed.
+
+### 4. Terminology Errors
+
+- 身心 → `body and heart mind` → should be `body and mind`
+- 甘露别院 → `Ganlu Bieyuan The Amrita Retreat Center` → official: `Amrita Retreat Center`
+- "六字" → `six characters` → should be `six words`
+
+### 5. Detail-Level Issues
+
+- Tone appropriateness
+- Citation source authority
+- CN→EN conversion rules: numerals, units, date formats
+- Proper noun handling: transliteration vs. translation
+
+## Readability Issues
+
+Standard: fluent sentences, graceful expression, matching target reader habits.
+"Conform to what foreign readers find readable."
+"Concise, clear, accurate, not draggy, meaning clear."
+
+### 1. Redundancy & Wordiness
+
+#### a. Long / nested sentences
+
+Sentences with too many clauses that can't be read in one breath. Split them.
+
+#### b. Passive voice overuse
+
+> 如果对肉食消费得少，乃至完全不消费，那么从业者自然随之减少，很多动物就可以摆脱被屠宰、割截的厄运。
+>
+> If meat consumption were reduced, or even eliminated entirely, then the number of
+> businesses involved would naturally be lessened. This way, many animals could be
+> spared from the adversity of being slaughtered and butchered.
+>
+> *Issue:* Three passives in two sentences. Active rewrite is more concise:
+> *If people consume less meat — or none at all — fewer will work in the industry,
+> and many animals will be spared the fate of slaughter and dismemberment.*
+
+#### c. Nominalization (动词名词化)
+
+> 我们前期倡导的正念禅修，比较重视培养觉知。
+>
+> In the early stages of mindfulness practice, we placed greater emphasis on
+> cultivating awareness.
+>
+> *Issue:* "placed greater emphasis on" is nominalized. → `we emphasized`
+
+### 2. Poor Structure (结构不当)
+
+Top-heavy sentences — the main clause arrives too late.
+
+> 没有如法的生活，不能严格要求自己，在今天这个红尘滚滚的时代，我们简直是没希望的。
+>
+> Weathering today's deluge of worldly temptation, without a Dharma-aligned lifestyle
+> and strict self-discipline, we have no hope.
+>
+> *Issue:* The condition clauses pile up before the main point. Restructure:
+> *Without a Dharma-aligned lifestyle and strict self-discipline, we have no hope
+> of weathering today's deluge of worldly temptations.*
+
+### 3. Wrong Word Choice (用词不当)
+
+> 当一个人开始关心生命的大问题时，对小问题自然云淡风轻。
+>
+> when we begin to contemplate the profound questions of existence, the banal
+> iterations of daily life lose their significance.
+>
+> *Issue:* "banal iterations" is obscure. Simplify.
+
+### 4. Weak Flow & Transitions (流畅性 & 连接词)
+
+> 修行的核心就是止恶行善。为什么某种行为能成为生命的主导？就是因为不断重复。
+>
+> The essence of cultivation lies in ceasing unwholesome actions and promoting
+> wholesome ones. But why do certain behaviors dominate our lives? Because of
+> constant repetition.
+>
+> *Issue:* Adding transition words improves flow. The "But" here creates a false
+> contrast — the second sentence is explanation, not counterpoint.
+
+## Detection Checklist
+
+When reviewing, run through:
+- [ ] Any missing content? (compare paragraph-by-paragraph)
+- [ ] Any added content? (parentheticals, expansions not in source)
+- [ ] Any over-literal renderings that don't work in English? (idioms, set phrases)
+- [ ] Any over-free renderings that lose specificity? (concrete examples → abstract)
+- [ ] Subject confusion? (two actors merged into one)
+- [ ] Buddhist terminology checked against terms DB?
+- [ ] Sentences too long to read in one breath? (split at natural breaks)
+- [ ] Passive voice clustering? (3+ in a paragraph)
+- [ ] Nominalized verbs? ("placed emphasis on" → "emphasized")
+- [ ] Top-heavy sentence structure? (main clause buried after long preamble)
+- [ ] Obscure word choices? (would a general reader understand?)
+- [ ] Missing or misleading transition words?

skills/self-review/references/deliberation-protocol.md

@@ -0,0 +1,90 @@
+# Deliberation Protocol (审议规程)
+
+From 东方译场审议手册 (Oriental Translation Review Manual, 2026-03-25).
+Guidance from Ven. Jiqun on translation review culture and process.
+
+## Core Standards
+
+1. **Accuracy (准确性)** — primary. Convey content-level meaning, not word-by-word
+   correspondence. Deepen accuracy iteratively; grasp the "inner spirit" rather than
+   fixating on literal vs free translation.
+
+2. **Readability (可读性)** — secondary. Fluent sentences, graceful expression. Do
+   not pursue endless revision; what is achievable now is the best for now.
+
+**Test**: invite 5–10 readers to read aloud and give feedback — is it clear or
+obscure? Is it concise? Is the meaning clear?
+
+## Decision-Making
+
+When opinions differ:
+- Compare which version is more accurate and more readable
+- Do NOT cling to your own view (不执着于己见)
+- If two people cannot decide, invite 5–10 to read and vote
+- Remember: the translation is for the READERS, not for yourself
+
+## Core Principle: Rejoice First (随喜)
+
+Before flagging issues, affirm what is good. Professional translation companies
+emphasize positive feedback to build translator confidence and good collaboration.
+Criticism alone — however precise — breeds resistance and harms the final result.
+
+In the Oriental Translation Workshop, rejoicing in others' merits cultivates
+sympathetic joy (随喜), compassion, selflessness, and boundless merit — while
+encouraging others' wholesome efforts.
+
+**Practice**: develop eyes that SEE the merits in a translation. Rejoice with
+genuine gladness. Examples:
+- "So many technical terms and complex logic here — you handled it beautifully!"
+- "This word choice is exactly right."
+- "The prose is so clean and fluid!"
+- "Excellent!" / "Good word choice!"
+
+## Three Principles for Giving Suggestions
+
+1. **Lower the self** — use compassionate, gentle, skillful language
+2. **Be diligent and attentive** — find the actual problems
+3. **Tier your suggestions** — make clear what MUST change, what COULD improve,
+   and what needs DISCUSSION
+
+## Three-Tier Issue System
+
+### Level 1: Spelling, Grammar, Format (拼写、语法、格式)
+
+**Action**: fix directly, keep tracked changes visible.
+
+May annotate for translator development:
+- "MPI format uses American English — `toward` not `towards`."
+- "Use `Chan` here, not `Zen`."
+- "MPI format does not use diacritics — `Yogacara` not `Yogācāra`."
+
+### Level 2: Omission, Mistranslation, Wordiness, Tone, Word Choice
+(漏翻、错翻、啰嗦、语气问题、用词问题)
+
+**Action**: preferably flag for translator to fix themselves; may also fix
+directly with tracked changes.
+
+Suggested language:
+- "Did we miss something here?"
+- "This sentence is quite long — could we split it into 2–3?"
+- "The translation here differs from the original. It reads as if the teacher
+  is saying X, but I think he means Y. What do you think?"
+- "This word feels a bit strong here — your thoughts?"
+- (After direct adjustment) "I tightened this sentence — does this work?"
+
+### Level 3: Citation Versions, Slightly Awkward Wording, Marginal Readability
+(引用版本不当、用词稍微不当、阅读体验略差)
+
+**Action**: flag and discuss with translator.
+
+Suggested language:
+- "I found another citation version that may be more concise — which do you prefer?"
+- "Could we use this word instead? Might fit better."
+- "Could you take another look at this passage? I feel readability could improve
+  but I'm not sure how — your thoughts?"
+
+## Tone Guide
+
+- Address the translator as 菩萨 (Bodhisattva) — respectful peer address
+- Use questions, not commands: "Could we...?" "What do you think?" "Would this work?"
+- Frame as collaborative inquiry, not correction

skills/self-review/references/docx-md-extraction.md

@@ -0,0 +1,171 @@
+# Extracting `.docx.md` to `source.dj` + `bilingual.dj`
+
+When the input is a `.docx.md` file (already pandoc markdown, not plain text), the
+extraction differs from the standard DOCX→plain workflow. The markdown preserves
+formatting artifacts that need specific handling.
+
+## CN/EN boundary detection in merged lines
+
+Pandoc markdown often merges CN and EN text on heading lines where the DOCX had
+multiple runs in the same paragraph:
+
+```
+# **三、重视文化教育，重塑人生价值** Prioritize Cultural Education; Reshape Life Values {#三、...}
+```
+
+The boundary regex must account for whitespace on BOTH sides of `**` markers:
+
+```python
+# WRONG: \** then \s* — fails when there's space BEFORE * (值 *Realizing)
+r'[\u4e00-\u9fff](?:\*{0,2})\s*([A-Za-z])'
+
+# WRONG: \s* then \** — fails when there's space AFTER ** (值** Prioritize)
+r'[\u4e00-\u9fff]\s*\**([A-Za-z])'
+
+# CORRECT: whitespace on both sides of optional *
+r'[\u4e00-\u9fff]\s*\**\s*([A-Za-z])'
+```
+
+Include CJK punctuation ranges in the character class:
+`[\u4e00-\u9fff\u3000-\u303f\uff00-\uffef]`
+
+## Anchor stripping
+
+Strip BOTH `{#anchor}` (pandoc heading anchors) AND `[text](#link)` (markdown TOC links)
+before further processing:
+
+```python
+def strip_anchors(s):
+    s = re.sub(r'\{#[^}]*\}', '', s)           # {#anchor}
+    s = re.sub(r'\[([^\]]*)\]\([^)]*\)', r'\1', s)  # [text](#link) → text
+    return s
+```
+
+## TOC handling — two critical guards
+
+### Guard 1: Don't pair TOC CN entries with following EN lines
+
+TOC entries often look like CN-heading-followed-by-EN (the standard pair pattern),
+but the following EN is actually the next TOC entry or heading:
+
+```
+[一、营造禅意氛围，优化工作环境\t1](#anchor)
+[二、重视慈善关爱...]
+...
+1、Create a Chan(Zen) Atmosphere; Optimize Your Environment
+2、Focus on Compassion and Care; Build Good Relationships
+```
+
+If the last CN TOC entry is followed by a blank line then the first EN TOC entry,
+the "CN → blank → EN" heading pattern will consume the first EN TOC entry.
+Guard against this:
+
+```python
+def is_toc_line(line, cn_raw):
+    if re.search(r'\[.*\]\(.*\)', line):  # markdown link
+        return True
+    if re.search(r'\t\d+', cn_raw):       # tab + page number
+        return True
+    return False
+```
+
+Skip the CN→EN and CN→blank→EN patterns when `is_toc_line()` returns True.
+These CN TOC entries should stay unpaired (en='') and get their EN from the
+separate EN TOC list.
+
+### Guard 2: Pair TOC EN entries forward, not backward
+
+EN TOC entries appear as standalone non-CJK lines. They must be paired with
+the FIRST unpaired CN (not the last):
+
+```python
+# CORRECT: forward iteration
+for j in range(len(pairs)):
+    if not pairs[j][1] and has_cjk(pairs[j][0]):
+        pairs[j] = (pairs[j][0], en)
+        break
+
+# WRONG: reverse iteration (pairs last-first, shifting everything)
+for j in range(len(pairs)-1, -1, -1):
+    ...
+```
+
+## Orphaned trailing `*` from italic splits
+
+When a line has italicized EN text (`*Realizing Ultimate Value*`) and the split
+point is at the `R` (after the opening `*` is consumed by the CN cleanup), the
+EN text retains a trailing `*`: `Realizing Ultimate Value*`. Strip it:
+
+```python
+def clean_en(s):
+    s = re.sub(r'^\d+[、,.]\s*', '', s)
+    s = re.sub(r'\*+$', '', s)                # orphaned italic close
+    return s.strip()
+```
+
+## `clean_cn` — operation order matters
+
+Strip leading numbers BEFORE heading markers. A line like `1. # **营造禅意...`
+starts with a digit, so `^#+\s*\**` won't match until the number is gone:
+
+```python
+def clean_cn(s):
+    s = re.sub(r'^\d+\.\s*', '', s)           # leading number FIRST
+    s = re.sub(r'^#+\s*\**', '', s)           # then heading markers
+    s = re.sub(r'\**\s*$', '', s)             # trailing **
+    s = re.sub(r'\t\d+$', '', s)              # trailing page number
+    return s.strip()
+```
+
+## Full extraction recipe
+
+1. Read `.docx.md` text
+2. For each line: strip anchors, detect CJK/EN
+3. CJK lines: try `split_cnen()` first (merged CN+EN). If that gives EN, use it.
+   Otherwise look ahead for EN on next line or next+blank — but SKIP if TOC-like.
+4. EN-only lines: pair forward with first unpaired CN.
+5. Clean: strip heading markers, leading numbers, trailing page numbers.
+6. Write `source.dj` (CN only) and `bilingual.dj` with proper markdown structure.
+
+## bilingual.dj output format
+
+Must follow the project convention (see reference bilingual.dj in any completed article):
+
+```
+# CN Title
+# EN Title
+
+---CN Subtitle
+---EN Subtitle
+
+CN Author
+EN Author
+
+- CN TOC item 1
+- CN TOC item 2
+...
+
+- EN TOC item 1
+- EN TOC item 2
+...
+
+CN body paragraph
+EN body paragraph
+
+CN section heading            ← plain text, no # prefix
+EN section heading
+
+...
+
+CN sub-heading                ← plain text, e.g. "1. 创造精神财富"
+EN sub-heading
+```
+
+- Title: `# ` prefix on both CN and EN
+- Subtitle: `---` prefix (3 dashes, no space after)
+- Author: plain text, one pair
+- TOC: `- ` bullet, CN block then EN block (not interleaved), blank line between blocks
+- Body headings: plain text, no `#` or `##` markers. Hierarchy conveyed by numbering:
+  `一、二、三、` for major sections, `1. 2. 3.` for sub-sections
+- Body paragraphs: interleaved (CN line, EN line, blank)
+- `source.dj` uses `# Title`, `---subtitle`, `## section headings` — different from bilingual.dj which uses plain body headings

skills/self-review/references/edit-suggestions-in-bilingual.md

@@ -0,0 +1,65 @@
+# Edit Suggestions in bilingual.dj
+
+When the user asks to add edit suggestions directly into bilingual.dj, use this
+two-tier approach.
+
+## Two files
+
+- `bilingual.dj` — clean: inline corrections applied, NO `{% %}` comment lines
+- `commented.dj` — same content + `{% ... %}` comment lines interleaved
+
+This lets the user diff them side by side.
+
+## Inline corrections
+
+Apply directly to the English text. These are fixes the reviewer is confident about:
+
+- Typos, mechanical issues (Chinese punctuation in EN, double periods, stray `*`, unbalanced quotes)
+- Grammar fixes (subject-verb agreement, missing articles)
+- Wording improvements (clunky literal translations → idiomatic English)
+
+Apply with `patch` (mode='replace'). Verify uniqueness before
+replacing — many EN paragraphs are long single lines, so match a unique
+substring. Never use regex-based string replacement in `execute_code`.
+
+## `{% %}` comment lines
+
+For translation decisions worth documenting but not "correct" per se:
+
+```
+CN paragraph
+EN paragraph
+{% reason for the choice, alternative renderings %}
+(blank)
+```
+
+The comment goes AFTER the EN line (before the blank separator). One comment per
+issue. Keep them terse.
+
+What to comment on:
+- Literal translation of idioms (`单线程` → "single-threaded")
+- Translation choices that differ from literal meaning (`精神利益` → "nourishing the spirit")
+- Standard Buddhist terminology (`正命` → "Right Livelihood")
+- Glosses added/dropped (`道场` — "(Dojo)" parenthetical removed)
+- Idiom translations (`甘之若饴` → "as sweet as syrup")
+- Paired terms where one rendering influences the other (`魔性` → "demon-nature" to parallel "Buddha-nature")
+- Scripture citations with Sanskrit titles (`普贤行愿品` → "Gaṇḍavyūha Sūtra")
+
+What NOT to comment on:
+- Obvious corrections (typos, grammar)
+- Names, dates, simple connectives
+- Standard renderings with no interesting choice
+
+## Pitfalls
+
+- **Never split a paragraph mid-sentence** with a `{% %}` comment. Comments go
+  AFTER the full EN paragraph, before the blank separator. If a string
+  replacement inserts `\n{% %}` in the middle of a paragraph, it breaks the
+  interleaved structure.
+
+- **Check for merged comments**: after inserting, scan for `{% ... %} ` followed
+  by EN text on the same line. These must be split so the EN text continues on
+  the next line.
+
+- **Collapse triple+ blank lines**: comment insertions can create extra blanks.
+  Run `re.sub(r'\n\n\n+', '\n\n', text)` after all insertions.

skills/self-review/references/proofreading-patterns.md

@@ -0,0 +1,116 @@
+# Proofreading: Manuscript vs Typeset
+
+AGENTS.md defines two workflows: Translation (A) and Proofread (B).
+The workflows below are Proofread mode — English comes from an existing
+manuscript and is authoritative. Only flag mechanical/manuscript-level issues.
+
+## Two extraction workflows
+
+### A. Bilingual from DOCX (standard)
+
+When the DOCX manuscript has both Chinese and English in 1:1 paragraph
+correspondence, generate `bilingual.dj` directly from the DOCX:
+
+1. `pandoc docx → plain text`
+2. Extract Chinese-English pairs from body
+3. Apply fixes: italicize Sanskrit on first occurrence, fix `N.Letter` → `N. Letter` spacing
+4. Write bilingual.dj
+The DOCX English is the authoritative target text. No PDF needed.
+
+**Extraction approach**: start by adapting `scripts/gen-bilingual-docx.py`.
+For articles where the body has strict CN→EN→CN→EN alternation, the simple
+extraction in that script (CN line, blank, EN line, blank) works directly.
+
+### A2. Bilingual from `.docx.md` (pandoc markdown output)
+
+When working with an already-converted `.docx.md` file (pandoc markdown, not plain
+text), use the techniques in `references/docx-md-extraction.md`. Key differences
+from plain-text extraction: merged CN+EN on heading lines, `{#anchor}` and
+`[text](#link)` artifacts, TOC pairing guards.
+
+### B. Bilingual from PDF (when PDF is the typeset target)
+sections that order CN content before EN content (CN heading → CN body → EN heading →
+EN body), the simple alternation fails. Use block-based extraction instead:
+
+1. Tag each non-blank line as CN or EN (via `has_cjk()`)
+2. Join page-break split paragraphs: merge consecutive same-language paragraphs
+   only when the first is long (>30 chars), doesn't end with CJK/ASCII terminal
+   punctuation (`[。！？：）.?!]$`), and isn't heading-like (starts with
+   `^[\dIVX]+[\.\s]` and <60 chars)
+3. Group consecutive same-language items into blocks
+4. Walk blocks: for each CN block, pair with the next EN block via `zip()`.
+   `min(len(cn), len(en))` handles translator-introduced paragraph splits.
+
+**Page-break splits in pandoc plain-text output**: the DOCX→plain conversion
+sometimes splits a Chinese paragraph mid-sentence (e.g. `白居` + `易、苏轼…`).
+These appear as two consecutive CN lines separated by a blank. The join heuristic
+above catches these reliably. For EN text, page-break splits are rare; the heading
+detection (`^[\dIVX]+[\.\s]`, <60 chars) prevents false merges of EN headings
+with following EN body paragraphs.
+The DOCX English is the authoritative target text. No PDF needed.
+
+### B. Bilingual from PDF (when PDF is the typeset target)
+
+When the PDF English is the typeset "final" version and should be the target:
+
+1. Extract DOCX Chinese paragraphs (source)
+2. Extract PDF body text via `pdftotext -layout`
+3. Clean PDF: remove slug lines, headers, page numbers, join hyphenation breaks
+4. Match DOCX English paragraphs against PDF body to find positions
+5. Segment PDF body at matched positions
+6. Write bilingual.dj with Chinese source + PDF English target
+
+**Pitfalls in PDF extraction:**
+- Consecutive hyphenation breaks (e.g. `thou-` + `sand...al-` + `leviate`) — the join
+  loop must be recursive: after joining pair N, check if result still ends with `-`
+  and join with line N+2
+- Lines with leading whitespace: use `lstrip()` before checking `n[0].islower()`
+- Drop-cap artifacts: `L iving` → `Living`
+- Trailing section numbers: `...viewpoints. 1)` — the ` 1)` is a PDF section marker
+  bleeding into the previous paragraph
+
+### C. Edit suggestions (edit-suggestions.dj)
+
+After generating bilingual.dj, scan for issues and write `edit-suggestions.dj`:
+
+**Format**: follow the original document's section/chapter layout. Group suggestions
+under the chapter headings where the issues occur. Use diff-style `-/+` notation.
+
+**What to flag:**
+- Garbled Chinese text (merged duplicate edits in source DOCX)
+- Repeated words (`the The`)
+- Chapter numbering mismatches (e.g. `九` ↔ `VIII`)
+- Translator notes in headings (`（某某翻，某某审）`)
+- Missing quotes around dialogue/speech
+
+## Common source DOCX issues
+
+- Translator notes in Chinese headings: `（某某翻，某某审）` — delete for publication
+- Merged duplicate edits: cut-paste errors where old+new text appear together
+- `N.Letter` without space: `2.How` → `2. How`
+- `the The` double article
+- **Numbering mismatches**: CN and EN headings sometimes disagree (e.g. CN `3．` vs EN `2.`).
+  The TOC usually has the correct number — flag the body heading for correction.
+- **Doubled names**: `岳麓书院岳麓书院` — cut-paste artifacts in Chinese body text.
+- **EN paragraph splits without CN counterpart**: translator sometimes renders one CN
+  paragraph as two EN paragraphs. The block-based extractor drops the extra EN paragraph
+  (as `min(len_cn, len_en)`). Flag in edit-suggestions so it can be manually merged or
+  the CN paragraph can be split.
+
+## Sanskrit italicization
+
+On first occurrence in body text, wrap with `*term*`. Track seen terms across
+the full body. Terms: bodhisattva, bodhicitta, samsara, Dharma, karma, nirvana,
+Sangha, sutra, Mahayana, Sravaka, Vinaya, Lamrim, Ksitigarbha, Samantabhadra,
+Chan, Arhatship, Theravada.
+
+## Proofread scope boundary
+
+When proofreading a DOCX manuscript:
+- **DO flag**: typos, double words, double punctuation, numbering mismatches,
+  garbled text, translator notes, duplicate names, capitalization errors.
+- **Do NOT flag**: em-dash formatting (`—` vs `---`), terminology choices,
+  translation style, calques, word order. The manuscript English is authoritative.
+- **Do NOT apply fixes** — write `edit-suggestions.dj` only.
+- If the user asks for translation review separately, write findings to
+  `translation-findings.dj`.

skills/self-review/references/terms-db-alignment.md

@@ -0,0 +1,64 @@
+# Terms Database Alignment
+
+Batch-align translation glossary entries and body text against the MPI terms database.
+
+## Module API (preferred)
+
+Import directly in `execute_code` scripts — no subprocess, no server, no text parsing:
+
+```python
+import sys
+sys.path.insert(0, '$MPI_PROJECT_ROOT/terms-search')
+from search import search
+
+results = search("三级修学", limit=5)
+results = search("空性", loc="心经", src="DoT定稿", limit=5)
+# returns list of {zh, en, loc, source} dicts
+```
+
+## Batch lookup pattern
+
+```python
+import sys
+sys.path.insert(0, '$MPI_PROJECT_ROOT/terms-search')
+from search import search
+
+terms = ["三无漏学", "八步三禅", "闻思修", ...]
+author_sources = {"DoT定稿", "内部特色词", "佛教术语", "经论名"}
+
+for term in terms:
+    results = search(term, limit=10)
+    relevant = [r for r in results if r["source"] in author_sources]
+    for r in relevant:
+        print(f"{r['zh']} → {r['en']}  [{r['source']}]")
+```
+
+Or filter to a single authoritative source directly:
+```python
+results = search("三级修学", src="DoT定稿", limit=5)
+```
+
+## Priority ranking
+
+When the same term has entries in multiple source tables, prefer:
+1. DoT定稿 (highest authority — final translation decisions)
+2. 内部特色词 (MPI internal terminology)
+3. 佛教术语 (general Buddhist terminology)
+4. 经论名 (sutra/shastra titles)
+
+## Alignment workflow
+
+1. Extract all Chinese glossary terms from `{% "TERM" ... %}` blocks in the .dj file
+2. Extract body-text domain terms that may not have glossary entries
+3. Batch-query each term against the HTTP API
+4. Filter results to authoritative source tables
+5. Compare DB canonical translation against current file translation
+6. Flag mismatches where DB entry differs materially from current
+7. Apply fixes with `patch` tool — fix both glossary comments AND body text occurrences
+8. Verify with `grep` that no old terms remain
+
+## Pitfalls
+
+- `replace_all` can create doubled words when the surrounding context already contains the replacement string (e.g., "The Eight Steps" → "The The Eight Steps"). Prefer targeted single-replacement patches.
+- Start patches from the bottom of the file upward to preserve line numbers.
+- Some DB entries are contextual phrases (e.g., "珍惜法缘" → a full sentence), not standalone term translations. Use standalone term entries where available.

skills/self-review/references/translation-pitfalls.md

@@ -0,0 +1,284 @@
+# Translation Pitfalls — MPI Buddhist Texts
+
+Patterns found in CN→EN translation review. Add to this file as new patterns emerge.
+
+## Terminology conflation
+
+### 人生佛教/人间佛教 — distinct concepts
+
+人生佛教 (Taixu's "Buddhism for Human Life") and 人间佛教 (Yinshun's "Humanistic Buddhism")
+are distinct doctrinal positions in modern Chinese Buddhism. Do not conflate both into
+"Humanistic Buddhism." When the source uses 人生佛教, render as "Buddhism for Human Life"
+or "Human Life Buddhism." User may propose a deliberately non-standard rendering
+("Buddhism for daily lives") — that's their call. Don't unilaterally pick from
+the standard set without checking.
+
+### 心性论 → buddha-nature (WRONG)
+
+心性 (mind-nature) is broader than 佛性 (buddha-nature / tathāgatagarbha).
+When a text discusses 心性 in the context of Confucian self-cultivation or general
+Buddhist psychology, use "mind-nature" or "nature of mind." Reserve "buddha-nature"
+only when the text explicitly references tathāgatagarbha doctrine.
+
+### 恨 → resentment (WRONG)
+
+恨 means "hatred," not "resentment." In the triad 羡慕嫉妒恨 (envy, jealousy, hatred),
+the force is strong. "Resentment" is too mild.
+
+### 感悟 → conversant / heartfelt (WRONG)
+
+感悟 means experiential insight or realization. It is not intellectual familiarity
+("conversant") or emotional warmth ("heartfelt"). Render as "insight," "realization,"
+or "deep understanding."
+
+### 关爱/关怀 → compassion (WRONG)
+
+Chinese 关爱 and 关怀 mean "care" or "loving care." They are NOT 慈悲 (compassion / karuṇā).
+Conflating them obscures two distinct Buddhist concepts.
+
+Check every occurrence of "compassion" in a translation against the source:
+- If source is 关爱/关怀 → "care"
+- If source is 慈悲 → "compassion" (correct)
+- If source is 悬壶济世 → "compassionate mission" (correct — the healing spirit)
+
+### 生存层面 → making a living (WRONG)
+
+生存层面 = the existential/survival dimension. Not just earning wages.
+→ "survival-level needs" or "the level of basic existence"
+
+### 修学 → "practice" (drops study dimension)
+
+修学 combines 修 (practice/cultivation) and 学 (study/learning). Rendering only
+as "practice" loses the study dimension. Use "practice and study" or
+"cultivation and learning" — especially in 静心学堂/Dharma study contexts where
+the academic dimension is emphasized.
+
+### 信仰者 → "believers" (Christian connotation)
+
+信仰者 = "people of faith" but in Buddhist context, "believers" carries Christian
+overtones. Use "practitioners" or "adherents" to avoid the implication. Reserve
+"believers" for contexts where the source explicitly uses 信徒 or where the
+Christian parallel is the point.
+
+## Loss of Dharma meaning
+
+### 生生增上 → continuously elevate our life (INCOMPLETE)
+
+生生 = life after life (multi-life Buddhist perspective). The single-life rendering
+"continuously elevate our life" loses the Dharma meaning entirely.
+→ "continuously elevate our life, life after life"
+
+## Degree / register shifts
+
+### "tantamount to" drops 几乎
+
+几乎 = "almost." "tantamount to" = "is in fact." Dropping 几乎 strengthens the
+claim. Source "几乎等同于一次全球化运动" = "almost equivalent to a globalization
+movement," NOT "tantamount to a globalization movement."
+
+### "untenable" for 都是不行的 (overstates)
+
+都是不行的 = "is not acceptable / won't do." "Untenable" = "indefensible" or
+"cannot be maintained." Too strong. Use "impermissible" or "is not acceptable."
+
+### "monumental event" for 大事 (slight stretch)
+
+大事 = "a major event" or "an event of importance." "Monumental" is typically
+reserved for tasks, errors, or achievements (e.g. "monumental task,"
+"monumental mistake"). "Event of great importance" or "landmark event" is more
+idiomatic.
+
+### "perennial" for 永恒 (wrong register)
+
+永恒 = eternal/ultimate. "Perennial" = recurring (per year, per season). They
+are not synonyms. Use "eternal" or "ultimate."
+
+### 学部委员 → Member (UNDERSTATES)
+
+学部委员 is CASS's highest academic title, equivalent to "Academician."
+"Member" understates the prestige significantly. → "Academician" or
+"Member of the Academic Divisions."
+
+### 文明 → culture (WRONG)
+
+文明 is "civilization," not 文化 "culture." When a text discusses 文明传播
+(civilizational transmission), do not substitute "cultural transmission."
+
+### 教制建设 → reforming (ADDS CONNOTATION)
+
+教制建设 means "developing/building monastic institutions." Adding "reforming"
+introduces a connotation of fixing something broken that is not in the source.
+→ "developing monastic institutions" or "institutional development."
+
+### 一荣俱荣、一损俱损 → too loose
+
+This idiom has a conditional structure: "if one prospers, all prosper; if one
+suffers, all suffer." Rendering as "thrived together and suffered together"
+loses the mutual-dependence logic. → "shared prosperity and adversity alike"
+or "rose and fell together."
+
+### 成圣成贤 collapsing
+
+圣 (sage) and 贤 (worthy) are distinct Confucian categories. Collapsing both
+to "a sage" loses the distinction. → "sagehood and worthiness" or "a sage or worthy."
+
+### "must" overuse from 应当/倡导
+
+Source patterns 应当 (should), 倡导 (advocate/champion), 今后要 (going forward, should)
+often get rendered as "must" by reflex. "must" in English is a strong directive
+appropriate only for 一定要, 必须, 务必. Default to "should" or "ought to" for
+recommendations. A whole section may be 倡导 without 一定 anywhere — don't
+manufacture urgency the source doesn't have.
+
+### 文明 → culture (WRONG)
+
+文明 is "civilization," not 文化 "culture." When a text discusses 文明传播
+(civilizational transmission), do not substitute "cultural transmission."
+
+#### 因病返贫 → "back into poverty"
+"返贫" means becoming poor due to illness, not returning to previous poverty. Use "into poverty" or "driven into poverty."
+
+#### Diacritics: use DB form, not academic Sanskrit
+| Wrong | Right | Source |
+|---|---|---|
+| `Mahāsthāmaprāpta` | `Mahasthamaprapta` | 佛教术语 |
+| `Yogācārabhūmi Śāstra` | `Yogacarabhumi-Sastra` | 经论名 |
+| `Avalokiteśvara` | `Guanyin` | 佛教术语 |
+| `pravāraṇā` | `Pavarana` | BAICKZ |
+
+Exception: `Kṣitigarbha` — DoT定稿 uses diacritics, so keep them.
+When in doubt, search the DB and follow the highest-priority source. See dharma-translation skill `references/diacritics-convention.md`.
+
+返贫 = become poor (from a non-poor state) due to medical costs. "Back" implies
+the person was previously poor — not necessarily true. This is about medical bankruptcy.
+→ "into poverty" or "fall into poverty" (no "back")
+
+## Subject-shift calques
+
+When the source uses an abstract/system noun as the grammatical subject (佛教,
+文明, 宗教, 文化) and the target reflexively substitutes a more concrete
+agent (Buddhist practitioners, civilization-builders, religious people, etc.),
+the English is wrong: the source is *not* talking about people, it's talking
+about the system.
+
+**Pattern**: 佛教都被社会大众赋予期望 → EN drifts to "Buddhist practitioners
+are looked upon with hope." Wrong subject — source is 佛教, not 佛教徒. Right:
+"Buddhism is regarded with hope by the broader society."
+
+**Detection**: For each translated sentence, find the *grammatical subject*
+in the English and check it matches the *grammatical subject* in the Chinese.
+If the English subject is a concrete agent and the Chinese is an abstract
+system noun, it's a subject-shift calque.
+
+## Factual inconsistencies across paired descriptions
+
+When the same person, place, or thing is described in two different paragraphs
+(intro + later reference), check that *every* descriptor matches: title,
+affiliation, role, credentials. A translation can be factually inconsistent
+even when each individual sentence is correct in isolation.
+
+**Pattern**: Prof. Wei's credentials in paragraph 1: "Academician of the
+Chinese Academy of Social Sciences (CASS) and research fellow at the
+Institute of World Religions." Paragraph 5 (same person, same source
+reference): "a member of the CASS academic committee and Director of the CASS
+Buddhist Research Center." Two different titles for the same CASS affiliation.
+One is right, the other is wrong.
+
+**Detection**: For each named person, build a dict {name → {credential:
+sentence_refs}} and check that all credentials in the dict match. Same
+institution or title can have different renderings in different paragraphs.
+
+## Calque checklist (subtle English calques of Chinese verbs)
+
+These are common Chinese-verb → English-verb pairs where the English word
+sounds natural in isolation but is a direct calque of the Chinese. They're
+easy to miss in a first-pass review because each one parses correctly:
+
+| Chinese verb | Wrong (calque) | Right (idiomatic) |
+|---|---|---|
+| 赋予 (entrust with) | "look to with hope" | "regard with hope" |
+| 得到 (obtain) | "draw forth" | "draw on" / "gain" |
+| 发挥 (bring into play) | "bring into full play" | "make the most of" |
+| 承担 (assume) | "shoulder" | "take on" |
+| 重视 | "attach importance to" | "value" / "emphasize" |
+| 体现 | "embody" / "reflect" | "show" / "demonstrate" (when abstract) |
+
+**Detection**: When the English uses an unusual verb that maps 1:1 to a
+Chinese word, and the Chinese word is a high-frequency academic verb (发挥,
+承担, 体现, 重视), check whether the English reads as a calque. A common
+smell: the English verb is "correct" but more formal/dramatic than the
+surrounding prose.
+
+## Tonal coherence inside a parallel list
+
+When a list of items in a section should have parallel structure (e.g. three
+"champion X, oppose Y" items; six "developing X, strengthening Y" items),
+check that the *verb choice* is consistent across the list. Inconsistency
+within a parallel structure is a strong signal of drift.
+
+**Pattern**: Section V lists "First, Buddhism should serve as... Second, it
+must serve as... Third, it must serve as..." Source has 应当/要做 for all
+three. The English should match: all three "should serve as" or all three
+"must serve as." Mixing is a tell.
+
+**Detection**: For each parallel-list section (First/Second/Third, etc.),
+extract the verb (or other repeated slot) and verify it's identical. Drift
+inside a parallel list is one of the easiest flow issues to catch
+mechanically — just look for variance.
+
+## Multi-pass review structure
+
+Translation review benefits from three distinct passes, run separately, each
+catching a different category of error:
+
+1. **Pass 1: terminology + consistency + line count** — fast, mechanical.
+   Catches: 人生佛教/人间佛教 conflation, 修行/修学, 恨→resentment,
+   paired inconsistencies (Buddhism vs Buddhist practitioners), missing
+   content, wrong numerals.
+
+2. **Pass 2: mechanical/formatting** — em-dash convention, double punctuation,
+   numbering mismatches, capitalisation typos, garbled text. Often skipped
+   if the file "looks clean." This pass is what makes the file safe to
+   publish; do it even when no content issues are obvious.
+
+3. **Pass 3: flow/tonal/calques** — *read the full English as a piece of
+   prose*. Catch: dramatic verbs that read as calques ("draw forth,"
+   "into full play," "shoulder"), intensifier drift ("profoundly important"
+   × 3 in one section), consistency of "must"/"should" inside parallel
+   lists, subject misattribution in calque. Often the user will prompt
+   this pass with "are you sure it reads well?" or "do the words hang
+   together?" Treat that prompt as a signal to re-read the whole English
+   target, not just spot-check.
+
+Pass 3 in particular is the one that catches the *most embarrassing* errors
+— the ones where the English is grammatical and faithful but reads as
+"translationese." Don't skip it.
+
+## DoT定稿 term drift
+
+### 念死 → recollection of death (WRONG per DoT定稿)
+
+DoT定稿 has "Cultivating mindfulness of death" / 佛教术语 has "contemplating the
+impermanence of death". The established term is "mindfulness of death", not
+"recollection of death." → "mindfulness of death" / "death-mindfulness"
+
+### 三级修学 → Three-Level Study Program (WRONG per DoT定稿)
+
+DoT定稿 has "Three-Stage Practice." → "Three-Stage Practice"
+
+### 下士道/中士道/上士道
+
+DoT定稿: "Path for Persons of Small/Medium/Great Capacity" — not "path of the
+initial/middle/great scope."
+
+### 观音菩萨 → Avalokiteśvara (AVOID in MPI translations)
+
+佛教术语 has "Guanshiyin/Guanyin Bodhisattva." Use "Guanyin Bodhisattva."
+
+## Workflow pitfall
+
+### Translating before consulting terms DB
+
+Always search key terms BEFORE translating. The dharma-translation skill says to do
+this, but it's easy to skip. Use the CLI: `$MPI_PROJECT_ROOT/terms-search/search.py <query>`.
+Prioritize DoT定稿 > 内部特色词 > 佛教术语 > 经论名.

skills/self-review/scripts/review_csv.py

@@ -0,0 +1,78 @@
+#!/usr/bin/env python3
+"""Reference: reusable detection rules for Chinese→English translation review.
+Adapt the rules list for each project's domain vocabulary."""
+
+import csv, re, sys
+
+def review_csv(csv_path):
+    rows = []
+    with open(csv_path) as f:
+        reader = csv.DictReader(f)
+        for r in reader:
+            rows.append((r.get('page', '') or '', r['Chinese'] or '', r['English'] or ''))
+
+    issues = []
+
+    def add(row_idx, cn, en, problem, suggestion):
+        issues.append({
+            'row': row_idx + 2,
+            'page': rows[row_idx][0],
+            'cn': cn, 'en': en,
+            'problem': problem,
+            'suggestion': suggestion
+        })
+
+    for i, (page, cn, en) in enumerate(rows):
+        if not cn or not en:
+            continue
+
+        # ── Add project-specific detection rules below ──
+
+        # Example: "Is we" → machine translation artifact
+        if re.search(r'\bIs we\b', en):
+            add(i, cn, en,
+                "'Is we' — literal MT of 是否/如果. Should be 'if we' or 'whether we'",
+                re.sub(r'\bIs we\b', 'if we', en))
+
+        # Example: Chinese punctuation in English text
+        if re.search(r'[，。；：！？、]', en):
+            add(i, cn, en,
+                "Chinese punctuation in English text",
+                "[Replace with English punctuation]")
+
+        # Example: unbalanced HTML tags
+        if en.count('<b>') != en.count('</b>'):
+            add(i, cn, en,
+                f"Unbalanced <b> tags (open={en.count('<b>')}, close={en.count('</b>')})",
+                "[Balance tags]")
+
+        # Example: unbalanced double quotes
+        if en.count('"') % 2 != 0:
+            add(i, cn, en,
+                f"Unbalanced quotes ({en.count(chr(34))} total)",
+                "[Balance quotation marks]")
+
+        # Example: term inconsistency check
+        # if re.search(r'TermA', en) and re.search(r'TermB', en) and ...
+
+    # ── Sanity checks ──
+    for i, (page, cn, en) in enumerate(rows):
+        if cn and not en:
+            print(f"WARNING row {i+2}: CN present but EN empty: {cn[:80]}")
+        cn_chars = re.findall(r'[\u4e00-\u9fff]', en)
+        if cn_chars:
+            print(f"WARNING row {i+2}: Chinese chars in EN: {cn_chars}")
+        if cn and en and cn.strip() == en.strip():
+            print(f"WARNING row {i+2}: CN==EN (untranslated): {cn[:60]}")
+
+    return issues
+
+if __name__ == '__main__':
+    issues = review_csv(sys.argv[1])
+    print(f"Issues found: {len(issues)}")
+    for iss in issues:
+        print(f"\nCSV_ROW_{iss['row']} [{iss['page']}]")
+        print(f"  CN: {iss['cn'][:120]}")
+        print(f"  EN: {iss['en'][:120]}")
+        print(f"  PROBLEM: {iss['problem']}")
+        print(f"  SUGGEST: {iss['suggestion'][:150]}")

skills/self-review/scripts/sweep.py

@@ -0,0 +1,139 @@
+#!/usr/bin/env python3
+"""Mechanical sweep for .dj translation review — run after patches or as final verification.
+
+Usage: python3 sweep.py <source.dj> <target.dj> [--stale term1,term2] [--new term1,term2]
+
+Checks:
+  1. Non-empty line count parity (source == target)
+  2. Heading count parity
+  3. Zero Markdown bold (**) in target (djot uses single *)
+  5. Zero common Chinese punctuation in target
+  6. Zero [text](#anchor) link artifacts in target TOC area (first 15 lines)
+  7. Zero unbalanced double-quotes in target
+  8. --stale: each listed string must appear ZERO times in target
+  9. --new: each listed string must appear at least once in target
+"""
+
+import re
+import sys
+
+
+CN_PUNCT = re.compile(r'[\u3000-\u303f\uff00-\uffef\u201c\u201d\u2018\u2019]')
+
+def read_nonempty(path):
+    with open(path) as f:
+        return [l for l in f.read().rstrip('\n').split('\n') if l.strip()]
+
+
+def main():
+    if len(sys.argv) < 3:
+        print("Usage: sweep.py <source.dj> <target.dj> [--stale a,b,c] [--new x,y,z]")
+        sys.exit(2)
+
+    src_path = sys.argv[1]
+    tgt_path = sys.argv[2]
+
+    stale_terms = []
+    new_terms = []
+    i = 3
+    while i < len(sys.argv):
+        if sys.argv[i] == '--stale' and i + 1 < len(sys.argv):
+            stale_terms = [t.strip() for t in sys.argv[i+1].split(',') if t.strip()]
+            i += 2
+        elif sys.argv[i] == '--new' and i + 1 < len(sys.argv):
+            new_terms = [t.strip() for t in sys.argv[i+1].split(',') if t.strip()]
+            i += 2
+        else:
+            i += 1
+
+    src_lines = read_nonempty(src_path)
+    tgt_lines = read_nonempty(tgt_path)
+    tgt_raw = open(tgt_path).read()
+
+    errors = 0
+
+    # 1. Line count
+    if len(src_lines) != len(tgt_lines):
+        print(f"[FAIL] Line count: src={len(src_lines)} tgt={len(tgt_lines)}")
+        errors += 1
+    else:
+        print(f"[OK] Line count: {len(src_lines)}")
+
+    # 2. Heading count
+    src_h = sum(1 for l in src_lines if l.startswith('## '))
+    tgt_h = sum(1 for l in tgt_lines if l.startswith('## '))
+    if src_h != tgt_h:
+        print(f"[FAIL] Headings: src={src_h} tgt={tgt_h}")
+        errors += 1
+    else:
+        print(f"[OK] Headings: {src_h}")
+
+    # 3. Unicode em/en-dash
+    em = tgt_raw.count('\u2014')
+    en = tgt_raw.count('\u2013')
+    if em or en:
+        print(f"[FAIL] Unicode dashes: em-dash={em} en-dash={en}")
+        errors += 1
+    else:
+        print("[OK] No Unicode em/en-dashes")
+
+    # 4. Markdown bold
+    bold = sum(1 for l in tgt_lines if '**' in l)
+    if bold:
+        print(f"[FAIL] Markdown bold (**): {bold} lines")
+        errors += 1
+    else:
+        print("[OK] No Markdown bold")
+
+    # 5. Chinese punctuation
+    cn = [(i+1, l[:60]) for i, l in enumerate(tgt_lines) if CN_PUNCT.search(l)]
+    if cn:
+        print(f"[FAIL] Chinese/smart punct: {len(cn)} lines")
+        for ln, snippet in cn[:5]:
+            print(f"  L{ln}: {snippet}")
+        errors += 1
+    else:
+        print("[OK] No Chinese punctuation")
+
+    # 6. TOC link artifacts (first 15 lines)
+    toc_links = sum(1 for l in tgt_lines[:15] if re.search(r'\[.*?\]\(#', l))
+    if toc_links:
+        print(f"[FAIL] TOC has [text](#anchor) links: {toc_links}")
+        errors += 1
+    else:
+        print("[OK] TOC clean (no link artifacts)")
+
+    # 7. Unbalanced quotes
+    for i, l in enumerate(tgt_lines):
+        if l.count('"') % 2 != 0:
+            print(f"[FAIL] L{i+1}: Unbalanced quotes: {l[:80]}")
+            errors += 1
+    if errors == sum(1 for l in tgt_lines if l.count('"') % 2 != 0):
+        pass  # errors already counted above
+    elif not any(l.count('"') % 2 != 0 for l in tgt_lines):
+        print("[OK] No unbalanced quotes")
+
+    # 8. Stale terms (must be absent)
+    for term in stale_terms:
+        count = tgt_raw.count(term)
+        if count > 0:
+            print(f"[FAIL] Stale term '{term}' still present: {count}")
+            errors += 1
+        else:
+            print(f"[OK] Stale term '{term}' absent")
+
+    # 9. New terms (must be present)
+    for term in new_terms:
+        count = tgt_raw.count(term)
+        if count == 0:
+            print(f"[FAIL] New term '{term}' not found")
+            errors += 1
+        else:
+            print(f"[OK] New term '{term}' found: {count}")
+
+    print(f"\n{'ALL CLEAN' if errors == 0 else f'{errors} ISSUE(S) FOUND'}")
+    sys.exit(0 if errors == 0 else 1)
+
+
+if __name__ == '__main__':
+    main()