deck2video-skill / ingest_parser.py
VanyaJ's picture
Release v4.5: Seamless One-Pass Sentence Prosody + Acoustic Valley Subtitle Sync & Balanced Line-Break + Context-Aware Korean Numeral G2P
775de68 verified
Raw History Blame Contribute Delete
62.9 kB
""" Universal Document-to-Deck Ingestion Parser (`ingest_parser.py`) Supports: - PowerPoint (.pptx) β€” Slides, bullets, tables, speaker notes - Word (.docx) β€” Headings, paragraphs, bullet lists, quotes, tables - Markdown (.md/.txt) β€” #, ##, -, >, tables - HTML (.html/.htm) β€” Reveal.js <section>, .slide, or <h1>/<h2>/<ul>/<p>/<blockquote> - JSON (.json) β€” Direct DeckSpec schema Maps every extracted slide into one of 5 visual motion archetypes: 1. `arc_cards` β€” 3D cylindrical card carousel + glowing big counter 2. `mockup_steps` β€” Left app/slide list mockup + Right 3-card sequential focus glow 3. `book_quote` β€” 3D perspective open manual/book spread + cyan highlighter sweep 4. `terminal_flow` β€” Dark glass CLI/code window typing + glowing node flow diagram 5. `metric_summary` β€” 3 neon glass pillars + final takeaway callout badge """ from __future__ import annotations import json import re import zipfile from pathlib import Path from typing import Any class DRMProtectedFileError(RuntimeError): """Raised when an OpenXML (.pptx/.docx) file is encrypted by enterprise DRM or Office protection.""" def __init__(self, file_path: Path, reason: str = "") -> None: self.file_path = file_path self.reason = reason super().__init__(f"DRM or encrypted document detected: {file_path} ({reason})") AGENT_HANDOFF_JSON_SCHEMA_EXAMPLE = """{ "title": "2026 사내 AI λ„μž… μ „λž΅ λ³΄κ³ μ„œ", "slides": [ { "title": "μΆ”μ§„ λ°°κ²½ 및 핡심 과제 5κ°€μ§€", "chapter": "01 μΆ”μ§„ λ°°κ²½", "bullets": [ "λ¬Έμ„œ λ³΄μ•ˆ μ€€μˆ˜ | 사내 DRM μ—΄λžŒ μŠ€ν‚¬κ³Ό νŒŒμ΄ν”„λΌμΈ μ•ˆμ „ 연동", "생산성 ν–₯상 | λ³΄κ³ μ„œ λΈŒλ¦¬ν•‘ μ˜μƒ μ œμž‘ μ‹œκ°„ 90% 단좕", "ν‘œμ€€ν™” | 16μ’… λͺ¨μ…˜ κ·Έλž˜ν”½ λ ˆμ΄μ•„μ›ƒ μžλ™ λ§€ν•‘" ], "notes": "사내 DRM μŠ€ν‚¬λ‘œ 읽은 λ°œν‘œμž λ…ΈνŠΈ λ˜λŠ” μŠ¬λΌμ΄λ“œ 핡심 μ„€λͺ… λŒ€λ³Έμž…λ‹ˆλ‹€. λ°œν‘œμž λ…ΈνŠΈκ°€ μ—†λŠ” μž₯ν‘œλŠ” μ—μ΄μ „νŠΈκ°€ μŠ¬λΌμ΄λ“œ λ‚΄μš©μ„ λ°”νƒ•μœΌλ‘œ μžμ—°μŠ€λŸ¬μš΄ 2~3λ¬Έμž₯ ꡬ어체 λ‚΄λ ˆμ΄μ…˜μ„ μž‘μ„±ν•΄ λ„£μœΌμ„Έμš”." }, { "title": "κΈ°μ‘΄ 방식 vs λͺ¨μ…˜ νŒŒμ΄ν”„λΌμΈ 비ꡐ", "chapter": "02 λ„μž… 효과", "table": [ ["비ꡐ ν•­λͺ©", "λ‹¨μˆœ μŠ¬λΌμ΄λ“œ 캑처", "Deck2Video λͺ¨μ…˜ μ—”μ§„"], ["μ‹œκ° λͺ°μž…도", "βœ•", "βœ“"], ["단어 ν•˜μ΄λΌμ΄νŠΈ μžλ§‰", "βœ•", "βœ“"], ["사내 폐쇄망 지원", "β–³", "βœ“"] ], "notes": "ν‘œ 데이터도 table λ°°μ—΄λ‘œ μ „λ‹¬ν•˜λ©΄ λ„€μ˜¨ 비ꡐ λ§€νŠΈλ¦­μŠ€λ‚˜ 데이터 차트 λͺ¨μ…˜μœΌλ‘œ μžλ™ λ³€ν™˜λ©λ‹ˆλ‹€." } ] }""" def check_openxml_drm(file_path: Path) -> None: """Verify whether a .pptx/.docx file is a valid OpenXML ZIP or encrypted by corporate DRM.""" try: header = file_path.read_bytes()[:64] except Exception as exc: raise DRMProtectedFileError(file_path, f"file read blocked: {exc}") from exc if not header: raise DRMProtectedFileError(file_path, "empty file (0 bytes)") # Standard OpenXML (.pptx, .docx, .xlsx) starts with ZIP magic bytes PK\x03\x04 (or PK\x05\x06 / PK\x07\x08) if header[:2] == b"PK" and zipfile.is_zipfile(file_path): return upper_hdr = header.upper() vendor = "Corporate DRM / Encrypted Container" if b"FASOO" in upper_hdr or b"FSN" in upper_hdr: vendor = "Fasoo DRM (νŒŒμˆ˜λ‹·μ»΄ λ¬Έμ„œλ³΄μ•ˆ)" elif b"MARKANY" in upper_hdr or b"DOCUMENT SAFER" in upper_hdr: vendor = "MarkAny DRM (λ§ˆν¬μ• λ‹ˆ λ¬Έμ„œλ³΄μ•ˆ)" elif b"SOFTCAMP" in upper_hdr or b"SCDSA" in upper_hdr: vendor = "Softcamp DRM (μ†Œν”„νŠΈμΊ ν”„ λ¬Έμ„œλ³΄μ•ˆ)" elif header[:8] == b"\xd0\xcf\x11\xe0\xa1\xb1\x1a\xe1": vendor = "OLE2 Encrypted Package (MS Purview AIP / Office μ•”ν˜Έν™” λ˜λŠ” κ΅¬ν˜• .ppt/.doc λ°”μ΄λ„ˆλ¦¬)" raise DRMProtectedFileError( file_path, f"{vendor} β€” magic={header[:12]!r} (python-pptx/python-docx cannot open encrypted binary directly)", ) ARCHETYPES = [ "hero_3d_briefing", "media_showcase_split", "cinema_broll_fullscreen", "dossier_stamp_card", "archify_architecture", "anatomy_mechanic", "anime_svg_chart", "custom_viz_stage", "changelog_vertical", "mission_staircase", "swarm_network", "pricing_matrix", "arc_cards", "bento_grid", "mockup_steps", "vs_comparison", "book_quote", "data_bars", "notebook_flip", "process_timeline", "terminal_flow", "metric_summary", ] def _split_sentences(text: str) -> list[str]: """Splits paragraph text ONLY at true terminal sentence punctuation (never mid-clause on words like '보닀 ').""" text = re.sub(r"\s+", " ", text.strip()) if not text: return [] parts = re.split(r"(?<=[.!?β€¦γ€‚γ€‚οΌοΌŸ])\s+(?![0-9])", text) cleaned = [p.strip() for p in parts if p.strip()] return cleaned if cleaned else [text] def _is_unfinished_clause(line: str) -> bool: """Returns True if a note line ends mid-sentence (e.g., '온천물이 μ†Ÿμ•„', 'λ–¨μ–΄μ§„ 곀좩이', '사라지면,') and should flow into the next line.""" s = line.strip() if not s or "||" in s: return False if s[-1] in ".!?β€¦γ€‚γ€‚οΌοΌŸ": return False if s.endswith(("λ‹€", "μš”", "μ£ ", "까", "λ„€", "음", "함", "됨", "μž„")): return False return True def _clean_chapter_label(raw_ch: str) -> str: cleaned = re.sub(r"^\d+\s*[.Β·\-:]?\s*", "", raw_ch.strip()).strip() return (cleaned or raw_ch.strip())[:16] def _infer_archetype( idx: int, total: int, title: str, bullets: list[str], notes: str, has_media: bool = False, has_stamp: bool = False, has_broll: bool = False, has_custom_viz: bool = False, ) -> str: combined = f"{title} {' '.join(bullets)} {notes}".lower() if has_custom_viz: return "custom_viz_stage" if has_broll: return "cinema_broll_fullscreen" if has_media: return "media_showcase_split" if has_stamp or any(k in combined for k in ["νŠΉν—ˆ", "곡식 ν•΄λͺ…", "곡식 μž…μž₯", "판결", "곡문", "μ„±λͺ…μ„œ", "patent", "official statement"]): return "dossier_stamp_card" if any(k in combined for k in ["μˆ˜μ‹¬", "단면", "μž μˆ˜μ •", "ν•΄λ°œ", "μ§€μΈ΅", "칼데라", "심해", "μ§€ν•˜", "고도"]): return "depth_cross_section" if any(k in combined for k in ["음ν–₯ 츑심", "λ“±κ³ μ„ ", "λ°”λ‹₯ 지도", "탐지", "μ†Œλ‚˜", "κ΄΄λ¬Ό νŒμ •", "ν˜„μž₯ 검증", "흔적 검증"]): return "sonar_factcheck_map" if any(k in combined for k in ["μ•„ν‚€ν…μ²˜", "μ‹œμŠ€ν…œ ꡬ쑰", "계측 ꡬ쑰", "데이터 νŒŒμ΄ν”„λΌμΈ", "κ²Œμ΄νŠΈμ›¨μ΄", "λ§ˆμ΄ν¬λ‘œμ„œλΉ„μŠ€", "architecture"]): return "archify_architecture" if any(k in combined for k in ["μž‘λ™ 원리", "λ‚΄λΆ€ 원리", "λ©”μ»€λ‹ˆμ¦˜", "기계", "κΈ°μ–΄", "밸브", "잠금", "ν•΄λΆ€", "anatomy", "mechanism"]): return "anatomy_mechanic" if any(k in combined for k in ["좔이", "μ„±μž₯ 곑선", "νŠΈλ Œλ“œ κ·Έλž˜ν”„", "μ‹œκ³„μ—΄", "λ³€ν™” μΆ”μ„Έ", "예츑 곑선"]): return "anime_svg_chart" if any(k in combined for k in ["μš”κΈˆ", "ν”Œλžœ", "κ°€κ²©ν‘œ", "ν• λ‹ΉλŸ‰", "무료", "유료", "pricing"]) and any("|" in b for b in bullets): return "pricing_matrix" date_bullets = sum(1 for b in bullets if re.match(r"^(?:\d{1,2}\.\d{1,2}|20\d{2}|v\d+\.\d+)", b.strip(), re.I)) if date_bullets >= 2 or any(k in combined for k in ["λ³€κ²½ 기둝", "changelog", "μ—°ν˜", "νžˆμŠ€ν† λ¦¬", "릴리즈 λ…ΈνŠΈ"]): return "changelog_vertical" if sum(1 for b in bullets if "mission" in b.lower() or "λ―Έμ…˜" in b.lower()) >= 2 or "λ―Έμ…˜" in title.lower(): return "mission_staircase" if any(k in combined for k in ["/swarm", "μŠ€μ›œ", "λ©€ν‹° μ—μ΄μ „νŠΈ", "코디넀이터", "μ§€νœ˜μž", "ꡰ체 λ„€νŠΈμ›Œν¬"]): return "swarm_network" if idx == 0 and (re.search(r"\d+", title) or len(bullets) >= 6): return "arc_cards" if idx == total - 1 and total >= 3 and any(k in combined for k in ["μš”μ•½", "정리", "κ²°λ‘ ", "마무리", "summary"]): return "metric_summary" if any(k in combined for k in [" vs ", "비ꡐ", "μ „ν›„", "as-is", "to-be", "차이점", "λŒ€λΉ„"]): return "vs_comparison" numeric_bullets = sum(1 for b in bullets if re.search(r"\d+\s*(%|λ°°|λΆ„|초|μ‹œκ°„|건|개|μ’…|점|만|μ–΅|x)", b.lower())) if numeric_bullets >= 2 or any(k in combined for k in ["μ§€ν‘œ", "μ„±κ³Ό", "톡계", "차트", "κ·Έλž˜ν”„", "데이터 비ꡐ"]): return "data_bars" if any(k in combined for k in ["λͺ…λ Ήμ–΄", "μ½”λ“œ", "cli", "터미널", "μ‹€ν–‰", "command", "python", "μ„€μΉ˜"]): return "terminal_flow" if any(k in combined for k in ["체크리슀트", "λ…ΈνŠΈ", "μ‹€μŠ΅", "μ κ²€ν‘œ", "μ‹€μ²œ 팁", "κ°€μ΄λ“œλΆ"]): return "notebook_flip" if any(k in combined for k in ["원칙", "μ±…", "맀뉴얼", "μ •μ˜", "인용", "핡심 λ¬Έμž₯", "chapter", "법칙", "μ² ν•™"]): return "book_quote" if any(k in combined for k in ["λ‘œλ“œλ§΅", "ν”„λ‘œμ„ΈμŠ€", "절차", "흐름도", "μˆœμ„œ", "νƒ€μž„λΌμΈ"]): return "process_timeline" if len(bullets) == 4 or any(k in combined for k in ["ꡬ성", "κΈ°λŠ₯", "νŠΉμ§•", "λ²€ν† ", "λͺ¨λ“ˆ", "μš”μ†Œ"]): return "bento_grid" if any(k in combined for k in ["μš”μ•½", "정리", "κ²°λ‘ ", "3κ°€μ§€"]): return "metric_summary" if len(bullets) >= 6: return "arc_cards" fallback_cycle = ["bento_grid", "mockup_steps", "book_quote", "process_timeline", "metric_summary"] return fallback_cycle[idx % len(fallback_cycle)] def _parse_notes_into_beats(title: str, bullets: list[str], notes_lines: list[str]) -> list[dict[str, Any]]: beats: list[dict[str, Any]] = [] cleaned_lines = [ln.strip() for ln in notes_lines if ln.strip()] idx = 0 while idx < len(cleaned_lines): ln = cleaned_lines[idx] if "||" in ln: ko_part, en_part = [x.strip() for x in ln.split("||", 1)] beats.append({"narration": ko_part, "sub_ko": ko_part, "sub_en": en_part}) idx += 1 continue # Merge consecutive continuation clauses into ONE continuous narration sentence while keeping explicit visual subtitle cues! merged_cues = [ln] while _is_unfinished_clause(merged_cues[-1]) and (idx + 1) < len(cleaned_lines) and ("||" not in cleaned_lines[idx + 1]): idx += 1 merged_cues.append(cleaned_lines[idx]) if len(merged_cues) > 1: joined_narr = " ".join(merged_cues) beats.append({ "narration": joined_narr, "sub_ko": joined_narr, "sub_en": "", "sub_cues": merged_cues, }) else: for sent in _split_sentences(ln): beats.append({"narration": sent, "sub_ko": sent, "sub_en": ""}) idx += 1 if not beats: intro_sent = title if title.endswith(("λ‹€.", "μš”.", "μ£ .", "?")) else f"{title}에 λŒ€ν•΄ ν•΅μ‹¬λ§Œ μ§šμ–΄λ³΄κ² μŠ΅λ‹ˆλ‹€." beats.append({"narration": intro_sent, "sub_ko": title, "sub_en": ""}) for b in bullets[:3]: display_b = b.replace("|", " Β· ").strip() spoken_b = b.replace("|", ", ").replace("β†’", "μ—μ„œ ").strip() if not spoken_b.endswith(("λ‹€.", "μš”.", "μ£ .", "!", "?")): spoken_b = f"{spoken_b}μž…λ‹ˆλ‹€." beats.append({"narration": spoken_b, "sub_ko": display_b, "sub_en": ""}) return beats def _compile_raw_slides_to_spec(deck_title: str, raw_slides: list[dict[str, Any]]) -> dict[str, Any]: if not raw_slides: raw_slides = [{ "title": deck_title, "bullets": ["핡심 λ‚΄μš© μš”μ•½ | μžλ™ 생성 μŠ¬λΌμ΄λ“œ"], "notes_lines": [f"{deck_title} ν”„λ ˆμ  ν…Œμ΄μ…˜μ„ μ‹œμž‘ν•©λ‹ˆλ‹€."], "meta": {}, }] chapters: list[dict[str, str]] = [] seen_chapters: dict[str, int] = {} for item in raw_slides: raw_ch = item.get("meta", {}).get("CHAPTER") or item.get("chapter") or item["title"] ch_label = _clean_chapter_label(raw_ch) if ch_label not in seen_chapters and len(chapters) < 6: seen_chapters[ch_label] = len(chapters) chapters.append({ "num": f"{len(chapters) + 1:02d}", "label": ch_label, }) slides_spec: list[dict[str, Any]] = [] for s_idx, item in enumerate(raw_slides): meta = item.get("meta", {}) raw_ch = meta.get("CHAPTER") or item.get("chapter") or item["title"] ch_label = _clean_chapter_label(raw_ch) ch_idx = seen_chapters.get(ch_label, min(s_idx, max(len(chapters) - 1, 0))) bullets = item.get("bullets") or [item["title"]] notes_lines = item.get("notes_lines", []) broll_src = meta.get("BROLL") or "" media_src = broll_src or meta.get("MEDIA") or item.get("media_src", "") stamp_text = meta.get("STAMP") or "" custom_svg = meta.get("CUSTOM_SVG") or item.get("custom_svg", "") custom_html = meta.get("CUSTOM_HTML") or item.get("custom_html", "") archify_json = meta.get("ARCHIFY_JSON") or item.get("archify_json", "") custom_anime_js = item.get("custom_anime_js", "") has_custom_viz = bool(custom_svg or custom_html or archify_json) archetype = meta.get("ARCHETYPE") or _infer_archetype( s_idx, len(raw_slides), item["title"], bullets, " ".join(notes_lines), has_media=bool(media_src) and not bool(broll_src), has_stamp=bool(stamp_text), has_broll=bool(broll_src), has_custom_viz=has_custom_viz, ) beats = _parse_notes_into_beats(item["title"], bullets, notes_lines) num_match = re.search(r"^(.*?)\s*(\d+)\s*([^\d]*)$", item["title"]) stat_prefix = meta.get("STAT_PREFIX", item["title"][:16]) stat_number = ( int(meta["STAT_NUM"]) if "STAT_NUM" in meta and str(meta["STAT_NUM"]).isdigit() else (int(num_match.group(2)) if num_match else len(bullets)) ) stat_suffix = meta.get("STAT_SUFFIX", num_match.group(3).strip()[:6] if num_match else "핡심") if num_match and "STAT_PREFIX" not in meta and num_match.group(1).strip(): stat_prefix = num_match.group(1).strip() pronounce_tag = meta.get("PRONOUNCE") or meta.get("LEXICON") or meta.get("GLOSSARY") or "" slide_pronounce: dict[str, str] = {} if pronounce_tag: from deck_video_pipeline.pronunciation_engine import parse_inline_pronounce_tag slide_pronounce.update(parse_inline_pronounce_tag(pronounce_tag)) if isinstance(item.get("pronounce"), dict): for pk, pv in item["pronounce"].items(): if pk and pv: slide_pronounce[str(pk)] = str(pv) slides_spec.append({ "id": f"slide_{s_idx + 1:02d}", "chapter_index": ch_idx, "archetype": archetype, "bg_preset": meta.get("BG", ""), "lang": meta.get("LANG", ""), "pronounce_map": slide_pronounce, "badge": meta.get("BADGE", f"SECTION {s_idx + 1:02d}"), "title": item["title"], "subtitle": meta.get("SUBTITLE", ""), "hero_word": meta.get("HERO_WORD", ""), "scramble_text": meta.get("SCRAMBLE", ""), "source_note": meta.get("SOURCE", ""), "stamp_text": stamp_text, "custom_svg": custom_svg, "custom_html": custom_html, "archify_json": archify_json, "custom_anime_js": custom_anime_js, "media_src": media_src, "media_caption": meta.get("MEDIA_CAPTION", ""), "plans": meta.get("PLANS", "무료 β‚©0 | Pro μ›” β‚©29,000 | Ultra μ›” β‚©119,000~"), "stat_prefix": stat_prefix, "stat_number": stat_number, "stat_suffix": stat_suffix, "bullets": bullets, "highlight_quote": meta.get("QUOTE") or item.get("quote") or (bullets[0] if bullets else item["title"]), "command": meta.get("COMMAND", "deck2video --input document.pptx --sfx on"), "flow_left": meta.get("FLOW_LEFT", "원본 κ΅μ•ˆ λ¬Έμ„œ"), "flow_left_sub": meta.get("FLOW_LEFT_SUB", "PPTX Β· DOCX Β· MD Β· HTML"), "flow_right": meta.get("FLOW_RIGHT", "1080p λͺ¨μ…˜ MP4"), "flow_right_sub": meta.get("FLOW_RIGHT_SUB", "HTML λ Œλ”λ§ + 효과음 싱크"), "beats": beats, }) return { "title": deck_title, "voice": "ko-KR-InJoonNeural", "voice_rate": "+6%", "bgm_mood": "focused", "chapters": chapters, "slides": slides_spec, } # --------------------------------------------------------------------------- # 1. PowerPoint (.pptx) Parser (with GroupShape recursion & Logo Filter) # --------------------------------------------------------------------------- def parse_pptx_to_spec(pptx_path: Path, extract_images: bool = True) -> dict[str, Any]: check_openxml_drm(pptx_path) from pptx import Presentation from pptx.enum.shapes import MSO_SHAPE_TYPE try: prs = Presentation(str(pptx_path)) except Exception as exc: raise DRMProtectedFileError(pptx_path, f"python-pptx open failed: {exc}") from exc slide_w = float(prs.slide_width or 9144000) slide_h = float(prs.slide_height or 5143500) raw_slides: list[dict[str, Any]] = [] extracted_dir = pptx_path.parent / f".{pptx_path.stem}_extracted_media" for s_idx, slide in enumerate(prs.slides): title = "" title_shape = slide.shapes.title if title_shape and title_shape.text.strip(): title = title_shape.text.strip().replace("\n", " ") bullets: list[str] = [] meta_tags: dict[str, str] = {} candidate_images: list[tuple[float, bool, Any, int]] = [] shape_counter = 0 def _walk_shapes(shapes_iter: Any) -> None: nonlocal shape_counter for shape in shapes_iter: shape_counter += 1 if shape == title_shape: continue if shape.shape_type == MSO_SHAPE_TYPE.GROUP and hasattr(shape, "shapes"): _walk_shapes(shape.shapes) continue if extract_images and shape.shape_type == MSO_SHAPE_TYPE.PICTURE: w_ratio = float(getattr(shape, "width", 0) or 0) / slide_w h_ratio = float(getattr(shape, "height", 0) or 0) / slide_h area_ratio = w_ratio * h_ratio is_full_wallpaper = (w_ratio >= 0.96 and h_ratio >= 0.96) # Ignore tiny corner corporate logos / header icons (< 16% width or height) if w_ratio >= 0.16 and h_ratio >= 0.16 and area_ratio >= 0.04: candidate_images.append((area_ratio, is_full_wallpaper, shape, shape_counter)) if getattr(shape, "has_text_frame", False): for para in shape.text_frame.paragraphs: line = para.text.strip() if not line: continue m_tag = re.match(r"^\[([A-Z_]+):\s*(.+)\]$", line) if m_tag: meta_tags[m_tag.group(1).upper()] = m_tag.group(2).strip() else: bullets.append(line) elif getattr(shape, "has_table", False): for r_idx, row in enumerate(shape.table.rows): cells = [c.text.strip().replace("\n", " ") for c in row.cells if c.text.strip()] if not cells: continue if r_idx == 0 and len(cells) >= 3 and "PLANS" not in meta_tags and any( k in " ".join(cells).lower() for k in ["무료", "유료", "pro", "λͺ¨λ“œ", "방식", "ꡬ뢄", "μ˜΅μ…˜", "비ꡐ"] ): meta_tags["PLANS"] = " | ".join(cells[1:4]) continue bullets.append(" | ".join(cells)) _walk_shapes(slide.shapes) extracted_img_path = "" # Pick the most prominent content chart/screenshot while ignoring full-slide template wallpaper if text exists filtered_imgs = [ item for item in candidate_images if not (item[1] and len(bullets) >= 2) ] if filtered_imgs: filtered_imgs.sort(key=lambda x: x[0], reverse=True) _, _, best_shape, best_idx = filtered_imgs[0] try: extracted_dir.mkdir(parents=True, exist_ok=True) ext = best_shape.image.ext or "png" img_file = extracted_dir / f"slide_{s_idx + 1:02d}_img_{best_idx}.{ext}" img_file.write_bytes(best_shape.image.blob) extracted_img_path = str(img_file.resolve()) except Exception: pass if not title and bullets: title = bullets.pop(0) if not title: title = f"μŠ¬λΌμ΄λ“œ {s_idx + 1}" notes_lines: list[str] = [] if slide.has_notes_slide and slide.notes_slide.notes_text_frame: raw_notes = slide.notes_slide.notes_text_frame.text.strip() notes_lines = [ln.strip() for ln in raw_notes.splitlines() if ln.strip()] raw_slides.append({ "title": title, "bullets": bullets, "notes_lines": notes_lines, "media_src": extracted_img_path, "meta": meta_tags, }) deck_title = raw_slides[0]["title"] if raw_slides else pptx_path.stem return _compile_raw_slides_to_spec(deck_title, raw_slides) # --------------------------------------------------------------------------- # 2. Microsoft Word (.docx) Parser # --------------------------------------------------------------------------- def parse_docx_to_spec(docx_path: Path) -> dict[str, Any]: check_openxml_drm(docx_path) from docx import Document from docx.oxml.table import CT_Tbl from docx.oxml.text.paragraph import CT_P from docx.table import Table from docx.text.paragraph import Paragraph try: doc = Document(str(docx_path)) except Exception as exc: raise DRMProtectedFileError(docx_path, f"python-docx open failed: {exc}") from exc deck_title = docx_path.stem current_chapter = "" raw_slides: list[dict[str, Any]] = [] cur_slide: dict[str, Any] | None = None def flush_slide() -> None: nonlocal cur_slide if cur_slide and (cur_slide["title"] or cur_slide["bullets"] or cur_slide["notes_lines"]): if not cur_slide["title"]: cur_slide["title"] = cur_slide["bullets"].pop(0) if cur_slide["bullets"] else f"μ„Ήμ…˜ {len(raw_slides) + 1}" raw_slides.append(cur_slide) cur_slide = None for child in doc.element.body.iterchildren(): if isinstance(child, CT_P): para = Paragraph(child, doc) text = para.text.strip() if not text: continue style_name = (para.style.name if para.style else "").lower() if "title" in style_name and not raw_slides and cur_slide is None: deck_title = text continue if style_name.startswith("heading 1"): flush_slide() current_chapter = text cur_slide = { "title": text, "chapter": current_chapter, "bullets": [], "notes_lines": [], "quote": "", "meta": {}, } continue if style_name.startswith(("heading 2", "heading 3")) or re.match(r"^(?:\d+\.|Step\s*\d+|μŠ¬λΌμ΄λ“œ\s*\d+)\s+\S+", text, re.I): # If cur_slide was just created from Heading 1 and still empty, adopt this Heading 2 as its title if cur_slide and not cur_slide["bullets"] and not cur_slide["notes_lines"] and cur_slide["title"] == current_chapter: cur_slide["title"] = text else: flush_slide() cur_slide = { "title": text, "chapter": current_chapter or text, "bullets": [], "notes_lines": [], "quote": "", "meta": {}, } continue if cur_slide is None: cur_slide = { "title": text if len(text) <= 36 else deck_title, "chapter": current_chapter or deck_title, "bullets": [] if len(text) <= 36 else [text], "notes_lines": [], "quote": "", "meta": {}, } continue m_tag = re.match(r"^\[([A-Z_]+):\s*(.+)\]$", text) if m_tag: cur_slide["meta"][m_tag.group(1).upper()] = m_tag.group(2).strip() elif "quote" in style_name or text.startswith((">", "πŸ’¬", "[λ°œν‘œλ…ΈνŠΈ]", "[λ‚΄λ ˆμ΄μ…˜]", "λ…ΈνŠΈ:")): cleaned_note = re.sub(r"^(?:>|πŸ’¬|\[λ°œν‘œλ…ΈνŠΈ\]|\[λ‚΄λ ˆμ΄μ…˜\]|λ…ΈνŠΈ:)\s*", "", text).strip() if "quote" in style_name and not cur_slide["quote"]: cur_slide["quote"] = cleaned_note cur_slide["notes_lines"].append(cleaned_note) elif "list" in style_name or text.startswith(("- ", "β€’ ", "* ", "β–ͺ ")): cur_slide["bullets"].append(re.sub(r"^[-β€’*β–ͺ]\s*", "", text).strip()) else: # Long prose sentence becomes spoken narration; concise phrase becomes bullet if len(text) > 52 and text.endswith(("λ‹€.", "μš”.", "μ£ .", "!", "?")): cur_slide["notes_lines"].append(text) else: cur_slide["bullets"].append(text) # Auto-chunk if a flat Word doc has too many bullets in one section if len(cur_slide["bullets"]) >= 7 and not current_chapter: flush_slide() elif isinstance(child, CT_Tbl): tbl = Table(child, doc) if cur_slide is None: cur_slide = { "title": deck_title, "chapter": current_chapter or deck_title, "bullets": [], "notes_lines": [], "quote": "", "meta": {}, } for r_i, row in enumerate(tbl.rows): cells = [c.text.strip().replace("\n", " ") for c in row.cells if c.text.strip()] if not cells: continue if r_i == 0 and any(h in cells[0].lower() for h in ["ꡬ뢄", "ν•­λͺ©", "단계", "λͺ¨λ“ˆ", "category", "step"]): continue cur_slide["bullets"].append(" | ".join(cells[:2]) if len(cells) >= 2 else cells[0]) flush_slide() if raw_slides and deck_title == docx_path.stem: deck_title = raw_slides[0]["title"] return _compile_raw_slides_to_spec(deck_title, raw_slides) # --------------------------------------------------------------------------- # 3. Markdown / Plain Text (.md / .txt / stdin) Parser # --------------------------------------------------------------------------- def parse_markdown_text_to_spec(text: str, default_title: str = "Presentation") -> dict[str, Any]: lines = text.splitlines() h1_count = sum(1 for l in lines if l.strip().startswith("# ") and not l.strip().startswith("## ")) deck_title = default_title current_chapter = "" raw_slides: list[dict[str, Any]] = [] cur_slide: dict[str, Any] | None = None def flush_slide() -> None: nonlocal cur_slide if cur_slide and (cur_slide["title"] or cur_slide["bullets"] or cur_slide["notes_lines"]): raw_slides.append(cur_slide) cur_slide = None for raw_ln in lines: ln = raw_ln.strip() if not ln or ln.startswith("---") or re.match(r"^\|[\s\-:|]+\|$", ln): continue if ln.startswith("# ") and not ln.startswith("## "): h1 = ln[2:].strip() if not raw_slides and cur_slide is None and deck_title == default_title: deck_title = h1 if h1_count > 1: current_chapter = h1 else: flush_slide() current_chapter = h1 continue if ln.startswith(("## ", "### ")): flush_slide() title = ln.lstrip("#").strip() cur_slide = { "title": title, "chapter": current_chapter or title, "bullets": [], "notes_lines": [], "quote": "", "meta": {}, } continue if cur_slide is None: cur_slide = { "title": deck_title, "chapter": current_chapter or deck_title, "bullets": [], "notes_lines": [], "quote": "", "meta": {}, } m_tag = re.match(r"^\[([A-Z_]+):\s*(.+)\]$", ln) if m_tag: cur_slide["meta"][m_tag.group(1).upper()] = m_tag.group(2).strip() elif ln.startswith(">"): cur_slide["notes_lines"].append(ln.lstrip("> ").strip()) elif ln.startswith(("- ", "* ", "β€’ ")): cur_slide["bullets"].append(ln[2:].strip()) elif ln.startswith("|") and ln.endswith("|"): cells = [c.strip() for c in ln.strip("|").split("|") if c.strip()] if cells and cells[0].lower() not in ("ꡬ뢄", "ν•­λͺ©", "μ˜΅μ…˜", "방식", "κΈ°λŠ₯"): cur_slide["bullets"].append(" | ".join(cells)) elif len(cells) >= 3 and "PLANS" not in cur_slide["meta"]: cur_slide["meta"]["PLANS"] = " | ".join(cells[1:4]) else: if len(ln) > 50 and ln.endswith(("λ‹€.", "μš”.", "μ£ .", "!", "?")): cur_slide["notes_lines"].append(ln) else: cur_slide["bullets"].append(ln) flush_slide() return _compile_raw_slides_to_spec(deck_title, raw_slides) def parse_markdown_to_spec(md_path: Path) -> dict[str, Any]: return parse_markdown_text_to_spec(md_path.read_text(encoding="utf-8-sig"), default_title=md_path.stem) # --------------------------------------------------------------------------- # 4. HTML Presentation / Web Article (.html / .htm) Parser # --------------------------------------------------------------------------- def parse_html_to_spec(html_path: Path) -> dict[str, Any]: from bs4 import BeautifulSoup soup = BeautifulSoup(html_path.read_text(encoding="utf-8"), "html.parser") for tag in soup(["script", "style", "noscript", "svg", "nav", "footer"]): tag.decompose() deck_title = (soup.title.string.strip() if soup.title and soup.title.string else "") or html_path.stem h1_first = soup.find("h1") if h1_first and h1_first.get_text(strip=True): deck_title = h1_first.get_text(" ", strip=True) raw_slides: list[dict[str, Any]] = [] # Case A: Slide-based HTML (<section>, .slide, [data-slide]) slide_nodes = soup.select("section, div.slide, article.slide, [data-slide]") # Filter out outer wrapper sections that contain nested <section> slides (like Reveal.js vertical stacks) leaf_slides = [ node for node in slide_nodes if not node.find(["section"]) and node.get_text(strip=True) ] if leaf_slides: for idx, node in enumerate(leaf_slides): heading = node.find(["h1", "h2", "h3", "h4"]) title = heading.get_text(" ", strip=True) if heading else f"μŠ¬λΌμ΄λ“œ {idx + 1}" chapter = node.get("data-chapter") or title notes_lines: list[str] = [] if node.get("data-notes"): notes_lines.append(str(node["data-notes"]).strip()) for aside in node.select("aside.notes, .speaker-notes, [data-narration]"): nt = aside.get_text("\n", strip=True) if nt: notes_lines.extend([x.strip() for x in nt.splitlines() if x.strip()]) aside.decompose() quote_txt = "" for bq in node.find_all("blockquote"): qt = bq.get_text(" ", strip=True) if qt and not quote_txt: quote_txt = qt bq.decompose() bullets: list[str] = [] for li in node.find_all("li"): t = li.get_text(" ", strip=True) if t: bullets.append(t) for tr in node.find_all("tr"): cells = [td.get_text(" ", strip=True) for td in tr.find_all(["td", "th"]) if td.get_text(strip=True)] if len(cells) >= 2 and tr.find("td"): bullets.append(" | ".join(cells[:2])) if not bullets: for p in node.find_all("p"): pt = p.get_text(" ", strip=True) if pt and pt != title: if len(pt) > 55: notes_lines.append(pt) else: bullets.append(pt) meta = {} if node.get("data-archetype"): meta["ARCHETYPE"] = str(node["data-archetype"]).strip() if node.get("data-badge"): meta["BADGE"] = str(node["data-badge"]).strip() if node.get("data-command"): meta["COMMAND"] = str(node["data-command"]).strip() raw_slides.append({ "title": title, "chapter": chapter, "bullets": bullets or [title], "notes_lines": notes_lines, "quote": quote_txt, "meta": meta, }) return _compile_raw_slides_to_spec(deck_title, raw_slides) # Case B: General Web Document / Article HTML (<h2>/<h3> sections) current_chapter = deck_title cur_slide: dict[str, Any] | None = None def flush_html_slide() -> None: nonlocal cur_slide if cur_slide and (cur_slide["bullets"] or cur_slide["notes_lines"] or cur_slide["title"]): raw_slides.append(cur_slide) cur_slide = None body = soup.body or soup for el in body.find_all(["h1", "h2", "h3", "li", "p", "blockquote", "pre", "code"], recursive=True): txt = el.get_text(" ", strip=True) if not txt: continue if el.name == "h1": current_chapter = txt elif el.name in ("h2", "h3"): flush_html_slide() cur_slide = { "title": txt, "chapter": current_chapter, "bullets": [], "notes_lines": [], "quote": "", "meta": {}, } else: if cur_slide is None: cur_slide = { "title": deck_title, "chapter": current_chapter, "bullets": [], "notes_lines": [], "quote": "", "meta": {}, } if el.name == "blockquote": cur_slide["quote"] = cur_slide["quote"] or txt cur_slide["notes_lines"].append(txt) elif el.name in ("pre", "code") and len(txt) < 120: cur_slide["meta"]["COMMAND"] = txt.splitlines()[0] cur_slide["meta"].setdefault("ARCHETYPE", "terminal_flow") elif el.name == "li": if txt not in cur_slide["bullets"]: cur_slide["bullets"].append(txt) elif el.name == "p" and not el.find_parent(["li", "blockquote"]): if len(txt) > 50: cur_slide["notes_lines"].append(txt) elif txt not in cur_slide["bullets"]: cur_slide["bullets"].append(txt) flush_html_slide() return _compile_raw_slides_to_spec(deck_title, raw_slides) # --------------------------------------------------------------------------- # 5. Agent Handoff / DRM Bridge JSON Parser (.json / --stdin-json) # --------------------------------------------------------------------------- def parse_agent_json_to_spec(payload: Any, default_title: str = "Presentation") -> dict[str, Any]: """ Accepts EITHER: 1. Already-compiled DeckSpec JSON (contains 'chapters' + slides with 'beats' & 'archetype') 2. Simplified Agent-Extracted DRM JSON: {"title": "...", "slides": [{"title": "...", "bullets": [...], "table": [[...]], "notes": "..."}]} or even a bare list of slide dicts: [{"title": "...", "bullets": [...], "notes": "..."}] """ if isinstance(payload, dict) and "chapters" in payload and isinstance(payload.get("slides"), list): slides_check = payload["slides"] if slides_check and all(isinstance(s, dict) and "beats" in s and "archetype" in s for s in slides_check): return payload if isinstance(payload, list): deck_title = default_title slides_in = payload top_voice = "" top_rate = "" top_lang = "" top_glossary: dict[str, str] = {} elif isinstance(payload, dict): deck_title = str(payload.get("title") or payload.get("deck_title") or default_title).strip() slides_in = payload.get("slides") or payload.get("pages") or payload.get("sections") or [] top_voice = str(payload.get("voice") or "").strip() top_rate = str(payload.get("voice_rate") or payload.get("rate") or "").strip() top_lang = str(payload.get("lang") or payload.get("language") or "").strip() top_glossary = payload.get("glossary") if isinstance(payload.get("glossary"), dict) else {} else: raise ValueError("JSON input must be a JSON object {'title': ..., 'slides': [...]} or a list of slide objects.") raw_slides: list[dict[str, Any]] = [] for idx, s in enumerate(slides_in): if not isinstance(s, dict): continue title = str(s.get("title") or s.get("heading") or s.get("name") or f"μŠ¬λΌμ΄λ“œ {idx + 1}").strip() chapter = str(s.get("chapter") or s.get("section") or title).strip() meta: dict[str, str] = {} if isinstance(s.get("meta"), dict): for mk, mv in s["meta"].items(): if mv is not None and str(mv).strip(): meta[str(mk).upper()] = str(mv).strip() key_map = { "archetype": "ARCHETYPE", "bg": "BG", "bg_preset": "BG", "badge": "BADGE", "subtitle": "SUBTITLE", "hero_word": "HERO_WORD", "scramble": "SCRAMBLE", "scramble_text": "SCRAMBLE", "source": "SOURCE", "source_note": "SOURCE", "media_caption": "MEDIA_CAPTION", "plans": "PLANS", "command": "COMMAND", "stamp": "STAMP", "stamp_text": "STAMP", "broll": "BROLL", } for src_k, dst_k in key_map.items(): val = s.get(src_k) if val and str(val).strip() and str(val).strip().lower() != "auto": meta[dst_k] = str(val).strip() raw_bullets = s.get("bullets") or s.get("items") or s.get("points") or s.get("body") or [] if isinstance(raw_bullets, str): bullets = [re.sub(r"^[-β€’*β–ͺ]\s*", "", ln.strip()) for ln in raw_bullets.splitlines() if ln.strip()] elif isinstance(raw_bullets, list): bullets = [str(b).strip() for b in raw_bullets if str(b).strip()] else: bullets = [] tbl = s.get("table") or s.get("rows") if isinstance(tbl, list) and tbl: for r_i, row in enumerate(tbl): if isinstance(row, list): cells = [str(c).strip().replace("\n", " ") for c in row if str(c).strip()] elif isinstance(row, str): cells = [c.strip() for c in row.strip("|").split("|") if c.strip()] else: cells = [] if not cells: continue if r_i == 0 and len(cells) >= 3 and "PLANS" not in meta and any( h in cells[0].lower() for h in ["ꡬ뢄", "ν•­λͺ©", "비ꡐ", "κΈ°λŠ₯", "μ˜΅μ…˜", "방식", "category", "feature"] ): meta["PLANS"] = " | ".join(cells[1:4]) continue bullets.append(" | ".join(cells)) raw_notes = ( s.get("notes") or s.get("notes_lines") or s.get("narration") or s.get("speaker_notes") or s.get("script") or [] ) if isinstance(raw_notes, str): notes_lines = [ln.strip() for ln in raw_notes.splitlines() if ln.strip()] elif isinstance(raw_notes, list): notes_lines = [str(n).strip() for n in raw_notes if str(n).strip()] else: notes_lines = [] # Also support pre-structured beats [{"narration": "...", "sub_ko": "...", "sub_en": "..."}] if not notes_lines and isinstance(s.get("beats"), list): for b_obj in s["beats"]: if isinstance(b_obj, dict) and b_obj.get("narration"): narr = str(b_obj["narration"]).strip() en = str(b_obj.get("sub_en") or "").strip() notes_lines.append(f"{narr} || {en}" if en else narr) media_src = str(s.get("media_src") or s.get("media") or s.get("image") or "").strip() quote = str(s.get("quote") or s.get("highlight_quote") or "").strip() raw_slides.append({ "title": title, "chapter": chapter, "bullets": bullets, "notes_lines": notes_lines, "media_src": media_src, "quote": quote, "custom_svg": str(s.get("custom_svg") or "").strip(), "custom_html": str(s.get("custom_html") or "").strip(), "archify_json": str(s.get("archify_json") or "").strip(), "custom_anime_js": str(s.get("custom_anime_js") or "").strip(), "pronounce": s.get("pronounce") if isinstance(s.get("pronounce"), dict) else {}, "meta": meta, }) if raw_slides and deck_title == default_title and isinstance(payload, list): deck_title = raw_slides[0]["title"] compiled = _compile_raw_slides_to_spec(deck_title, raw_slides) if top_voice: compiled["voice"] = top_voice if top_rate: compiled["voice_rate"] = top_rate if top_lang: compiled["lang"] = top_lang if top_glossary: compiled["glossary"] = top_glossary return compiled # --------------------------------------------------------------------------- # Unified Dispatcher # --------------------------------------------------------------------------- def load_input_to_spec(input_path: Path, extract_ppt_images: bool = True) -> dict[str, Any]: suffix = input_path.suffix.lower() if suffix == ".json": raw_text = input_path.read_text(encoding="utf-8-sig") return parse_agent_json_to_spec(json.loads(raw_text), default_title=input_path.stem) if suffix == ".pptx": return parse_pptx_to_spec(input_path, extract_images=extract_ppt_images) if suffix == ".docx": return parse_docx_to_spec(input_path) if suffix in (".md", ".markdown", ".txt"): return parse_markdown_to_spec(input_path) if suffix in (".html", ".htm"): return parse_html_to_spec(input_path) raise ValueError( f"Unsupported input format '{suffix}'. Supported: .pptx, .docx, .md, .txt, .html, .htm, .json" ) # --------------------------------------------------------------------------- # Pre-Production Deck Analyzer & Style Recommender (--analyze) # --------------------------------------------------------------------------- def analyze_deck_spec(spec: dict[str, Any], lang: str = "auto", glossary_path: str = "") -> dict[str, Any]: from deck_video_pipeline.themes_and_presets import STYLE_PACKS from deck_video_pipeline.pronunciation_engine import ( LANGUAGE_REGISTRY, load_custom_glossary, prepare_display_and_tts_text, resolve_language_code, ) slides = spec.get("slides", []) full_blob = " ".join( f"{s.get('title', '')} {' '.join(s.get('bullets', []))} {' '.join(b.get('narration', '') for b in s.get('beats', []))}" for s in slides ) resolved_lang = resolve_language_code(lang if lang and lang != "auto" else spec.get("lang", ""), full_blob) base_glossary = load_custom_glossary(glossary_path or None, lang=resolved_lang) if isinstance(spec.get("glossary"), dict): base_glossary.update({str(k): str(v) for k, v in spec["glossary"].items()}) full_blob = full_blob.lower() # Score style packs dynamically so every deck receives context-matched recommendations scores: dict[str, int] = {k: 10 for k in STYLE_PACKS} if any(k in full_blob for k in ["보고", "싀적", "μ˜ˆμ‚°", "λΆ„κΈ°", "μž„μ›", "경영", "곡곡", "μ„±κ³Ό", "λŒ€ν•™"]): scores["executive_light"] += 35 scores["apple_keynote"] += 15 if any(k in full_blob for k in ["μ—…λ°μ΄νŠΈ", "릴리즈", "버전", "λͺ¨λΈ", "κΈ°λŠ₯", "μΆœμ‹œ", "api", "μ½”λ”©", "μ—μ΄μ „νŠΈ"]): scores["tech_deepdive"] += 32 scores["interactive_creator"] += 24 if any(k in full_blob for k in ["νŠΉν—ˆ", "λ…Όλž€", "μ‚¬νšŒ", "역사", "인간", "미래", "사건", "μŠ€ν† λ¦¬", "μ§„μ‹€", "μœ„ν—˜"]): scores["docu_storytelling"] += 38 if any(k in full_blob for k in ["μˆ˜μ‹¬", "탐사", "κ³Όν•™", "μ§€μ§ˆ", "ν™”μ‚°", "팩트", "검증", "츑심", "λ―ΈμŠ€ν„°λ¦¬", "기둝", "λ°”λ‹₯"]): if "deepfact_docu" in scores: scores["deepfact_docu"] += 42 if any(k in full_blob for k in ["μ•„ν‚€ν…μ²˜", "섀계", "νŒŒμ΄ν”„λΌμΈ", "원리", "인프라", "μ„œλ²„", "ν”„λ‘œν† μ½œ", "λ³΄μ•ˆ", "ꡬ쑰"]): scores["engineering_blueprint"] += 36 scores["tech_deepdive"] += 18 if any(k in full_blob for k in ["ꡐ윑", "κ°€μ΄λ“œ", "νŠœν† λ¦¬μ–Ό", "μ‹€μŠ΅", "μž…λ¬Έ", "μ˜¨λ³΄λ”©", "팁"]): scores["interactive_creator"] += 30 scores["executive_light"] += 14 if len(slides) <= 5 and sum(len(s.get("bullets", [])) for s in slides) <= 15: scores["apple_keynote"] += 22 ranked_packs = sorted(STYLE_PACKS.items(), key=lambda kv: scores[kv[0]], reverse=True) slide_plans: list[dict[str, Any]] = [] viz_candidates: list[dict[str, Any]] = [] total_chars = 0 for idx, s in enumerate(slides): s_txt = f"{s.get('title', '')} {' '.join(s.get('bullets', []))}".lower() narr_txt = " ".join(b.get("narration", "") for b in s.get("beats", [])) total_chars += len(narr_txt) viz_suggestion = "" if any(k in s_txt for k in ["μ•„ν‚€ν…μ²˜", "ꡬ쑰", "νŒŒμ΄ν”„λΌμΈ", "연동", "흐름", "λ ˆμ΄μ–΄", "λ„€νŠΈμ›Œν¬", "μ‹œμŠ€ν…œ"]): viz_suggestion = "Archify μ‹œμŠ€ν…œ μ•„ν‚€ν…μ²˜ λ‹€μ΄μ–΄κ·Έλž¨ (viz_skills/archify λ˜λŠ” archify_architecture)" elif any(k in s_txt for k in ["원리", "μž‘λ™", "λ©”μ»€λ‹ˆμ¦˜", "μ—”μ§„", "검증", "잠금", "ν•„ν„°", "λ‚΄λΆ€"]): viz_suggestion = "Anatomy μ•„μ΄μ†Œλ©”νŠΈλ¦­ 기계 μž‘λ™ 도식 (viz_skills/anatomy λ˜λŠ” anatomy_mechanic)" elif any(k in s_txt for k in ["좔이", "μ„±μž₯", "μ§€ν‘œ", "%", "λ°°", "톡계", "비ꡐ", "단좕"]): viz_suggestion = "Anime.js μ‹€μ‹œκ°„ λ“œλ‘œμž‰ νŠΈλ Œλ“œ 차트 (anime_svg_chart)" if viz_suggestion: viz_candidates.append({ "slide_num": idx + 1, "title": s.get("title", ""), "suggested_tool": viz_suggestion, }) slide_gloss = dict(base_glossary) if isinstance(s.get("pronounce_map"), dict): slide_gloss.update(s["pronounce_map"]) s_lang = resolve_language_code(s.get("lang") or resolved_lang, narr_txt) beat_pairs = [] for b in s.get("beats", []): disp_b, tts_b = prepare_display_and_tts_text(b.get("narration", ""), lang=s_lang, custom_glossary=slide_gloss) beat_pairs.append({"display": disp_b, "tts_phonetic": tts_b}) slide_plans.append({ "slide_num": idx + 1, "title": s.get("title", ""), "bullets": s.get("bullets", []), "bullets_count": len(s.get("bullets", [])), "default_archetype": s.get("archetype", ""), "has_media": bool(s.get("media_src")), "viz_upgrade_hint": viz_suggestion, "full_narration_beats": [bp["display"] for bp in beat_pairs], "tts_phonetic_beats": [bp["tts_phonetic"] for bp in beat_pairs], "narration_preview": narr_txt, }) est_sec = max(12, int(round(total_chars * 0.16 + len(slides) * 1.2))) lang_info = LANGUAGE_REGISTRY.get(resolved_lang, LANGUAGE_REGISTRY["ko"]) return { "title": spec.get("title", "Presentation"), "lang_code": resolved_lang, "lang_label": f"{lang_info['name_ko']} ({lang_info['name_en']})", "slide_count": len(slides), "estimated_duration_sec": est_sec, "recommended_style_packs": [ {"id": k, "score": scores[k], **v} for k, v in ranked_packs[:4] ], "viz_candidates": viz_candidates, "slides": slide_plans, } def format_analysis_report_md(analysis: dict[str, Any]) -> str: lines = [ f"# 🎬 ν”„λ¦¬ν”„λ‘œλ•μ…˜ 자료 뢄석 & λŒ€λ³Έ/발음 기획 리포트: {analysis['title']}", f"- **총 μŠ¬λΌμ΄λ“œ 수**: {analysis['slide_count']}μž₯ | **μ˜ˆμƒ μ˜μƒ 길이**: μ•½ {analysis['estimated_duration_sec']}초 | **감지/μ§€μ • μ–Έμ–΄**: `{analysis.get('lang_code', 'ko')}` ({analysis.get('lang_label', 'ν•œκ΅­μ–΄')})", "", "## 1. λ¬Έμ„œ λ‚΄μš© 기반 λ§žμΆ€ν˜• μΆ”μ²œ μŠ€νƒ€μΌ 팩 (Top 4)", ] for rank, sp in enumerate(analysis["recommended_style_packs"], 1): lines.append( f"{rank}. **`--style-pack {sp['id']}` ({sp['label']})**\n" f" - ꡬ성: λ°°κ²½=`{sp['bg']}`, 상단HUD=`{sp['hud_style']}`, μžλ§‰=`{sp['caption_style']}`, μΆ”μ²œμ„±μš°=`{sp['voice']}`\n" f" - νŠΉμ§•: {sp['desc']}" ) lines.append("\n## 2. μŠ¬λΌμ΄λ“œλ³„ ν™”λ©΄ ꡬ성 & 전체 λŒ€μ‚¬/발음 λ³€ν™˜ μ΄ˆμ•ˆ (μ‚¬μš©μž κ²€μˆ˜μš© μ „λ¬Έ)") for sp in analysis["slides"]: viz_note = f" βž” πŸ’‘ **[μ‹œκ°ν™” μΆ”μ²œ: {sp['viz_upgrade_hint']}]**" if sp["viz_upgrade_hint"] else "" lines.append( f"- **Slide {sp['slide_num']:02d} [{sp['default_archetype']}]** `{sp['title']}` (ν™”λ©΄ ν•­λͺ© {sp['bullets_count']}개){viz_note}" ) disp_list = sp.get("full_narration_beats", []) phon_list = sp.get("tts_phonetic_beats", []) for b_i, b_line in enumerate(disp_list, 1): lines.append(f" - [μžλ§‰ ν‘œκΈ° {b_i}] \"{b_line}\"") if b_i <= len(phon_list) and phon_list[b_i - 1] != b_line: lines.append(f" ↳ πŸ”Š [μ„±μš° μ‹€μ œ 발음 {b_i}] \"{phon_list[b_i - 1]}\"") lines.append( "\n## 3. [AI μ—μ΄μ „νŠΈ ν•„μˆ˜ μ•‘μ…˜] λ Œλ”λ§ μ „ μ‚¬μš©μž 확인 & 인터뷰 체크리슀트\n" "μœ„ 뢄석 κ²°κ³Όλ₯Ό λ°”νƒ•μœΌλ‘œ **μ ˆλŒ€ λ°”λ‘œ μ˜μƒμ„ λ Œλ”λ§ν•˜μ§€ 말고**, μ‚¬μš©μžμ—κ²Œ μ•„λž˜ 4κ°€μ§€λ₯Ό λ¨Όμ € ν™•μΈλ°›μœΌμ„Έμš”:\n" "1. **μ˜μƒ μŠ€νƒ€μΌ & λΆ„μœ„κΈ°**: μΆ”μ²œ μŠ€νƒ€μΌ 팩(1~4번) 쀑 μ–΄λ–€ λ””μžμΈμœΌλ‘œ μ œμž‘ν• κΉŒμš”?\n" "2. **λ‚΄λ ˆμ΄μ…˜(μŒμ„±) & μžλ§‰ ꡬ성 λͺ¨λ“œ**:\n" " - β‘  μ„±μš° λ‚΄λ ˆμ΄μ…˜ + ν•˜λ‹¨ μžλ§‰ λͺ¨λ‘ 포함 (κΈ°λ³Έ)\n" " - β‘‘ μ„±μš° λ‚΄λ ˆμ΄μ…˜λ§Œ 포함, ν•˜λ‹¨ μžλ§‰ μˆ¨κΉ€ (`--no-subtitles`)\n" " - β‘’ μ„±μš° μŒμ„± 없이 μžλ§‰ + λ°°κ²½μŒμ•…/효과음만 포함 (`--no-voice`)\n" " - β‘£ μŒμ„±Β·μžλ§‰ λͺ¨λ‘ 없이 순수 λͺ¨μ…˜ κ·Έλž˜ν”½ + λ°°κ²½μŒμ•…/효과음 쇼릴 (`--no-voice --no-subtitles`)\n" "3. **λŒ€μ‚¬/슀크립트 κ²€μˆ˜**: μœ„ 2번 μ„Ήμ…˜μ˜ μŠ¬λΌμ΄λ“œλ³„ λŒ€μ‚¬/μžλ§‰ μ „λ¬Έ 쀑 μˆ˜μ •ν•˜κ±°λ‚˜ 닀듬고 싢은 λ¬Έμž₯이 μžˆλ‚˜μš”? (μŒμ„± 포함 μ‹œ μ„±μš° `M1~M5` / `F1~F5` 선택 포함)\n" "4. **λ§žμΆ€ν˜• μ‹œκ°ν™” μΆ”κ°€ μ—¬λΆ€**: μΆ”μ²œλœ μŠ¬λΌμ΄λ“œμ— `Archify`(μ•„ν‚€ν…μ²˜), `Anatomy`(기계 단면도), `Anime.js`(μ‹€μ‹œκ°„ λ“œλ‘œμž‰ 차트) μ‹œκ°ν™”λ₯Ό μ œμž‘ν•΄ ν¬ν•¨ν• κΉŒμš”?" ) return "\n".join(lines) # --------------------------------------------------------------------------- # Sample Template Generators (PPTX, DOCX, HTML) # --------------------------------------------------------------------------- def create_sample_pptx(output_pptx: Path) -> Path: from pptx import Presentation output_pptx.parent.mkdir(parents=True, exist_ok=True) prs = Presentation() layout = prs.slide_layouts[1] sample_slides = [ { "title": "AI ꡐ윑 μ—μ΄μ „νŠΈ 9μ’…", "body": [ "[CHAPTER: 01 μ—μ΄μ „νŠΈ ꡰ체]", "[ARCHETYPE: arc_cards]", "[BADGE: PPTX Γ— HyperFrames Engine]", "[STAT_PREFIX: AI μ œμž‘ λͺ¨λ“ˆ]", "[STAT_NUM: 9]", "[STAT_SUFFIX: μ’…]", "κ΅μ•ˆλΆ„μ„ | μŠ¬λΌμ΄λ“œ ꡬ쑰 μΆ”μΆœ", "λŒ€λ³Έμ„€κ³„ | λ°œν‘œ λ…ΈνŠΈ 호흑 λΆ„ν• ", "λͺ¨μ…˜μ—”μ§„ | HTML 3D μΊ”λ²„μŠ€ 생성", "μ„±μš°ν•©μ„± | μžμ—°μŠ€λŸ¬μš΄ λ‚΄λ ˆμ΄μ…˜", "νš¨κ³ΌμŒλ΄‡ | μ „ν™˜Β·νŒμ—… μ‚¬μš΄λ“œ 싱크", "μžλ§‰μ—”μ§„ | ν•œΒ·μ˜ 이쀑 μžλ§‰ μ •λ ¬", "차트λͺ¨λ“ˆ | 숫자 μΉ΄μš΄νŠΈμ—… μ—°μΆœ", "BGM덕킹 | λ°°κ²½μŒμ•… λ³Όλ₯¨ μžλ™μ‘°μ ˆ", "λ Œλ”λŸ¬ | 1080p 무손싀 MP4 μΆ”μΆœ", ], "notes": ( "νŒŒμ›Œν¬μΈνŠΈ κ΅μ•ˆλ§Œ λ„£μœΌλ©΄ 아홉 개의 AI λͺ¨λ“ˆμ΄ λ™μ‹œμ— μ›€μ§μž…λ‹ˆλ‹€. || Put in a PowerPoint deck, and 9 AI modules spring into action.\n" "μ •μ§€λœ μŠ¬λΌμ΄λ“œλ₯Ό μ½”λ“œ 기반 λͺ¨μ…˜ μ˜μƒμœΌλ‘œ μžλ™ λ³€ν™˜ν•΄ μ£Όμ£ . || Turning static slides into code-rendered motion videos automatically." ), }, { "title": "μ™œ κΈ°μ‘΄ κ°•μ˜ μ˜μƒ μ œμž‘μ€ 였래 κ±Έλ Έμ„κΉŒ", "body": [ "[CHAPTER: 02 병λͺ© 진단]", "[ARCHETYPE: mockup_steps]", "[BADGE: Lecture Deck Analysis]", "μ •μ§€λœ PPT λ…Ήν™”μ˜ ν•œκ³„ | ν™”λ©΄ λ³€ν™”κ°€ μ—†μ–΄ 초반 μ‹œμ²­ 지속λ₯  급락", "μˆ˜λ™ μ˜μƒ νŽΈμ§‘μ˜ 병λͺ© | μžλ§‰κ³Ό 효과음 싱크λ₯Ό μ†μœΌλ‘œ λ§žμΆ”λŠλΌ μ‹œκ°„ μ†Œλͺ¨", "μ½”λ“œ 기반 SYNC μžλ™ν™” | λŒ€λ³Έ νƒ€μž„μŠ€νƒ¬ν”„μ— 맞좰 μΉ΄λ“œμ™€ μ‚¬μš΄λ“œκ°€ μ¦‰μ‹œ λ°˜μ‘", ], "notes": ( "기쑴에 PPTλ₯Ό λ„μ›Œλ†“κ³  ν™”λ©΄λ§Œ λ…Ήν™”ν•˜λ©΄ μ‹œμ²­μžκ°€ 금방 μ§€λ£¨ν•΄μ§‘λ‹ˆλ‹€. || Recording static PPT slides quickly loses viewer attention.\n" "κ·Έλ ‡λ‹€κ³  ν”„λ¦¬λ―Έμ–΄λ‘œ μΉ΄λ“œμ™€ νš¨κ³ΌμŒμ„ 일일이 λ§žμΆ”λ©΄ νŽΈμ§‘ μ‹œκ°„μ΄ μ—΄ 배둜 λŠ˜μ–΄λ‚˜μ£ . || Yet syncing cards and SFX manually in Premiere takes 10x longer.\n" "해결책은 μŒμ„±κ³Ό ν™”λ©΄, νš¨κ³ΌμŒμ„ μ½”λ“œλ‘œ ν•œ λ²ˆμ— λ™κΈ°ν™”ν•˜λŠ” κ²ƒμž…λ‹ˆλ‹€. || The answer is syncing voice, visuals, and SFX together in code." ), }, { "title": "λ°˜λ³΅λ˜λŠ” νŽΈμ§‘ κ·œμΉ™μ€ 맀뉴얼과 μ½”λ“œλ‘œ 남긴닀", "body": [ "[CHAPTER: 03 SYNC 원칙]", "[ARCHETYPE: book_quote]", "[BADGE: CHAPTER 03 Β· DESIGN SYSTEM]", "[QUOTE: 같은 λ””μžμΈκ³Ό νŽΈμ§‘μ„ 두 번 λ°˜λ³΅ν–ˆλ‹€λ©΄, 이제 κ·Έ μž‘μ—…μ€ νƒ€μž„λΌμΈμ΄ μ•„λ‹ˆλΌ μ½”λ“œ ν…œν”Œλ¦Ώμ— μžˆμ–΄μ•Ό ν•œλ‹€.]", "상단 챕터 진행바와 ν•˜λ‹¨ 이쀑 μžλ§‰ λ ˆμ΄μ•„μ›ƒμ„ κ³ μ • κ·œμΉ™μœΌλ‘œ μžμ‚°ν™”", "κ΅μ•ˆ ν…μŠ€νŠΈλ§Œ λ°”λ€Œμ–΄λ„ μΌκ΄€λœ λΈŒλžœλ“œ ν†€μ•€λ§€λ„ˆμ™€ 가독성 100% μœ μ§€", ], "notes": ( "맀번 같은 μŠ€νƒ€μΌμ˜ κ°•μ‘° νš¨κ³Όμ™€ μžλ§‰μ„ μ†μœΌλ‘œ 찍고 μžˆλ‹€λ©΄ λ°”κΏ”μ•Ό ν•©λ‹ˆλ‹€. || If you repeat the same highlights and captions by hand, change the workflow.\n" "λ””μžμΈ κ·œμΉ™κ³Ό 효과음 타이밍을 μ½”λ“œ ν…œν”Œλ¦ΏμœΌλ‘œ 남겨두면 κ΅μ•ˆλ§Œ λ°”κΏ” 계속 μ“Έ 수 μžˆμŠ΅λ‹ˆλ‹€. || Once baked into a code template, any new deck inherits the exact same polish." ), }, { "title": "λͺ…λ Ήμ–΄ ν•œ μ€„λ‘œ λλ‚˜λŠ” PPTX μ˜μƒ λ³€ν™˜ νŒŒμ΄ν”„λΌμΈ", "body": [ "[CHAPTER: 04 μžλ™ λ³€ν™˜]", "[ARCHETYPE: terminal_flow]", "[BADGE: Automated CLI Pipeline]", "[COMMAND: python3 produce_deck_video.py --input lecture.pptx --sfx on]", "[FLOW_LEFT: ꡐ윑자료 Β· PPTX]", "[FLOW_LEFT_SUB: μŠ¬λΌμ΄λ“œ Β· λ°œν‘œλ…ΈνŠΈ νŒŒμ‹±]", "[FLOW_RIGHT: 1080p λͺ¨μ…˜ MP4]", "[FLOW_RIGHT_SUB: HTML λ Œλ”λ§ + UI 효과음]", "μŠ¬λΌμ΄λ“œ 제λͺ©Β·λ³Έλ¬ΈΒ·λ°œν‘œμž λ…ΈνŠΈλ₯Ό 읽어 씬 λ ˆμ΄μ•„μ›ƒ μžλ™ λ°°μ •", "μ „ν™˜ μŠ€μš°μ‹œ, μΉ΄λ“œ 포컀슀 팝, 타이핑 μ‚¬μš΄λ“œλ₯Ό 0.01초 λ‹¨μœ„λ‘œ 배치", ], "notes": ( "μ‹€μ œ μ‚¬μš©λ²•μ€ μ•„μ£Ό κ°„λ‹¨ν•©λ‹ˆλ‹€. κ΅μœ‘μžλ£Œλ‚˜ νŒŒμ›Œν¬μΈνŠΈ 파일 ν•˜λ‚˜λ§Œ μ§€μ •ν•΄ μ£Όμ„Έμš”. || Usage is simple: just point the pipeline at your PPTX or lesson file.\n" "엔진이 μŠ¬λΌμ΄λ“œ ꡬ쑰λ₯Ό 뢄석해 HTML λͺ¨μ…˜κ³Ό νš¨κ³ΌμŒκΉŒμ§€ μž…νžŒ MP4 μ˜μƒμ„ λ°”λ‘œ λ½‘μ•„λƒ…λ‹ˆλ‹€. || It parses your slides and renders an MP4 complete with HTML motion and UI SFX." ), }, { "title": "μ•žμœΌλ‘œ ꡐ윑 μ˜μƒ μ œμž‘μ΄ λ°”λ€ŒλŠ” 3단계", "body": [ "[CHAPTER: 05 핡심 μš”μ•½]", "[ARCHETYPE: metric_summary]", "[BADGE: Ready For Production]", "[SUBTITLE: 자료 μž…λ ₯λΆ€ν„° 효과음 λ§ˆμŠ€ν„°λ§κΉŒμ§€ μ›μŠ€ν†± μžλ™ν™”]", "κ΅μ•ˆ μž…λ ₯ | PPTX Β· μ›Œλ“œ Β· λ§ˆν¬λ‹€μš΄ Β· HTML 지원", "λͺ¨μ…˜Β·νš¨κ³ΌμŒ | 3D μ›¨μ΄λΈŒΒ·λ„€μ˜¨ μΉ΄λ“œΒ·μ‚¬μš΄λ“œ μžλ™ 싱크", "κ³ ν™”μ§ˆ λ Œλ”λ§ | λŠκΉ€ μ—†λŠ” 1080p MP4 μ˜μƒ μ¦‰μ‹œ μ™„μ„±", ], "notes": ( "이제 νŒŒμ›Œν¬μΈνŠΈλ‚˜ μ›Œλ“œ κ΅μ•ˆλ§Œ μ€€λΉ„ν•˜μ‹œλ©΄ λͺ¨μ…˜ κ·Έλž˜ν”½κ³Ό νš¨κ³ΌμŒμ€ νŒŒμ΄ν”„λΌμΈμ΄ λ§‘μŠ΅λ‹ˆλ‹€. || Prepare your PPTX or Word docsβ€”the pipeline handles motion and SFX.\n" "μ–Έμ œλ“  μƒˆλ‘œμš΄ ꡐ윑자료λ₯Ό λ„£μ–΄ 같은 포맷의 κ³ ν’ˆμ§ˆ ν”„λ ˆμ  ν…Œμ΄μ…˜ μ˜μƒμ„ λ§Œλ“€μ–΄ λ³΄μ„Έμš”! || Feed in any future training deck to produce sleek presentation videos anytime!" ), }, ] for s in sample_slides: slide = prs.slides.add_slide(layout) slide.shapes.title.text = s["title"] tf = slide.placeholders[1].text_frame tf.clear() for idx, line in enumerate(s["body"]): p = tf.paragraphs[0] if idx == 0 else tf.add_paragraph() p.text = line slide.notes_slide.notes_text_frame.text = s["notes"] prs.save(str(output_pptx)) return output_pptx def create_sample_docx(output_docx: Path) -> Path: from docx import Document output_docx.parent.mkdir(parents=True, exist_ok=True) doc = Document() doc.add_heading("AI λ¬Έμ„œ μžλ™ μ˜μƒν™” κ°€μ΄λ“œ", 0) doc.add_heading("01 닀쀑 포맷 μž…λ ₯", level=1) doc.add_heading("λ„€ κ°€μ§€ 싀무 λ¬Έμ„œλ₯Ό κ·ΈλŒ€λ‘œ μ˜μƒμœΌλ‘œ", level=2) doc.add_paragraph("νŒŒμ›Œν¬μΈνŠΈ | λ°œν‘œ μŠ¬λΌμ΄λ“œμ™€ λ°œν‘œμž λ…ΈνŠΈ μžλ™ μΆ”μΆœ", style="List Bullet") doc.add_paragraph("μ›Œλ“œ λ¬Έμ„œ | 제λͺ© μŠ€νƒ€μΌκ³Ό ν‘œΒ·λ³Έλ¬Έ 단락 ꡬ쑰화", style="List Bullet") doc.add_paragraph("λ§ˆν¬λ‹€μš΄ Β· HTML | 기술 λ¬Έμ„œμ™€ μ›Ή μŠ¬λΌμ΄λ“œ μ¦‰μ‹œ νŒŒμ‹±", style="List Bullet") doc.add_paragraph("> μ‹€λ¬΄μ—μ„œ 이미 μ“°λ˜ μ›Œλ“œ κΈ°νšμ„œλ‚˜ ꡐ윑 κ΅μ•ˆμ„ μƒˆλ‘œ νŽΈμ§‘ν•  ν•„μš”κ°€ μ—†μŠ΅λ‹ˆλ‹€. || No need to re-edit existing Word specs or training docs.") doc.add_heading("02 핡심 원칙", level=1) doc.add_heading("λ¬Έμ„œ ꡬ쑰가 κ³§ λͺ¨μ…˜ μŠ€ν† λ¦¬λ³΄λ“œκ°€ λœλ‹€", level=2) doc.add_paragraph("같은 λ ˆμ΄μ•„μ›ƒ μž‘μ—…μ„ 두 번 λ°˜λ³΅ν•˜μ§€ 말고 μ½”λ“œ νŒŒμ΄ν”„λΌμΈμ— λ§‘κΈ΄λ‹€.", style="Quote") doc.add_paragraph("제λͺ©μ€ 챕터 λ°”λ‘œ, λͺ©λ‘μ€ λ„€μ˜¨ 포컀슀 μΉ΄λ“œλ‘œ μžλ™ λ§€ν•‘λ©λ‹ˆλ‹€.") doc.save(str(output_docx)) return output_docx def create_sample_html_slides(output_html: Path) -> Path: output_html.parent.mkdir(parents=True, exist_ok=True) content = """<!doctype html> <html lang="ko"> <head><meta charset="utf-8"><title>μ›Ή μŠ¬λΌμ΄λ“œ μ˜μƒ λ³€ν™˜ μƒ˜ν”Œ</title></head> <body> <section data-chapter="μ›Ή μŠ¬λΌμ΄λ“œ" data-badge="HTML Deck Import"> <h2>HTML ν”„λ ˆμ  ν…Œμ΄μ…˜λ„ κ·ΈλŒ€λ‘œ MP4둜</h2> <ul> <li>μ„Ήμ…˜ μžλ™ 인식 | Reveal.js 및 일반 μ›Ή μŠ¬λΌμ΄λ“œ νƒœκ·Έ νŒŒμ‹±</li> <li>λ°œν‘œ λ…ΈνŠΈ 연동 | aside.notes νƒœκ·Έλ₯Ό 읽어 μ„±μš° λ‚΄λ ˆμ΄μ…˜ ν•©μ„±</li> <li>효과음 κ²°ν•© | ν™”λ©΄ μ „ν™˜κ³Ό μΉ΄λ“œ ν¬μ»€μŠ€μ— 맞좰 UI μ‚¬μš΄λ“œ λ―Ήμ‹±</li> </ul> <aside class="notes"> μ›ΉμœΌλ‘œ λ§Œλ“  HTML λ°œν‘œ μžλ£Œλ„ λͺ…λ Ήμ–΄ ν•œ 쀄이면 효과음이 λ“€μ–΄κ°„ MP4 μ˜μƒμœΌλ‘œ λ³€ν™˜λ©λ‹ˆλ‹€. || Even HTML slide decks convert into SFX-synced MP4 videos in a single command. </aside> </section> </body> </html> """ output_html.write_text(content, encoding="utf-8") return output_html
"""
Universal Document-to-Deck Ingestion Parser (`ingest_parser.py`)
Supports:
- PowerPoint (.pptx) β€” Slides, bullets, tables, speaker notes
- Word (.docx) β€” Headings, paragraphs, bullet lists, quotes, tables
- Markdown (.md/.txt) β€” #, ##, -, >, tables
- HTML (.html/.htm) β€” Reveal.js <section>, .slide, or <h1>/<h2>/<ul>/<p>/<blockquote>
- JSON (.json) β€” Direct DeckSpec schema
Maps every extracted slide into one of 5 visual motion archetypes:
1. `arc_cards` β€” 3D cylindrical card carousel + glowing big counter
2. `mockup_steps` β€” Left app/slide list mockup + Right 3-card sequential focus glow
3. `book_quote` β€” 3D perspective open manual/book spread + cyan highlighter sweep
4. `terminal_flow` β€” Dark glass CLI/code window typing + glowing node flow diagram
5. `metric_summary` β€” 3 neon glass pillars + final takeaway callout badge
"""
from __future__ import annotations
import json
import re
import zipfile
from pathlib import Path
from typing import Any
class DRMProtectedFileError(RuntimeError):
"""Raised when an OpenXML (.pptx/.docx) file is encrypted by enterprise DRM or Office protection."""
def __init__(self, file_path: Path, reason: str = "") -> None:
self.file_path = file_path
self.reason = reason
super().__init__(f"DRM or encrypted document detected: {file_path} ({reason})")
AGENT_HANDOFF_JSON_SCHEMA_EXAMPLE = """{
"title": "2026 사내 AI λ„μž… μ „λž΅ λ³΄κ³ μ„œ",
"slides": [
{
"title": "μΆ”μ§„ λ°°κ²½ 및 핡심 과제 5κ°€μ§€",
"chapter": "01 μΆ”μ§„ λ°°κ²½",
"bullets": [
"λ¬Έμ„œ λ³΄μ•ˆ μ€€μˆ˜ | 사내 DRM μ—΄λžŒ μŠ€ν‚¬κ³Ό νŒŒμ΄ν”„λΌμΈ μ•ˆμ „ 연동",
"생산성 ν–₯상 | λ³΄κ³ μ„œ λΈŒλ¦¬ν•‘ μ˜μƒ μ œμž‘ μ‹œκ°„ 90% 단좕",
"ν‘œμ€€ν™” | 16μ’… λͺ¨μ…˜ κ·Έλž˜ν”½ λ ˆμ΄μ•„μ›ƒ μžλ™ λ§€ν•‘"
],
"notes": "사내 DRM μŠ€ν‚¬λ‘œ 읽은 λ°œν‘œμž λ…ΈνŠΈ λ˜λŠ” μŠ¬λΌμ΄λ“œ 핡심 μ„€λͺ… λŒ€λ³Έμž…λ‹ˆλ‹€. λ°œν‘œμž λ…ΈνŠΈκ°€ μ—†λŠ” μž₯ν‘œλŠ” μ—μ΄μ „νŠΈκ°€ μŠ¬λΌμ΄λ“œ λ‚΄μš©μ„ λ°”νƒ•μœΌλ‘œ μžμ—°μŠ€λŸ¬μš΄ 2~3λ¬Έμž₯ ꡬ어체 λ‚΄λ ˆμ΄μ…˜μ„ μž‘μ„±ν•΄ λ„£μœΌμ„Έμš”."
},
{
"title": "κΈ°μ‘΄ 방식 vs λͺ¨μ…˜ νŒŒμ΄ν”„λΌμΈ 비ꡐ",
"chapter": "02 λ„μž… 효과",
"table": [
["비ꡐ ν•­λͺ©", "λ‹¨μˆœ μŠ¬λΌμ΄λ“œ 캑처", "Deck2Video λͺ¨μ…˜ μ—”μ§„"],
["μ‹œκ° λͺ°μž…도", "βœ•", "βœ“"],
["단어 ν•˜μ΄λΌμ΄νŠΈ μžλ§‰", "βœ•", "βœ“"],
["사내 폐쇄망 지원", "β–³", "βœ“"]
],
"notes": "ν‘œ 데이터도 table λ°°μ—΄λ‘œ μ „λ‹¬ν•˜λ©΄ λ„€μ˜¨ 비ꡐ λ§€νŠΈλ¦­μŠ€λ‚˜ 데이터 차트 λͺ¨μ…˜μœΌλ‘œ μžλ™ λ³€ν™˜λ©λ‹ˆλ‹€."
}
]
}"""
def check_openxml_drm(file_path: Path) -> None:
"""Verify whether a .pptx/.docx file is a valid OpenXML ZIP or encrypted by corporate DRM."""
try:
header = file_path.read_bytes()[:64]
except Exception as exc:
raise DRMProtectedFileError(file_path, f"file read blocked: {exc}") from exc
if not header:
raise DRMProtectedFileError(file_path, "empty file (0 bytes)")
# Standard OpenXML (.pptx, .docx, .xlsx) starts with ZIP magic bytes PK\x03\x04 (or PK\x05\x06 / PK\x07\x08)
if header[:2] == b"PK" and zipfile.is_zipfile(file_path):
return
upper_hdr = header.upper()
vendor = "Corporate DRM / Encrypted Container"
if b"FASOO" in upper_hdr or b"FSN" in upper_hdr:
vendor = "Fasoo DRM (νŒŒμˆ˜λ‹·μ»΄ λ¬Έμ„œλ³΄μ•ˆ)"
elif b"MARKANY" in upper_hdr or b"DOCUMENT SAFER" in upper_hdr:
vendor = "MarkAny DRM (λ§ˆν¬μ• λ‹ˆ λ¬Έμ„œλ³΄μ•ˆ)"
elif b"SOFTCAMP" in upper_hdr or b"SCDSA" in upper_hdr:
vendor = "Softcamp DRM (μ†Œν”„νŠΈμΊ ν”„ λ¬Έμ„œλ³΄μ•ˆ)"
elif header[:8] == b"\xd0\xcf\x11\xe0\xa1\xb1\x1a\xe1":
vendor = "OLE2 Encrypted Package (MS Purview AIP / Office μ•”ν˜Έν™” λ˜λŠ” κ΅¬ν˜• .ppt/.doc λ°”μ΄λ„ˆλ¦¬)"
raise DRMProtectedFileError(
file_path,
f"{vendor} β€” magic={header[:12]!r} (python-pptx/python-docx cannot open encrypted binary directly)",
)
ARCHETYPES = [
"hero_3d_briefing",
"media_showcase_split",
"cinema_broll_fullscreen",
"dossier_stamp_card",
"archify_architecture",
"anatomy_mechanic",