v4.3.5 — Prompt optimization, pipeline improvements, config tuning | v4.3.5 Prompt 优化、流水线改进、配置调整

What's New

🔧 Prompt & Pipeline Optimization

Redesigned the chapter writing workflow to eliminate common quality issues at the source. Task 1 oracle now verifies outline completeness before proceeding. Task 3 chapter agent follows a structured subsection model (claim → evidence → analysis → counterpoint) and avoids redundant "In summary" filler. Per-section depth is now enforced naturally via tighter prompt constraints rather than post-hoc review.

🐛 Serial Progress Guard

Added a checkpoint file (serial_progress.txt) to the macOS serial mode workflow. When the main agent resumes after user interruption, it reads the file to determine the correct chapter to continue from — preventing accidental parallel dispatch or restart.

📏 Depth Balance Check

New depth-balance script command detects chapters with significantly fewer lines than average. Integrated into the QA pipeline as an informational check — flags thin chapters without blocking assembly.

📐 Character Count Alignment

Updated profiles.json with realistic character targets across all three modes (quick 15K / standard 25K / deep 45K). Chapter agents now write once without iterative trimming — a major source of latency. The stripped-char metric is maintained as the universal cross-language unit.

🧹 Word Count Display Fix

Fixed a bug where the final summary table displayed an incorrect word count value due to the agent extracting the wrong JSON field. Explicitly routed to checks.word_count.count from the qa-report JSON output.

🇨🇳 中文说明

更新内容

🔧 Prompt 与流水线优化

从源头消除常见的报告质量问题。Task 1 oracle 现先生成大纲后自动检查覆盖面完整性。Task 3 章节 agent 遵循结构化子节模型（结论→数据→因果→反方），不再写"综上所述"式重述。各节深度通过更精准的 prompt 约束保障，而非事后 Review。

🐛 串行进度文件保护

macOS 串行模式新增 checkpoints 文件。主 agent 中断恢复后读取进度文件确定从哪章继续，消除意外并行派发或重头开始的问题。

📏 章节深度均衡检查

新增 depth-balance 脚本命令，自动检测行数显著低于平均值的章节。集成到 QA 流水线中，告警但不阻断。

📐 字数目标合理化

更新 profiles.json 三档字数目标（quick 15K / standard 25K / deep 45K）。章节 agent 一次性写完，不再反复修剪——大幅降低耗时。去空格字符数作为全局统一标尺保持不变。

🧹 字数显示 bug 修复

修复最终汇报表中字数显示错误的问题——agent 原先误读了 JSON 的顶层字段。现已改为从 checks.word_count.count 读取。

v4.3.0 — Parallel chapter dispatch, report quality fixes, new reports | v4.3.0 Task 3 并行派发、报告质量修复、新增报告

What's New

🔧 Parallel Chapter Dispatch

Task 3 now supports parallel chapter writing on non-macOS platforms using OpenCode's native task(run_in_background=true). macOS automatically falls back to serial mode via platform detection. Task IDs are persisted to a JSON file for recovery after interruption.

🔍 Dual-Engine Search with Quality Triggered Fallback

SearXNG and Exa are now detected in parallel — both search simultaneously when available. Results are merged and deduplicated. When quality is insufficient (URL count, freshness, source diversity), Layer 3 free sources (DuckDuckGo, Bing, domestic Chinese sources) are automatically triggered as a fallback.

📊 Final Report Table

The completion summary now displays as a structured table with lang-adaptive labels. Includes quick-insights row from outline chapter 1, freshness data, and a search strategy description. The data_limited badge is inlined into the Data row.

📁 Report Management

Promo footer added to all 59 existing reports in one batch update. Showcase table in README lists featured reports with clickable titles. /research-update command now supports direct git pull and displays project Star History.

🐛 Quality Fixes

Fixed language contamination where non-English characters leaked into report metadata headers. Repaired TOC anchors in non-zh reports to match GitHub's auto-generated heading IDs. Restored missing promo footer on GenAI Enterprise report.

🇨🇳 中文说明

更新内容

🔧 Task 3 并行派发

支持非 macOS 平台并行撰写各章，使用 OpenCode 原生 task(run_in_background=true)。macOS 通过平台检测自动切换串行模式。task-id 持久化到 JSON 文件，中断后可恢复。

🔍 双引擎搜索 + 质量触发补强

SearXNG 和 Exa 并行检测，同时可用时同时搜索，结果合并去重。搜索结果质量不足时（URL 数量、时效性、来源多样性）自动触发 Layer 3 免费源补强。

📊 最终汇报表格

汇报总结改为结构化表格，标签按语言自适应。包含第 1 章观点速览行、数据新鲜度、搜索策略描述。data_limited 徽标并入 Data 行。

📁 报告管理

一次性为全部 59 份报告添加推广尾链。README 新增展示表陈列精选报告。/research-update 命令支持 git pull 直拉，显示项目 Star History。

🐛 质量修复

修复非英文字符泄漏到报告元数据的问题。修复非中文报告的 TOC 锚点与 GitHub 自动生成 ID 不匹配的问题。恢复 GenAI Enterprise 报告缺失的推广尾链。

v4.2.2 — platform-aware chapter dispatch (macOS serial workaround)

v4.2.2

What's Changed

• Platform detection: Task 3 now runs uname -s before dispatching chapters
  • macOS (Darwin): serial dispatch — task(run_in_background=false) in a loop, one chapter at a time. No dependency on oh-my-openagent's background-task notification system, which has known bugs on macOS (issues #4721, #4874, #5089 in code-yeongyu/oh-my-openagent)
  • Other platforms (Windows, Linux, Codex CLI): parallel dispatch — task(run_in_background=true) as before, keeps full speed
• New report: added 1 new Chinese research report (AI programming tools trends 2026)

更新内容

• 平台自适应：Task 3 现在执行 uname -s 检测操作系统后再派发章节
  • macOS（Darwin）：串行模式——循环中依次调用 task(run_in_background=false)，每章等写完再写下一章。不依赖 oh-my-openagent 的后台任务通知机制，绕过 macOS 上已知的 omo bug（#4721、#4874、#5089）
  • 其他平台（Windows、Linux、Codex CLI）：并行模式——task(run_in_background=true) 不变，保持最快速度
• 新报告：新增 1 篇中文调研报告（AI 编程助手工具发展趋势 2026）

v4.2.1 — update command, star history, minor fix

v4.2.1

What's Changed

• update command: now directly pulls latest code from main branch — no version check, no confirmation required
• README: added Star History chart at the bottom
• Reports: added 2 new Chinese research reports (toy industry, Douyin personal branding)
• AGENTS.md: added bilingual commit message requirement (English first, Chinese second)

Known Issue — macOS (oh-my-openagent only)

On macOS, due to a known bug in oh-my-openagent (issues #4721, #4874, #5089), after all chapters are written in parallel (Task 3), the main agent may not automatically proceed to assembly. Workaround: type 继续 or continue once to resume.

This only affects oh-my-openagent on macOS. Other platforms (Codex CLI, Windows omo) are not affected.

更新内容

• update 命令：改为直接从 main 分支拉取最新代码，不再比对版本号，无需用户确认
• README：底部添加 Star History 星标轨迹图
• 新报告：新增 2 篇中文调研报告（玩具行业、抖音个人自媒体生态）
• AGENTS.md：新增 git commit 双语提交要求（英文在前，中文在后）

已知问题 — macOS（仅 oh-my-openagent）

在 macOS 上，由于 oh-my-openagent 的已知 bug（#4721、#4874、#5089），章节并行撰写完毕后，主 agent 可能不会自动进入装配环节。临时解决方案：回复 继续 或 continue 即可恢复。

此问题仅影响 macOS 版 oh-my-openagent。其他平台（Codex CLI、Windows omo）不受影响。

v4.2.0

v4.2.0

What's Changed

Fix: Task 3 switched from parallel background dispatch to serial synchronous chapter writing to bypass OpenCode v1.16 background notification regression. Chapters are now written one at a time (run_in_background=false) — slightly slower but 100% reliable.

New Reports (all generated in this session):

• 8 new English reports: AI Chip, Cross-border E-commerce, EV Battery, GenAI Adoption, Data Center Energy, Autonomous Vehicles, Quantum Computing, Edge Computing
• 10 new non-English reports: Hindi/Indonesian/Italian/Dutch/Polish/Swedish/Thai/Turkish/Vietnamese — covering agriculture, digital economy, luxury fashion, semiconductors, IT outsourcing, biotech, tourism, energy, electronics

Other:

• Data-pool.json now uses Python json.dump (skip write tool size limit)
• Offline mode (Step 0.5) merged into main workflow
• Language output guard strengthened
• Documentation updated (parallel → serial references)

---

v4.2.0

改动内容

修复: Task 3 从并行后台派发改为串行同步撰写章节，绕过 OpenCode v1.16 后台通知回归问题。逐章同步写入（run_in_background=false）——稍慢但 100% 可靠。

新增报告（本次会话生成）：

• 8 篇英文报告：AI 芯片、跨境电商、EV 电池、GenAI 采用、数据中心能耗、自动驾驶、量子计算、边缘计算
• 10 篇多语言报告：印地语/印尼语/意大利语/荷兰语/波兰语/瑞典语/泰语/土耳其语/越南语——覆盖农业、数字经济、奢侈品、半导体、IT外包、生物科技、旅游、能源、电子

其他优化：

• data-pool.json 改用 Python json.dump 写入（绕过 write 工具大小限制）
• 离线模式（Step 0.5）合并到主流程
• 语言输出加固
• 文档更新（并行→串行表述）

v4.1.0 — Offline Mode for Local File Research

v4.1.0 - Offline Mode for Local File Research

What's New

🔥 Offline Mode

• The skill can now work with local files only — no internet connection required
• Automatically detects user intent from the first prompt (mentions of local files, "don't search", "offline mode") via LLM semantic understanding
• Supports .md, .txt, .pdf, .docx file formats
• Auto-installs PyPDF2 for PDF text extraction when model cannot read PDF natively
• Auto-installs python-docx for DOCX parsing

How it works
When a user includes local file paths and indicates no online search:

1. Detects offline intent via LLM semantic understanding (Step 0.5)
2. Skips all web search, scraping, and data collection (Task 2 offline branch)
3. Reads local files directly via read/glob tools
4. Builds the same structured data-pool.json as online mode
5. Reuses existing Task 3 (chapter writing) and Task 4 (assembly/QA) unchanged

Bug Fixes

• Fix duplicate report files with same title in reports/ directory
• Fix README cost estimates with DeepSeek pricing source URL
• Fix English README unit error in cost section

---

🇨🇳 中文说明

v4.1.0 - 本地文件离线调研模式

新增功能

🔥 离线模式

• Skill 现支持纯本地文件调研，无需联网
• 自动通过 LLM 语义理解识别用户意图（提到本地文件、要求不联网等），非关键词匹配
• 支持 .md、.txt、.pdf、.docx 四种文件格式
• AI 自动安装 PyPDF2 提取 PDF 文本（无需用户操作）
• AI 自动安装 python-docx 解析 DOCX 文件

工作原理
用户输入本地文件路径并指示不联网后，系统：

1. 识别离线意图（Step 0.5，LLM 语义理解）
2. 跳过所有搜索、抓取流程（Task 2 离线分支）
3. 读取本地文件
4. 构建标准 data-pool.json（与在线模式格式完全一致）
5. 复用现有的章节撰写（Task 3）和装配质检（Task 4），零改动

Bug 修复

• 修复 reports/ 目录同标题报告重复问题
• 修复 README 成本估算数据，补充 DeepSeek 定价来源 URL
• 修复英文版 README 成本单位错误

---

Full Changelog: v4.0.0...v4.1.0

v4.0.0 — Multilingual pipeline, language-aware search, reports reorganization

v4.0.0 — Multilingual pipeline, language-aware search, reports reorganization

Changes from v3.0.2

🌍 Fully multilingual final report

• All summary labels (outline/data/report/chapters/sources/facts/lines/chars/min) now dynamically translate to $LANG — zh→中文, en→English, fr→Français, ja→日本語, ru→Русский, etc.
• Chapter list heading also translated. Previously only zh/en were supported; now all 19 languages.

🔍 Language-aware search source filtering

• Non-Chinese research ($LANG != "zh") now skips Chinese-only search engines (cn.bing.com, sogou, 360) and all B-class Chinese sources (zhihu, 36kr, CSDN, etc.) to eliminate irrelevant results.
• Generic search engines now get locale parameters: Brave &country={COUNTRY}, Mojeek &lang={LANG}.
• Regional engines added: Yandex for Russian (ru), Yahoo JP for Japanese (ja).
• LANG→COUNTRY mapping table added to SKILL.md for Task 2 variable replacement.

📁 Reports organized by language

• Reports now saved to reports/$LANG/ subdirectories (e.g., reports/zh/, reports/en/, reports/fr/).
• Existing 38 reports classified and moved into their respective language directories.

🔧 Windows compatibility improvements

• Filename sanitization (dr_gen.py): Windows-invalid characters (<>:"/\|?*) replaced with -; trailing dots/spaces trimmed.
• Zero-byte file cleanup: Task 4 deletes stale 0-byte stubs before assembly to prevent silent failures.
• os.makedirs(dirname(output), exist_ok=True) added as safety net in dr_gen.py.
• .gitignore updated: tmp/, language.txt, start_time.txt now ignored.

🔄 Pipeline restructuring

• Setup phase extracted: TMPDIR creation, TOOLSDIR/PROMPTSDIR detection, and file reading now happen before Step 0 (language detection). Previously Step 0 referenced {TMPDIR} before it was created, causing language.txt to be written to the wrong location on first run.
• Language detection now announces result: 🌐 Language detected: en after completion.
• "禁止" rule updated: clarifies handoff file reads (outline.json, manifest.json) are allowed between tasks; only search calls and data processing must stay within sub-agents.

✅ QA improvements

• TOC heading whitelist expanded in dr_check.py: now includes all 19 language variants (目次/목차/Índice/Table des matières/Inhaltsverzeichnis/etc.) — previously only had English/Chinese/German.
• Final report template requires all labels to be in $LANG (no more Chinese labels appearing in French research output).

Files changed

• SKILL.md — Setup phase, search source filtering table, reports/$LANG/, final report multilingual template, updated variable mappings
• prompts/task2_data_collection.md — Search source language filtering, regional engines, LANG/COUNTRY variables
• prompts/task4_assembly.md — Output path changed to reports/{LANG}/
• tools/dr_gen.py — Filename sanitization, os.makedirs safety net
• tools/dr_check.py — TOC heading whitelist expanded to 19 languages
• .gitignore — New ignores for tmp/ and temp files
• VERSION — 3.0.2 → 4.0.0
• reports/ — Existing 38 files reorganized by language subdirectory

---

v4.0.0 — 全链路多语言、搜索源按语言过滤、报告按语言分类

相对于 v3.0.2 的变更

🌍 最终汇报完全多语言化

• 所有摘要标签（大纲/数据/报告/章/来源/事实/行/字/分钟）根据 $LANG 动态翻译——zh→中文、en→English、fr→Français、ja→日本語、ru→Русский……
• 章节列表标题同步翻译。此前仅支 zh/en 两种，现覆盖全部 19 种语言。

🔍 搜索源按语言过滤

• 非中文调研（$LANG != "zh"）跳过中文专用搜索引擎（cn.bing.com、搜狗、360）和 B 类中文源（知乎、36氪、CSDN 等），避免噪音结果。
• 通用搜索引擎加 locale 参数：Brave &country={COUNTRY}、Mojeek &lang={LANG}。
• 新增区域引擎：俄语用 Yandex，日语用 Yahoo JP。
• 在 SKILL.md 中添加 LANG→COUNTRY 映射表用于 Task 2 变量替换。

📁 报告按语言分类

• 报告保存到 reports/$LANG/ 子目录（如 reports/zh/、reports/en/、results/fr/）。
• 38 份现有报告已分类移入对应语言目录。

🔧 Windows 兼容性改进

• 文件名净化（dr_gen.py）：Windows 非法字符（<>:"/\|?*）替换为 -；尾部句点和空格去除。
• 零字节残留清理：Task 4 在装配前删除所有 0 字节文件，防止静默失败。
• dr_gen.py 写文件前加 os.makedirs(dirname(output), exist_ok=True) 兜底。
• .gitignore 更新：新增 tmp/、language.txt、start_time.txt。

🔄 流程重构

• 分离出 Setup 阶段：TMPDIR 创建、TOOLSDIR/PROMPTSDIR 确定、文件读取，现在都在 Step 0 语言判定之前完成。此前 Step 0 引用 {TMPDIR} 时目录还未创建，导致 language.txt 第一次被写到错误位置。
• 语言判定后向用户公告结果：🌐 Language detected: en。
• 更新"禁止"规则：明确 Task 间 handoff 文件读取（outline.json、manifest.json）不受限；只有搜索引擎调用和数据处理必须在子 agent 内完成。

✅ QA 改进

• dr_check.py 的 TOC 标题白名单扩展到 19 种语言（目次/목차/Índice/Table des matières/Inhaltsverzeichnis 等），此前只有英文/中文/德语三项。
• 最终汇报模板强制全部标签按 $LANG 翻译（不再出现法语调研结果显示中文标签的问题）。

变更文件

• SKILL.md — Setup 阶段、搜索源过滤表、reports/$LANG/、最终汇报多语言模板、变量映射更新
• prompts/task2_data_collection.md — 搜索源语言过滤、区域引擎、LANG/COUNTRY 变量
• prompts/task4_assembly.md — 输出路径改为 reports/{LANG}/
• tools/dr_gen.py — 文件名净化、os.makedirs 兜底
• tools/dr_check.py — TOC 标题白名单扩展到 19 种语言
• .gitignore — 新增 tmp/ 和临时文件忽略
• VERSION — 3.0.2 → 4.0.0
• reports/ — 38 份现有报告按语言子目录重组

v3.0.2

v3.0.2 — Encoding fixes & faster language detection

Changes from v3.0.1

🔧 Encoding (Windows CP936 pipe fix)

• oracle no longer returns JSON via response body (which goes through shell pipe and corrupts non-ASCII text on Windows). Instead, oracle outputs JSON in its response → main agent uses write tool to create the file, bypassing the CP936 encoding issue entirely.
• This affects Task 1 (outline generation). Task 2/3/4 already used file-based I/O and were not affected.

⚡ Language detection speedup

• Replaced detect_lang.py (213-line Python script) with LLM-based language detection. The main agent now determines the language directly from the topic text and writes language.txt via write tool.
• Eliminates the 1-3 second python3 → python fallback delay on Windows.
• Supports 19 languages with "en" default fallback.

📝 README improvements

• Non-OpenCode installation prompt updated: step 3 now adapts per-tool (Codex CLI → skill, Claude Code → Hook, Cursor → custom commands, others → auto-detect).
• Both Chinese and English READMEs updated.

Files changed

• SKILL.md: Step 0 rewritten (LLM detection, removed detect_lang.py); Task 1 flow fixed (oracle → response → main agent write); cross-platform encoding section updated.
• prompts/task1_oracle.md: Reverted to oracle outputting JSON in response body; added {LANG} variable.
• README.md / README_EN.md: Non-OpenCode install prompts adapted per tool.
• tools/detect_lang.py: Removed (replaced by LLM-based detection).
• VERSION: 3.0.0 → 3.0.2

---

v3.0.2 — 编码修复 & 语言检测提速

相对于 v3.0.1 的变更

🔧 编码修复（Windows CP936 pipe 问题）

• Task 1 oracle 不再在回答中用 JSON 代码块输出（回答通过 shell pipe 回传时，俄语/日语等非 ASCII 字符会被 Windows CP936 编码损坏）。改为 oracle 输出 JSON 到 response body → 主 agent 用 write 工具写文件，完全绕过 pipe。
• Task 2/3/4 原本就走文件 I/O，无此问题。

⚡ 语言检测提速

• 删除 detect_lang.py（213 行 Python 脚本），改为主 agent 直接根据主题文本判定语言，用 write 工具写 language.txt。
• 消除了 Windows 上 python3 → python 回退的 1-3 秒延迟。
• 支持 19 种语言，不确定时默认英文。

📝 README 改进

• 非 OpenCode 安装提示词按工具分路线（Codex CLI → skill、Claude Code → Hook、Cursor → 自定义指令）。
• 中英文版同步更新。

变更文件

• SKILL.md：Step 0 重写（LLM 判定、删除 detect_lang.py）；Task 1 流程修复（oracle → response → 主 agent write）；跨平台编码章节更新。
• prompts/task1_oracle.md：恢复 oracle 输出 JSON 到回答 body；添加 {LANG} 变量。
• README.md / README_EN.md：非 OpenCode 安装提示词按工具分路线。
• tools/detect_lang.py：已删除（LLM 判定替代）。
• VERSION：3.0.0 → 3.0.2

v3.0.1

v3.0.1 — Windows CP936 Encoding Compatibility + BOM Tolerance

🐛 Bug fix: Windows non-English language encoding crash (核心修复)

Root cause: Windows PowerShell 5.1 console encoding CP936(GBK) cannot represent 18 non-English languages (Cyrillic, CJK, Arabic, Thai, etc.), causing:

• detect_lang.py receiving non-ASCII topics via argv gets garbled
• PowerShell Set-Content -Encoding UTF8 adds BOM → Python json.load() crashes
• Shell redirect/pipe encodes via CP936, corrupting non-ASCII text

修复方案（三层防御）:

Layer	Fix	File
输入端 Input	detect_lang.py new --file/--output args, bypass shell argv entirely	tools/detect_lang.py
读容错 Read	All Python file reads use encoding='utf-8-sig' (BOM auto-stripped)	tools/dr_tools.py dr_check.py dr_gen.py
输出端 Output	sys.stdout.reconfigure(encoding='utf-8') for UTF-8 stdout	tools/detect_lang.py

📝 Documentation (文档)

• SKILL.md Step 0 updated to cross-platform --file + --output workflow
• New section 9: Cross-platform Encoding Standards (5 hard rules)
• SKILL.md 第 9 节新增「跨平台编码规范」5 条硬性规则

✅ Verified (验证)

10 languages end-to-end: zh/ja/ko/ru/ar/th/de/es/en/uk — all pass

All pass: True

Full changelog: git log v3.0.0..v3.0.1

v3.0.0: 多语言支持 + 并行验证 + 数据质量加固

v3.0.0 — 多语言支持 · 并行验证 · 数据质量加固 · 完整容错

🆕 多语言支持（核心新能力）

• Step 0 语言判定：detect_lang.py 自动检测主题语言，所有输出跟随 $LANG
• 19 种语言支持：通过 lang_config.py 配置表覆盖中日韩英等主流语言
• 双语文档：新增 README_EN.md 英文版
• 搜索关键词语言与主题一致：中文主题使用中文关键词，英文主题使用英文关键词
• 章节 agent 不硬编码语言：跟随 $LANG 变量，不做任何语言假设

⚡ 并行验证 & 数据质量

• validate-all-chapters：ThreadPoolExecutor 并行验证所有章节，替代逐章串行
• detect-engine：SearXNG / Exa / Web Sources 三种引擎自动探测
• check-datapool：返回 source_count / fact_count 结构化统计
• 数据受限检测：自动识别数据受限场景，在报告中插入声明
• Prompt 强制脚本统计 + 兜底：避免遗漏或空数据

🛡 完整容错机制

• Task 2 数据收集失败后自动重试 1 次
• 步骤失败时先删除残留产物再重新执行
• 所有脚本/命令调用都有兜底路径（换 sys.executable / 检查路径 / 直接 Python 实现）
• 三次失败后向用户报告具体问题

📁 报告目录重组

• 案例报告/ → reports/（20 份报告全部迁移，保留 git 历史）
• 新增 3 份多语言报告：
  • 🇬🇧 Quantum Computing Market Outlook 2026
  • 🇯🇵 日本のアニメ産業のグローバル市場戦略
  • 🇰🇷 한국 반도체 산업의 글로벌 경쟁력 분석
• 删除过时的 Global SaaS ERP Market 报告

🛠 工具链重构

• dr_gen.py：重构命令解析，集成并行验证
• dr_check.py：新增多种质量检查命令（check-citations, check-headers, check-sections 等）
• dr_tools.py：新增 validate-all-chapters / detect-engine / qa-report 等高阶命令
• 新增 json-get 命令，铁律禁止内联代码

📝 文档 & 配置

• SKILL.md：完全重写，双语 description，Step 0 语言判定完整流程
• README.md：FAQ 版本更新策略说明，命令列表表格化，流程图优化
• RULES.md：标题大众化，修复过期引用（Search Router → SearXNG）
• TYPES.md：更新分类标准与编号规范
• command/research.md / update.md：同步最新流程
• .gitignore：新增 .DS_Store

🧹 规范清理

• prompts 统一编号（task1_oracle / task2_data_collection / task3_chapter_agent / task4_assembly）
• 批量优化：合并年度搜索、去掉冗余 manifest、修正编号、修复引用格式
• 章节标题统一从 outline 生成，目录锚点可点击跳转
• 引用格式改为 (N) 可点击，修复所有旧报告

🔗 完整变更记录

详见 commit log：git log v2.1.1..v3.0.0