We need answer as editor read-only format analysis teammate. Need perform task. Must first read PDF skill doc before reading PDFs. Need maybe use todo because multi-step. Need read todo maybe beginning. We are subagent explorer? User specifically editor. Need analyze PDFs format. Need read /app/.agents/skills/pdf/SKILL.md. Then read PDFs. Since PDFs can be read via read_file converted to markdown; might need maybe shell pdfinfo/pdftotext etc. Must adhere skill. Need not modify. Need output Chinese with requested sections. Also final user adds hard constraints requiring output only “R2 增量格式规范清单（从本周开始生效）”, containing新增/修订条款、执行口径、放置位置、例外/边界、对 hunter 输入的要求. They ask update format spec: don't rewrite R1, incremental append/revision clauses, preserve R1 and mark effective from this week. We need produce only R2 incremental spec. But initial task also asks analyze two PDFs. We likely need still analyze PDFs to inform R1? However final says output only R2 incremental format checklist. We need maybe include relation to R1 based on PDF analysis. Need read skill and PDFs. Let's proceed. Use todo? Complex task yes. First todo_read maybe. Then todo_write. But final requested only spec; tool use ok. Need not show todos. We already did. Now call todo_read. Let's do.[
{
"function": {
"arguments": "{}",
"name": "mshtools-todo_read"
},
"id": "mshtools-todo_read_0",
"index": 0,
"type": "function"
}
]No detailed content available.
Need create todo list: read skill, inspect PDFs, extract common/different structures, synthesize R2 incremental checklist. Use todo_write.[
{
"function": {
"arguments": "{\"todos\":[{\"id\": \"t1\", \"content\": \"读取 PDF skill 文档并提取适用阅读/解析要求\", \"status\": \"in_progress\", \"priority\": \"high\"}, {\"id\": \"t2\", \"content\": \"只读解析两份参考 PDF 的结构、版式与格式细节\", \"status\": \"pending\", \"priority\": \"high\"}, {\"id\": \"t3\", \"content\": \"归纳 R1 基线格式规范及两份样本共同/差异点\", \"status\": \"pending\", \"priority\": \"high\"}, {\"id\": \"t4\", \"content\": \"按用户硬性约束生成仅含 R2 增量格式规范清单\", \"status\": \"pending\", \"priority\": \"high\"}]}",
"name": "mshtools-todo_write"
},
"id": "mshtools-todo_write_1",
"index": 0,
"type": "function"
}
]No detailed content available.
Need read skill doc. Use read_file absolute path.[
{
"function": {
"arguments": "{\"file_path\":\"/app/.agents/skills/pdf/SKILL.md\",\"limit\":1000,\"offset\":1}",
"name": "mshtools-read_file"
},
"id": "mshtools-read_file_2",
"index": 0,
"type": "function"
}
]No detailed content available.
Skill says process route must read route file before writing any code. We're not writing code, but working existing PDFs. Instruction: before reading/processing PDFs, read skill doc and obey. Need read process route? It says mandatory before implementation, before writing any code; user says before PDF read/parse. But as applicable maybe read route/process.md. Let's do to be safe.[
{
"function": {
"arguments": "{\"file_path\":\"/app/.agents/skills/pdf/routes/process.md\",\"limit\":1000,\"offset\":1}",
"name": "mshtools-read_file"
},
"id": "mshtools-read_file_3",
"index": 0,
"type": "function"
}
]No detailed content available.
Need update todo t1 completed, t2 in_progress.[
{
"function": {
"arguments": "{\"todos\":[{\"id\": \"t1\", \"content\": \"读取 PDF skill 文档并提取适用阅读/解析要求\", \"status\": \"completed\", \"priority\": \"high\"}, {\"id\": \"t2\", \"content\": \"只读解析两份参考 PDF 的结构、版式与格式细节\", \"status\": \"in_progress\", \"priority\": \"high\"}, {\"id\": \"t3\", \"content\": \"归纳 R1 基线格式规范及两份样本共同/差异点\", \"status\": \"pending\", \"priority\": \"high\"}, {\"id\": \"t4\", \"content\": \"按用户硬性约束生成仅含 R2 增量格式规范清单\", \"status\": \"pending\", \"priority\": \"high\"}]}",
"name": "mshtools-todo_write"
},
"id": "mshtools-todo_write_4",
"index": 0,
"type": "function"
}
]No detailed content available.
Need inspect PDFs. Need maybe use read_file directly converted markdown maybe enough. Could also use pdfplumber via python/shell to get page counts, text, font sizes/colors maybe. Since final requires format. Need no modifications. Can run shell commands to inspect file sizes, pdfinfo if available, maybe python script to extract pages, text positions fonts, links annotations, images. Need not create files? Can run python in ipython maybe. But route says setup before using pdf.py commands; but we can use python pdfplumber directly? Need maybe run setup? We can check packages. Since read-only. Use shell ls, pdfinfo maybe. Need maybe use read_file with PDFs? It converts to markdown and can handle PDF. But to get fonts/annotations maybe use python libraries. Let's first shell identify paths and maybe pdfinfo. Use shell not modifying.[
{
"function": {
"arguments": "{\"command\":\"ls -lh /mnt/agents/upload/金融行业一周要闻及舆情动态（2026年5月18日-2026年5月29日）.pdf /mnt/agents/upload/金融行业一周要闻及舆情动态（2026年5月25日-2026年6月1日）.pdf && (pdfinfo /mnt/agents/upload/金融行业一周要闻及舆情动态（2026年5月18日-2026年5月29日）.pdf || true) && (pdfinfo /mnt/agents/upload/金融行业一周要闻及舆情动态（2026年5月25日-2026年6月1日）.pdf || true)\",\"description\":\"Inspect PDF file metadata\",\"timeout\":60000}",
"name": "mshtools-shell"
},
"id": "mshtools-shell_5",
"index": 0,
"type": "function"
}
]No detailed content available.
Need parse text. We can use pdftotext -layout maybe available. Also read_file PDF maybe. Let's call read_file for both maybe converted markdown.[
{
"function": {
"arguments": "{\"file_path\":\"/mnt/agents/upload/金融行业一周要闻及舆情动态（2026年5月18日-2026年5月29日）.pdf\",\"limit\":1000,\"offset\":1}",
"name": "mshtools-read_file"
},
"id": "mshtools-read_file_6",
"index": 0,
"type": "function"
}
]No detailed content available.
Read_file output for first has content but no visual formatting except maybe lines. It includes title, department, date, summary, sections, numbered headings, source mentions inline, closing. Need second.[
{
"function": {
"arguments": "{\"file_path\":\"/mnt/agents/upload/金融行业一周要闻及舆情动态（2026年5月25日-2026年6月1日）.pdf\",\"limit\":1000,\"offset\":1}",
"name": "mshtools-read_file"
},
"id": "mshtools-read_file_7",
"index": 0,
"type": "function"
}
]No detailed content available.
Need visual format details. Need maybe use pdfplumber to get fonts, sizes, positions, headings, colors, links. Need inspect PDF page images maybe. Need perhaps use Python to extract chars metadata, annotations. Since final only R2, but to make informed need R1. Use python maybe import pdfplumber, PyMuPDF maybe. Need not modify. First maybe use shell to run python to extract page count, text per page, font sizes distribution, colors, bbox, links annotations. Need check dependencies. Let's run python script read-only. It can print. Use PyMuPDF? Let's see.[
{
"function": {
"arguments": "{\"command\":\"python3 - <<'PY'\\nimport importlib.util\\nfor m in ['pdfplumber','fitz','pikepdf']:\\n print(m, bool(importlib.util.find_spec(m)))\\nPY\",\"description\":\"Check PDF parsing libraries\",\"timeout\":60000}",
"name": "mshtools-shell"
},
"id": "mshtools-shell_8",
"index": 0,
"type": "function"
}
]No detailed content available.
Need maybe use PyMuPDF for detailed formatting. Extract font spans. Need maybe inspect first 3 pages and last pages, maybe every page. Need report with fonts, sizes, line spacing, page size, margin, color, title size, headings. We can write script no file writes. Let's extract text blocks spans with bbox/font/size/color for first and second. Also annotations links. Need maybe identify cover vs not? Read_file starts immediately title no obvious cover. Need maybe first page is report title at top; no separate cover? Need inspect screenshot image maybe. Use PyMuPDF to render first page maybe read image? We can use read_file PDF displayed? It extracted text only. Could render screenshots with Python? Need maybe not write? But read_file can read PDF directly maybe not display. We can use python to print page block details. Let's use fitz get_text('dict'), spans. Need pages dimensions. Use script to summarize font sizes/colors, first page spans sorted. Also detect link annotations.[
{
"function": {
"arguments": "{\"command\":\"python3 - <<'PY'\\nimport fitz, os, json, statistics, collections, re\\npaths=[\\n'/mnt/agents/upload/金融行业一周要闻及舆情动态（2026年5月18日-2026年5月29日）.pdf',\\n'/mnt/agents/upload/金融行业一周要闻及舆情动态（2026年5月25日-2026年6月1日）.pdf'\\n]\\nfor path in paths:\\n doc=fitz.open(path)\\n print('\\\\n===', os.path.basename(path), 'pages', doc.page_count, '===')\\n print('metadata', doc.metadata)\\n page=doc[0]\\n print('page0 rect', page.rect)\\n # sizes/colors/fonts count\\n cnt=collections.Counter(); colorcnt=collections.Counter(); fontcnt=collections.Counter()\\n for p in doc:\\n d=p.get_text('dict')\\n for b in d['blocks']:\\n if b.get('type')!=0: continue\\n for l in b['lines']:\\n for s in l['spans']:\\n text=s['text'].strip()\\n if not text: continue\\n cnt[round(s['size'],1)] += len(text)\\n colorcnt[s['color']] += len(text)\\n fontcnt[s['font']] += len(text)\\n print('font sizes weighted:', cnt.most_common(20))\\n print('colors weighted:', [(hex(k),v) for k,v in colorcnt.most_common(10)])\\n print('fonts weighted:', fontcnt.most_common(10))\\n # links\\n link_count=0\\n link_examples=[]\\n for i,p in enumerate(doc):\\n links=p.get_links()\\n if links:\\n link_count += len(links)\\n for ln in links[:5]:\\n link_examples.append((i+1, ln.get('uri'), ln.get('from')))\\n print('links count', link_count, 'examples', link_examples[:10])\\n # first page spans detail\\n for pi in [0,1,doc.page_count-1]:\\n if pi<0 or pi>=doc.page_count: continue\\n p=doc[pi]\\n print(f'-- page {pi+1} first spans --')\\n spans=[]\\n for b in p.get_text('dict')['blocks']:\\n if b.get('type')!=0: continue\\n for l in b['lines']:\\n line_text=''.join(s['text'] for s in l['spans']).strip()\\n if not line_text: continue\\n # aggregate style from first span\\n s=l['spans'][0]\\n spans.append((round(l['bbox'][1],1), round(l['bbox'][0],1), round(l['bbox'][2],1), round(l['bbox'][3],1), s['font'], round(s['size'],1), hex(s['color']), line_text[:100]))\\n for item in spans[:25]:\\n print(item)\\n if len(spans)>25: print('... total lines', len(spans))\\nPY\",\"description\":\"Extract PDF formatting metadata\",\"timeout\":120000}",
"name": "mshtools-shell"
},
"id": "mshtools-shell_9",
"index": 0,
"type": "function"
}
]No detailed content available.
Output truncated during second page; but enough. Need maybe re-run focused for second pages maybe last page, style details. Also get all headings maybe from fonts/size. Need determine R1 common structure and differences. We already have text. Need more visual details: margins first sample x=28.9, second x=72. First has no clickable links, inline source text, no reference list. Second has clickable '查看原文' links, reference list at end, 265 links, blue color 0x563c1? Need identify color 0x563c1 is maybe #0563C1 standard hyperlink blue. Font sizes: first body 12, title 25.5, level1 16.5, level2 15, level3 maybe 13.5? second body 11, title 18 bold, section title 16.1, subsection 15.1, reference maybe 11. Need margins: first left 28.9 pt (approx 1.02 cm), top 72 pt for title then content top 252 after summary? Wait page 2 starts y31, top margin 31 pt. It seems no page headers/footers. First title centered? x 28.9 full line, maybe due width not centered? Actually title line x 28.9 to 553.2, yes centered? second title x72 to 499, margin 72, left aligned? The text line spans nearly full, but title starts at left margin. It might not be centered. Need inspect alignment: first lines x28.9, not centered? In first page title line x28.9, maybe full width; second line x28.9 to 287.5, so left aligned. But title likely left aligned at 28.9 not center. Summary heading x28. Second left margin 72, right max 523 (page width 595), so left/right margins ~72pt. That seems Word default 1 inch. First has 0.4 inch margin. Need perhaps due HTML PDF Chromium with CSS 0.4in? Need perhaps first page content top after title x=28. Second generated by macOS Quartz from Word likely margins 72 pt, fonts DengXian/Arial. Need final only R2 incremental, but user says include R1 relation. Need maybe output as R2 spec with new/revised clauses, execution口径， placement, exceptions, hunter input requirements. Must not rewrite R1; can reference R1 existing. Need maybe include R1 observed baseline as relation? But final only R2; not R1. Need maybe mention R2 is incremental append to R1, effective 2026-06-01 to 2026-06-07. Need include 6 hard constraints and perhaps placement in report: 核心摘要 before existing 摘要与核心观察/观点. Footnote on homepage. Page <=30. Exclude stock trends. Time window. Must cover被查/被罚/人事任免舆情. Need hunter input requirements: Hunter should provide only in-window items with event/disclosure/announcement dates; tag item types especially被查/被罚/人事任免； supply source URLs; no stock走势； include status and relevance; deduplicate; include date fields; identify outside-window items as exclude/background only. Need mention priority: R2 overrides R1 when conflict: page limit, core summary, footnote, stock exclusion. R1 retained otherwise. Need perhaps specify homepage footnote placement not only final note. Also summary section mandatory. Need note if first sample had no core summary? It had 摘要与核心观察/观点； R2 adds 核心摘要 at very front before R1摘要. Maybe combine? User says report最前面加“核心摘要”章节，面向总裁及高管概览. Need place after编制日期 before R1摘要. Must be first substantive section. It may include 5-8 bullet points or 1页内？ We can specify. But output only incremental format spec, not necessarily content. Good.
Need perhaps inspect links in first? It had no link annotations despite source names. second has clickable links. Need R1 relation with links maybe no change? R2 should not rewrite but maybe include additional requirements about source dates and coverage. Need maybe mention for被查/被罚/人事任免 items: must appear in both relevant topical section and 舆情/风险 section if material; but avoid duplication. For examples: 丁向群任免 (5/29 in sample), 人行四川分行主要领导任免， 交银金科副总裁被查. If in current week 6/1-6/7, include only if occurred/disclosed/announced in range. Need mention not漏. For current week, must capture these categories across监管部门/银行/金科/信托/券商基金 etc. Need maybe require hunter input includes a dedicated '被查/被罚/人事任免' candidate list and flags severity.
Need maybe use todo updates. But maybe still need t2 comple
... [365 chars truncated][
{
"function": {
"arguments": "{\"command\":\"python3 - <<'PY'\\nimport fitz, re, os\\npaths=[\\n'/mnt/agents/upload/金融行业一周要闻及舆情动态（2026年5月18日-2026年5月29日）.pdf',\\n'/mnt/agents/upload/金融行业一周要闻及舆情动态（2026年5月25日-2026年6月1日）.pdf'\\n]\\npatterns=[r'^[一二三四五六七]、', r'^\\\\d+\\\\.\\\\d+', r'^\\\\d+\\\\.\\\\d+\\\\.\\\\d+', r'^摘要', r'^备注', r'^参考资料']\\nfor path in paths:\\n doc=fitz.open(path)\\n print('\\\\n===', os.path.basename(path), '===')\\n for i,p in enumerate(doc):\\n text=p.get_text('text')\\n lines=[ln.strip() for ln in text.splitlines() if ln.strip()]\\n hits=[]\\n for ln in lines:\\n if any(re.search(pt,ln) for pt in patterns):\\n hits.append(ln)\\n if hits:\\n print('Page', i+1)\\n for h in hits[:30]: print(' ', h)\\nPY\",\"description\":\"List PDF section heading lines\",\"timeout\":120000}",
"name": "mshtools-shell"
},
"id": "mshtools-shell_10",
"index": 0,
"type": "function"
}
]No detailed content available.
The heading regex missed first level for second because line starts 一 maybe in page 1? It printed 1.1 only. Maybe there is '一、监管层与...' but regex used normal Chinese numerals? Actually maybe starts with ⼀ compatibility char not matching. Fine. Need maybe inspect first page title and headings from screenshot. Need perhaps render page images? Could use python to save? But output not needed. Need perhaps use PyMuPDF to get blocks and colors. Need maybe second last pages to see reference formatting and end. Use script focused last page.[
{
"function": {
"arguments": "{\"command\":\"python3 - <<'PY'\\nimport fitz, os\\npath='/mnt/agents/upload/金融行业一周要闻及舆情动态（2026年5月25日-2026年6月1日）.pdf'\\ndoc=fitz.open(path)\\nfor pi in [0,12,13,17]:\\n if pi>=doc.page_count: continue\\n p=doc[pi]\\n print('\\\\n-- page',pi+1,'--')\\n lines=[]\\n for b in p.get_text('dict')['blocks']:\\n if b.get('type')!=0: continue\\n for l in b['lines']:\\n text=''.join(s['text'] for s in l['spans']).strip()\\n if not text: continue\\n s=l['spans'][0]\\n lines.append((round(l['bbox'][1],1), round(l['bbox'][0],1), round(l['bbox'][2],1), round(l['bbox'][3],1), s['font'], round(s['size'],1), hex(s['color']), text[:140]))\\n for item in lines[:50]: print(item)\\n print('total lines', len(lines))\\nPY\",\"description\":\"Inspect selected second PDF pages\",\"timeout\":120000}",
"name": "mshtools-shell"
},
"id": "mshtools-shell_11",
"index": 0,
"type": "function"
}
]No detailed content available.
Good. Need maybe for first last page note style. We have. Need maybe extract table? No tables. Need maybe detect no page headers/footers, no page numbers. Could use text lines not showing page numbers. Maybe there are no headers/footers. Need maybe inspect annotations first none, second many links. No cover; no tables/cards/columns. Colors black, hyperlink blue, grey note. Need maybe screenshots? Not necessary. Need perhaps detect if second has footnote at end, first note at end. R2 requires homepage footnote. Need note homepage footnote can be small gray at bottom or below 编制日期？ Since PDF page footnote at homepage. Need placement: at bottom of first page or immediately below compilation info? User says 首页加脚注； should be a footnote on first page. Recommend bottom-left or bottom-center of first page, grey 9-10pt, before正文不干扰. Maybe exact text: “本报告部分内容由 AI 生成，仅供参考”. This supersedes R1 final note “（注：文档部分内容可能由 AI 生成）”: move/copy to首页脚注； R1末尾AI说明可删除或保留？ User says加首页脚注； to avoid conflict, specify R2新增首页脚注，末尾 AI 说明如保留应一致但不得替代首页脚注； recommended delete duplicate末尾AI注 or keep备注中的风险提示？ Need output relation/override. But final only R2 incremental. Good.
Need maybe think about R1 common structures and differences: First structure seven sections: title，编制单位/日期，摘要与核心观察，一金融监管与政策动态，二金融基础设施运营动态，三金融机构市场要闻 with subsegments，四金融科技厂商动态，五行业舆情与风险提示，六总结与趋势预判，七备注，落款，AI注. Second: title，编制，摘要与核心观点，一监管层与金融基础设施动态 (merged first two)，二银行业机构动态，三金融科技厂商竞争分析，四信托与银行理财子行业动态，五行业舆情与重大风险动态，六行业趋势研判，备注，编制单位/日期，参考资料，AI注. Common: A4, no cover, title+编制 at front, summary, numbered sections, body paragraphs, risk/舆情， trend summary, remarks, sign-off, AI note. Differences: page count 12 vs 18; first no clickable links/reference list, inline source text; second clickable 查看原文 plus references; first more granular subheading levels including 3.1; second bullet numbering with nested 1. and 3. etc due Word, more links, larger margins; first uses black body no URLs; second has URL reference list. But final not require full R1? The final user says output only R2增量. They don't want R1 analysis? Need maybe not include sections from initial output requirements. But maybe include some relation to R1 observed? They asked only R2, so no chapter common/diff. We can mention R2 preserves R1 existing规范. Need maybe include in R2 a brief "与R1关系/覆盖优先级". Need not include common/diff. But maybe should mention based on samples, R1 remains baseline. The final output should have exactly R2 incremental checklist with sections: 新增/修订条款， 执行口径， 放置位置， 例外/边界， 对 hunter 输入的要求. Could have a table with each clause. That satisfies. We should not output R1 common/diff; the user explicitly now says output only R2增量格式规范清单. Need ensure not write long explanation. But perhaps include "本清单不重写R1". Good.
Need maybe update todos. Need t2 complete, t3 complete maybe. We need mark after final maybe. Need perhaps use todo_write.[
{
"function": {
"arguments": "{\"todos\":[{\"id\": \"t1\", \"content\": \"读取 PDF skill 文档并提取适用阅读/解析要求\", \"status\": \"completed\", \"priority\": \"high\"}, {\"id\": \"t2\", \"content\": \"只读解析两份参考 PDF 的结构、版式与格式细节\", \"status\": \"completed\", \"priority\": \"high\"}, {\"id\": \"t3\", \"content\": \"归纳 R1 基线格式规范及两份样本共同/差异点\", \"status\": \"completed\", \"priority\": \"high\"}, {\"id\": \"t4\", \"content\": \"按用户硬性约束生成仅含 R2 增量格式规范清单\", \"status\": \"in_progress\", \"priority\": \"high\"}]}",
"name": "mshtools-todo_write"
},
"id": "mshtools-todo_write_12",
"index": 0,
"type": "function"
}
]No detailed content available.
Need craft final. But before final maybe consider if any need to send message? We are teammate? There is send_message? User gave task, answer in final likely. Need maybe not. Need perhaps mark t4 completed before final? Use todo_write.[
{
"function": {
"arguments": "{\"todos\":[{\"id\": \"t1\", \"content\": \"读取 PDF skill 文档并提取适用阅读/解析要求\", \"status\": \"completed\", \"priority\": \"high\"}, {\"id\": \"t2\", \"content\": \"只读解析两份参考 PDF 的结构、版式与格式细节\", \"status\": \"completed\", \"priority\": \"high\"}, {\"id\": \"t3\", \"content\": \"归纳 R1 基线格式规范及两份样本共同/差异点\", \"status\": \"completed\", \"priority\": \"high\"}, {\"id\": \"t4\", \"content\": \"按用户硬性约束生成仅含 R2 增量格式规范清单\", \"status\": \"completed\", \"priority\": \"high\"}]}",
"name": "mshtools-todo_write"
},
"id": "mshtools-todo_write_13",
"index": 0,
"type": "function"
}
]No detailed content available.
Need final in Chinese. Must only output R2增量格式规范清单. Need include effective date. Need format maybe:
# R2 增量格式规范清单（从本周开始生效）
适用范围：2026-06-01至2026-06-07周报；保留R1全部既有规范；R2为增量追加/修订条款；冲突时R2优先。
## 1. 新增/修订条款
Table: 编号， 条款， 类型， 与R1关系/优先级， 执行口径， 放置位置， 例外/边界， 对 hunter 输入要求.
Could be a big table but readability. Need include all required fields. Maybe each clause with subheadings.
Need include six clauses:
R2-01 日期窗口收敛：只保留起止日期内信息，不要过多非当周内容. 放置位置： 全文；备注可声明统计周期；核心摘要也不得超窗. Exception: 非当周背景最多一句话作为必要上下文；若事件发生日期不在窗口但公告/披露在窗口，纳入并以披露日说明；若招标公告在窗口、项目起始/预算之前，纳入；中标结果如结果公告在窗口；计划性维护如发生在窗口或公告在窗口且影响窗口； but no old周回顾. Hunter input: 每条标注事件日期/披露日期/公告日期/来源发布时间；超出窗口单独列入“剔除/背景”；给 inclusion_reason.
R2-02 被查/被罚/人事任免必覆盖. 执行口径： 设独立校验清单，至少扫描监管机构、纪委监察、行政处罚、任职资格批复、银行/信托/金科/券商基金高管人事等；重大舆情如 examples. Placement: 可放在“行业舆情与重大风险动态/风险提示”及对应机构章节；核心摘要中若重大需列1条； not only厂商. Boundary: 仅在窗口内发生/披露/公告；已任前传闻未公告不纳；非金融行业或与周报无关不纳. Hunter: 标记类型【被查】【被罚】【人事任免】、主体、职务、发布机构、日期、来源URL、严重性； must not漏； with confidence. Need mention交银金科副总裁被查 if within current week? For current week 6/1-6/7, only if within window; examples from sample? User says例如 these不能漏 but 成稿受第1条； maybe in current week if they are in window? Example Ding 5/29 not in 6/1-6/7? But user says 本周指 6/1-6/7; examples include 5/29? They may be hard examples from previous week. Need phrase: examples为类别示例，仅在本周窗口内发生/披露/公告时纳入；不作为跨周补录理由. But user says“这类信息不能漏；但成稿仍须受第1条约束，仅纳入起止日期内发生/披露/公告的信息.” So yes.
R2-03 核心摘要 mandatory. Place after编制日期 before R1摘要/核心观察. Executive summary. Format maybe heading“核心摘要”; 3-6条/≤1页；面向总裁及高管；结论先行；包含监管/机构/科技/风险/本周关注. Not content summary? But spec. Relation: 新增，不替代R1摘要； R1摘要可保留为行业概览. Hunter: 提供 top 5-8 high-impact items and why matters, risk level.
R2-04 首页脚注. Place first page bottom or under compilation info. Text exact. Font small gray. If R1末尾AI note exists, homepage footnote is mandatory；末尾note可保留但文案一致 or remove duplicate. Hunter not relevant maybe sources can indicate AI-assisted?
R2-05 上市金融机构股票走势不纳入. Execution: exclude stock price涨跌幅、K线、股价表现、市值波动； exception if监管处罚/人事/重大舆情导致股价波动？ Do not include stock走势； can mention market reaction only if directly tied to舆情？ But user says股票走势信息不纳入； so no股价走势. Financial metrics? It says上市金融机构股票走势； not all上市公司业务. Exclude资本市场行情 sections. Hunter must filter out stock trends; if source includes股价， strip.
R2-06 页数≤30. Place overall. Execution: A4, existing style; common sample 12-18 pages; upper limit 30 including references. If >30, prioritize summarization, reduce references, merge low-impact items, preserve mandatory categories. Boundary: page count includes封面？ No cover. includes参考资料/备注/落款. Hunter input maybe cap raw items and rank importance; not dump >30 pages; mark must-have/optional.
Need also maybe include R2-00 precedence. Could include as own clause. Need mention not rewrite R1.
Need maybe specify "从本周开始生效" in title and under scope. This week = 2026-06-01至2026-06-07. Good.
Need maybe include "对 hunter 输入的要求" across all. Could have separate section at end to be clear: Hunter must submit structured cards. But user asks each contains; we can include both table and aggregate. Maybe not too long but enough. Desired oververbosity 5. We can produce concise but complete.
Need maybe not include R1 common/diff; okay.
Need maybe mention no file modifications? Not in final unless necessary? Not needed.
Need maybe not claim read PDFs? Could mention? User asked output only spec; no need.
Need final with Chinese. Avoid R1 full rewrite. Ensure "R2 增量格式规范清单（从本周开始生效）" exactly. Include "R1已有规范继续沿用，不重写；以下条款为追加/修订；冲突时R2优先." Good.
Need maybe include section "执行口径" etc. Let's formulate elegantly.
Potential issue: User wanted "新增/修订条款、执行口径、放置位置、例外/边界、对 hunter 输入的要求". A table with columns exactly. Good.
Maybe use Markdown table. Long table may be wide. Could use numbered clauses each with bullets for the fields. Maybe more readable and still includes fields. Let's do headings:
# R2 增量格式规范清单（从本周开始生效）
适用周期：2026-06-01至2026-06-07周报。
与R1关系：R1保留；R2追加/修订；冲突优先；未涉事项按R1。
## R2-00 覆盖优先级
- 新增/修订条款：R2为R1增量条款.
... [1359 chars truncated]# R2 增量格式规范清单（从本周开始生效）
**适用范围**：2026-06-01 至 2026-06-07 周报。
**与 R1 关系**：本清单**不重写 R1**；R1 既有章节、标题层级、链接、落款、备注、版式等规范继续沿用。以下为**增量追加/修订条款**；如与 R1 冲突，**R2 优先**。
---
## R2-01 日期窗口收敛条款
- **新增/修订条款**：只保留周报起止日期内的信息，不得堆砌过多非当周内容。
- **执行口径**：
- 以事件发生日、监管披露日、公告发布日、处罚/任免披露日为准。
- 只有落在本周起止日期内的信息，才进入正文、摘要、核心摘要、舆情与风险章节。
- 非当周信息原则上不纳入；如确为背景，最多一句话，不得展开。
- **放置位置**：全文通用；备注中继续保留“本报告统计周期为……”表述。
- **例外/边界**：
- 事件发生在本周前，但公告/披露/处罚/任免在本周，可纳入，并明确“披露/公告日期”。
- 招标、中标、系统维护等事项，以公告日、结果公示日、维护执行日是否在窗口内判断。
- 不为补充完整性而回溯上周或更早内容。
- **对 hunter 输入的要求**：
- 每条素材必须标注：`事件日期 / 披露日期 / 公告日期 / 来源发布时间`。
- 同步标注“纳入理由”或“剔除原因：超出本周窗口”。
- 不得把非当周素材混入当周正文素材池。
---
## R2-02 被查 / 被罚 / 人事任免舆情必覆盖条款
- **新增/修订条款**：必须覆盖本周内的被查、被罚、人事任免类舆情；如“金融监管总局丁向群任免”“人行四川分行主要领导任免”“交银金科副总裁被查”这类事项，属于高优先级素材，不得遗漏。
- **执行口径**：
- 将“被查、被罚、人事任免”设为强制检索与强制校验类别。
- 覆盖对象包括但不限于：监管部门、银行、银行理财子、信托、券商、基金、金融科技子公司、支付及消费金融等金融机构。
- 重大事项应同时体现在对应机构章节与“行业舆情 / 重大风险 / 风险提示”章节中；如足够重要，应进入“核心摘要”。
- **放置位置**：
- 首选：`行业舆情与重大风险动态`、`行业舆情与风险提示`、`重点风险舆情`。
- 同步可放入：`监管动态`、`银行业机构动态`、`金融机构市场要闻`、`金融科技厂商动态`等相关章节。
- **例外/边界**：
- 仍受 R2-01 约束：仅纳入本周内发生、披露、公告或官方通报的信息。
- 传闻、未经证实的市场消息不纳入。
- 上周已充分披露且本周无新增进展的，不重复展开。
- **对 hunter 输入的要求**：
- 必须单独输出“被查 / 被罚 / 人事任免”候选清单。
- 每条至少包含：主体机构、人员/岗位、事项类型、官方来源、披露日期、影响判断、是否建议进入核心摘要。
- 对重大廉政、监管处罚、监管高层任免，标记为 `MUST_INCLUDE`。
---
## R2-03 新增“核心摘要”章节条款
- **新增/修订条款**：报告最前面必须新增“核心摘要”章节，面向公司总裁及高管快速概览。
- **执行口径**：
- “核心摘要”采用结论先行写法，不做新闻堆砌。
- 建议控制在 3–6 条，原则上不超过 1 页。
- 每条应包含：本周核心判断、影响对象、风险/机会、需管理层关注点。
- 重点呈现监管、重大人事、重大处罚/被查、金融科技风险、重要业务趋势。
- **放置位置**：
- 顺序为：报告标题 → 编制单位 / 编制日期 → 首页脚注 → **核心摘要** → R1 原有“摘要与核心观察 / 摘要与核心观点 / 本期概览” → 正文。
- “核心摘要”是第一个实质性章节。
- **例外/边界**：
- 不替代 R1 原有摘要章节；R1 摘要可保留为更完整的行业概览。
- 若本周重大事项较少，也不得删除该章节，可压缩为 3 条以内。
- **对 hunter 输入的要求**：
- hunter 需提供“高管必看 Top 事项”清单，并按重要性排序。
- 每条 Top 事项需给出：一句话结论、影响范围、风险等级、建议是否入核心摘要。
---
## R2-04 首页 AI 生成脚注条款
- **新增/修订条款**：报告首页必须增加脚注：**“本报告部分内容由 AI 生成，仅供参考”**。
- **执行口径**：
- 文案保持原句，不改写、不弱化。
- 样式应与正文区分，建议小字号、灰色、页脚或标题页底部呈现。
- 该脚注是首页强制元素，不得只放在文末备注中。
- **放置位置**：
- 首页底部脚注区域；如首页空间紧张，可放在编制单位 / 编制日期下方、核心摘要上方。
- 不得放到正文后段或最后一页替代首页脚注。
- **例外/边界**：
- R1 如已有文末“文档部分内容可能由 AI 生成”提示，可保留，但不能替代首页脚注。
- 如为避免重复需删减，优先保留首页脚注。
- **对 hunter 输入的要求**：
- hunter 不需处理该脚注，但输入素材不得包含“已确认完全人工生成”等与之冲突的表述。
---
## R2-05 上市金融机构股票走势排除条款
- **新增/修订条款**：上市金融机构的股票走势信息不纳入报告。
- **执行口径**：
- 不写股价涨跌幅、K 线走势、市值波动、短期股价表现、资本市场行情解读。
- 不设置“上市金融机构股价表现”“资本市场走势”类小节。
- 若来源文章中夹带股价信息，成稿时应剔除。
- **放置位置**：全文通用；尤其适用于银行、券商、保险、信托、金融科技上市公司相关章节。
- **例外/边界**：
- 上市金融机构的监管处罚、人事任免、业务调整、招投标、系统建设、重大舆情仍可纳入。
- 排除的是“股票走势 / 股价表现”，不是排除上市公司本身的经营与舆情信息。
- 如重大舆情引发市场反应，只可概括“引发市场关注”，不得写具体股价走势。
- **对 hunter 输入的要求**：
- hunter 输入时应过滤股票行情类素材。
- 若素材来源含股价段落，应标注“股价段落剔除，仅保留经营/监管/舆情信息”。
---
## R2-06 页数上限条款
- **新增/修订条款**：报告总页数控制在 **30 页以内**。
- **执行口径**：
- 30 页为硬上限，包含正文、核心摘要、备注、参考资料、落款等全部内容。
- 两份样本常见页数约为 12–18 页；本周新增“核心摘要”后仍应控制在 30 页以内。
- 超出时优先压缩低价值事项、合并同类项、精简参考链接和低影响厂商动态。
- **放置位置**：全文总控；排版、编辑、导出 PDF 前均需校验。
- **例外/边界**：
- 不得因压缩页数删除 R2-02 要求的被查 / 被罚 / 人事任免重大事项。
- 不得删除首页 AI 脚注。
- 不得删除核心摘要。
- 页数压力优先由低优先级招投标、重复性厂商动态、冗余背景信息承担。
- **对 hunter 输入的要求**：
- hunter 应对素材分级：`MUST_INCLUDE / SHOULD_INCLUDE / OPTIONAL`。
- 同一主题多条来源应去重，只保留权威来源和关键差异。
- 不得无筛选倾倒长清单；需提供可压缩空间建议。
---
## hunter 输入最低字段要求
为便于执行 R2，hunter 本周起提交素材时，每条至少包含：
1. `标题`
2. `机构 / 主体`
3. `事项类型`：监管政策 / 被查 / 被罚 / 人事任免 / 招投标 / IT 建设 / 业务动态 / 舆情风险 / 其他
4. `事件日期`
5. `披露日期 / 公告日期`
6. `是否在本周窗口内`
7. `来源名称`
8. `来源 URL`
9. `重要性等级`：MUST_INCLUDE / SHOULD_INCLUDE / OPTIONAL
10. `是否建议进入核心摘要`
11. `是否涉及股票走势`：如涉及，需标注并剔除股价段落
12. `纳入建议 / 剔除原因`