OXI-717/ai-native-toolkit / plugins/research-ensemble/skills/research-ensemble
research-ensemble
VerifiedMulti-agent adversarial research with Exa AI. Two levels: standard (3-5 agents, 30-60 min) and deep (10-15 agents, full ensemble with 3 cycles). Triggers: "research X", "исследуй", "deep research", "find out about", "compare options for", "deep dive into", "gather information on", "write a report about", "проанализируй", "разберись в", "investigate", "analyze topic", "сравни варианты", "изучи тему"
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 6.4 KB |
| references/agent-prompts/critic.md | markdown | 2.8 KB |
| references/agent-prompts/deep-diver.md | markdown | 2.5 KB |
| references/agent-prompts/domain-reviewer.md | markdown | 2.8 KB |
| references/agent-prompts/fact-checker.md | markdown | 2.7 KB |
| references/agent-prompts/interaction-mapper.md | markdown | 3.4 KB |
| references/agent-prompts/scout.md | markdown | 4.4 KB |
| references/agent-prompts/statistician.md | markdown | 2.5 KB |
| references/agent-prompts/synthesizer.md | markdown | 3.0 KB |
| references/deep-workflow.md | markdown | 9.9 KB |
| references/domains.md | markdown | 6.2 KB |
| references/exa-search-guide.md | markdown | 4.3 KB |
| references/generate-pdf-report.pyexecutable | script | 19.0 KB |
| references/profile-template.md | markdown | 1.6 KB |
| references/standard-workflow.md | markdown | 5.9 KB |
references/generate-pdf-report.py
View on GitHub#!/usr/bin/env python3
"""
Generate a PDF research report from synthesis.md and supporting files.
Usage: python3 generate-pdf-report.py <output_dir> <topic> <level> <domain> [language]
Requires: reportlab (pip install reportlab)
Fonts: Uses Arial (macOS system font) for broad Unicode support.
"""
import sys
import os
import re
import textwrap
from datetime import datetime
from reportlab.lib.pagesizes import A4
from reportlab.lib.units import mm
from reportlab.lib.colors import HexColor
from reportlab.lib.styles import ParagraphStyle
from reportlab.lib.enums import TA_LEFT, TA_CENTER, TA_JUSTIFY
from reportlab.platypus import (
SimpleDocTemplate, Paragraph, Spacer, PageBreak,
Table, TableStyle, HRFlowable, KeepTogether
)
from reportlab.pdfbase import pdfmetrics
from reportlab.pdfbase.ttfonts import TTFont
# --- Font Registration ---
# macOS system fonts with broad Unicode support
FONT_DIR = "/System/Library/Fonts/Supplemental"
try:
pdfmetrics.registerFont(TTFont("Arial", f"{FONT_DIR}/Arial.ttf"))
pdfmetrics.registerFont(TTFont("Arial-Bold", f"{FONT_DIR}/Arial Bold.ttf"))
pdfmetrics.registerFont(TTFont("Arial-Italic", f"{FONT_DIR}/Arial Italic.ttf"))
pdfmetrics.registerFontFamily("Arial", normal="Arial", bold="Arial-Bold", italic="Arial-Italic")
except Exception as e:
sys.stderr.write(f"Error: Arial fonts not found at {FONT_DIR}: {e}\n")
sys.stderr.write("This script requires macOS with Arial fonts for Unicode support.\n")
sys.exit(1)
# --- Colors ---
PRIMARY = HexColor("#1a1a2e")
ACCENT = HexColor("#4a6fa5")
LIGHT_BG = HexColor("#f0f4f8")
BORDER = HexColor("#d0d8e0")
TEXT_COLOR = HexColor("#2c3e50")
MUTED = HexColor("#7f8c8d")
# --- Styles ---
STYLES = {
"title": ParagraphStyle(
"Title", fontName="Arial-Bold", fontSize=22, leading=28,
textColor=PRIMARY, alignment=TA_CENTER, spaceAfter=6*mm,
),
"subtitle": ParagraphStyle(
"Subtitle", fontName="Arial", fontSize=11, leading=14,
textColor=MUTED, alignment=TA_CENTER, spaceAfter=10*mm,
),
"h1": ParagraphStyle(
"H1", fontName="Arial-Bold", fontSize=16, leading=22,
textColor=PRIMARY, spaceBefore=8*mm, spaceAfter=4*mm,
),
"h2": ParagraphStyle(
"H2", fontName="Arial-Bold", fontSize=13, leading=18,
textColor=ACCENT, spaceBefore=6*mm, spaceAfter=3*mm,
),
"h3": ParagraphStyle(
"H3", fontName="Arial-Bold", fontSize=11, leading=15,
textColor=TEXT_COLOR, spaceBefore=4*mm, spaceAfter=2*mm,
),
"body": ParagraphStyle(
"Body", fontName="Arial", fontSize=10, leading=15,
textColor=TEXT_COLOR, alignment=TA_JUSTIFY, spaceAfter=2*mm,
),
"bullet": ParagraphStyle(
"Bullet", fontName="Arial", fontSize=10, leading=15,
textColor=TEXT_COLOR, leftIndent=8*mm, bulletIndent=3*mm,
spaceAfter=1.5*mm,
),
"tldr": ParagraphStyle(
"TLDR", fontName="Arial", fontSize=10.5, leading=16,
textColor=PRIMARY, leftIndent=4*mm, rightIndent=4*mm,
spaceBefore=3*mm, spaceAfter=3*mm, backColor=LIGHT_BG,
borderPadding=(3*mm, 3*mm, 3*mm, 3*mm),
),
"footer": ParagraphStyle(
"Footer", fontName="Arial-Italic", fontSize=8, leading=10,
textColor=MUTED, alignment=TA_CENTER,
),
}
LABELS = {
"en": {
"report": "RESEARCH REPORT", "level": "Level", "domain": "Domain", "date": "Date",
"method": "Method: Multi-Agent Adversarial Ensemble (Exa AI)", "contents": "Contents",
"takeaways": "Key Takeaways", "main_report": "Main Report", "fact_check": "Fact Check",
"critic_review": "Critic Review", "open_questions": "Open Questions and Next Steps",
"tldr_missing": "TL;DR section not found in synthesis.md", "synthesis_missing": "synthesis.md not found",
"fact_check_missing": "_fact_check.md not found", "critic_missing": "_critic_review.md not found",
"unknowns_missing": "unknowns_and_next.md not found", "footer": "Deep Research",
"tldr_headings": ("TL;?DR", "Executive Summary", "Key Takeaways"),
},
"ru": {
"report": "ОТЧЁТ ОБ ИССЛЕДОВАНИИ", "level": "Уровень", "domain": "Домен", "date": "Дата",
"method": "Метод: мультиагентный состязательный анализ (Exa AI)", "contents": "Содержание",
"takeaways": "Ключевые выводы", "main_report": "Основной отчёт", "fact_check": "Проверка фактов",
"critic_review": "Критический обзор", "open_questions": "Открытые вопросы и следующие шаги",
"tldr_missing": "Раздел с ключевыми выводами не найден в synthesis.md", "synthesis_missing": "synthesis.md не найден",
"fact_check_missing": "_fact_check.md не найден", "critic_missing": "_critic_review.md не найден",
"unknowns_missing": "unknowns_and_next.md не найден", "footer": "Глубокое исследование",
"tldr_headings": ("TL;?DR", "Ключевые выводы", "Краткие выводы", "Резюме"),
},
"es": {
"report": "INFORME DE INVESTIGACIÓN", "level": "Nivel", "domain": "Dominio", "date": "Fecha",
"method": "Método: conjunto adversarial multiagente (Exa AI)", "contents": "Contenido",
"takeaways": "Conclusiones clave", "main_report": "Informe principal", "fact_check": "Verificación de datos",
"critic_review": "Revisión crítica", "open_questions": "Preguntas abiertas y próximos pasos",
"tldr_missing": "No se encontró la sección de conclusiones en synthesis.md", "synthesis_missing": "No se encontró synthesis.md",
"fact_check_missing": "No se encontró _fact_check.md", "critic_missing": "No se encontró _critic_review.md",
"unknowns_missing": "No se encontró unknowns_and_next.md", "footer": "Investigación profunda",
"tldr_headings": ("TL;?DR", "Resumen ejecutivo", "Conclusiones clave"),
},
}
def language_key(language):
"""Normalize the selected report language to a supported label set."""
normalized = language.strip().lower().replace("_", "-")
if normalized.startswith(("ru", "russian", "рус")):
return "ru"
if normalized.startswith(("es", "spanish", "español")):
return "es"
return "en"
def read_file(path):
"""Read file content, return empty string if not found."""
try:
with open(path, "r", encoding="utf-8") as f:
return f.read()
except FileNotFoundError:
return ""
def escape_xml(text):
"""Escape XML special characters for ReportLab Paragraph."""
text = text.replace("&", "&")
text = text.replace("<", "<")
text = text.replace(">", ">")
return text
def md_inline_to_rl(text):
"""Convert inline markdown (bold, italic, code) to ReportLab XML tags."""
text = escape_xml(text)
# Bold: **text** or __text__
text = re.sub(r'\*\*(.+?)\*\*', r'<b>\1</b>', text)
text = re.sub(r'__(.+?)__', r'<b>\1</b>', text)
# Italic: *text* or _text_
text = re.sub(r'\*(.+?)\*', r'<i>\1</i>', text)
text = re.sub(r'(?<!\w)_(.+?)_(?!\w)', r'<i>\1</i>', text)
# Inline code: `text` — use Arial for Cyrillic support (Courier has no Cyrillic glyphs)
text = re.sub(r'`(.+?)`', r'<font face="Arial" size="9" color="#c0392b">\1</font>', text)
# Links: [text](url) → text
text = re.sub(r'\[([^\]]+)\]\([^\)]+\)', r'\1', text)
return text
def parse_markdown_to_flowables(md_text):
"""Parse markdown text into ReportLab flowables."""
flowables = []
lines = md_text.split("\n")
i = 0
buffer = []
def flush_buffer():
if buffer:
text = " ".join(buffer)
flowables.append(Paragraph(md_inline_to_rl(text), STYLES["body"]))
buffer.clear()
while i < len(lines):
line = lines[i]
stripped = line.strip()
# Skip empty lines (flush buffer)
if not stripped:
flush_buffer()
i += 1
continue
# Headings
if stripped.startswith("### "):
flush_buffer()
flowables.append(Paragraph(md_inline_to_rl(stripped[4:]), STYLES["h3"]))
i += 1
continue
if stripped.startswith("## "):
flush_buffer()
flowables.append(Paragraph(md_inline_to_rl(stripped[3:]), STYLES["h2"]))
i += 1
continue
if stripped.startswith("# "):
flush_buffer()
flowables.append(Paragraph(md_inline_to_rl(stripped[2:]), STYLES["h1"]))
i += 1
continue
# Horizontal rule
if stripped in ("---", "***", "___"):
flush_buffer()
flowables.append(Spacer(1, 2*mm))
flowables.append(HRFlowable(width="100%", thickness=0.5, color=BORDER))
flowables.append(Spacer(1, 2*mm))
i += 1
continue
# Bullet points
if re.match(r'^[-*+]\s', stripped):
flush_buffer()
bullet_text = re.sub(r'^[-*+]\s+', '', stripped)
flowables.append(Paragraph(
f"• {md_inline_to_rl(bullet_text)}", STYLES["bullet"]
))
i += 1
continue
# Numbered list
m = re.match(r'^(\d+)\.\s+(.+)', stripped)
if m:
flush_buffer()
num, text = m.group(1), m.group(2)
flowables.append(Paragraph(
f"{num}. {md_inline_to_rl(text)}", STYLES["bullet"]
))
i += 1
continue
# Code blocks — render as preformatted
if stripped.startswith("```"):
flush_buffer()
code_lines = []
i += 1
while i < len(lines) and not lines[i].strip().startswith("```"):
code_lines.append(escape_xml(lines[i]))
i += 1
i += 1 # skip closing ```
code_text = "<br/>".join(code_lines) if code_lines else " "
code_style = ParagraphStyle(
"Code", fontName="Arial", fontSize=8.5, leading=12,
textColor=TEXT_COLOR, backColor=LIGHT_BG,
leftIndent=4*mm, rightIndent=4*mm,
spaceBefore=2*mm, spaceAfter=2*mm,
borderPadding=(2*mm, 2*mm, 2*mm, 2*mm),
)
flowables.append(Paragraph(code_text, code_style))
continue
# Markdown table
if "|" in stripped and stripped.startswith("|"):
flush_buffer()
table_rows = []
while i < len(lines) and lines[i].strip().startswith("|"):
row = lines[i].strip()
# Skip separator rows (|---|---|)
if re.match(r'^\|[\s\-:|]+$', row):
i += 1
continue
cells = [c.strip() for c in row.split("|")[1:-1]]
table_rows.append(cells)
i += 1
if table_rows:
# Convert to Paragraphs for wrapping
data = []
for ri, row in enumerate(table_rows):
style = STYLES["body"] if ri > 0 else ParagraphStyle(
"TableHeader", parent=STYLES["body"],
fontName="Arial-Bold", fontSize=9,
)
data.append([Paragraph(md_inline_to_rl(c), style) for c in row])
ncols = max(len(r) for r in data) if data else 1
col_width = (A4[0] - 40*mm) / ncols
t = Table(data, colWidths=[col_width]*ncols)
t_style = [
("BACKGROUND", (0, 0), (-1, 0), LIGHT_BG),
("GRID", (0, 0), (-1, -1), 0.5, BORDER),
("VALIGN", (0, 0), (-1, -1), "TOP"),
("TOPPADDING", (0, 0), (-1, -1), 2*mm),
("BOTTOMPADDING", (0, 0), (-1, -1), 2*mm),
("LEFTPADDING", (0, 0), (-1, -1), 2*mm),
("RIGHTPADDING", (0, 0), (-1, -1), 2*mm),
]
t.setStyle(TableStyle(t_style))
flowables.append(Spacer(1, 2*mm))
flowables.append(t)
flowables.append(Spacer(1, 2*mm))
continue
# Regular paragraph text — accumulate
buffer.append(stripped)
i += 1
flush_buffer()
return flowables
def slugify_filename(text, max_len=80):
"""Create a safe filename from text, preserving non-ASCII characters."""
# Remove characters unsafe for filenames
safe = re.sub(r'[<>:"/\\|?*]', '', text)
# Collapse whitespace
safe = re.sub(r'\s+', ' ', safe).strip()
return safe[:max_len]
def build_pdf(output_dir, topic, level, domain, language="English"):
"""Build PDF report from research output files."""
labels = LABELS[language_key(language)]
date_str = datetime.now().strftime("%Y-%m-%d")
# Prefix is deployment-specific (vault slug and the like), so it comes from
# the environment instead of being hardcoded here.
prefix = slugify_filename(os.environ.get("DEEP_RESEARCH_FILENAME_PREFIX", "").strip())
prefix = f"{prefix} " if prefix else ""
filename = f"{prefix}{{research}} {slugify_filename(topic)} – {date_str} – DeepResearch.pdf"
pdf_path = os.path.join(output_dir, filename)
timestamp = datetime.now().strftime("%d.%m.%Y %H:%M")
doc = SimpleDocTemplate(
pdf_path,
pagesize=A4,
topMargin=20*mm,
bottomMargin=20*mm,
leftMargin=20*mm,
rightMargin=20*mm,
title=f"{labels['report']}: {topic}",
author="Deep Research — Adversarial Ensemble",
)
story = []
# --- Cover Page ---
story.append(Spacer(1, 30*mm))
story.append(Paragraph(labels["report"], STYLES["title"]))
story.append(Spacer(1, 4*mm))
story.append(Paragraph(md_inline_to_rl(topic), ParagraphStyle(
"TopicTitle", fontName="Arial-Bold", fontSize=18, leading=24,
textColor=ACCENT, alignment=TA_CENTER, spaceAfter=8*mm,
)))
story.append(HRFlowable(width="60%", thickness=1, color=ACCENT))
story.append(Spacer(1, 6*mm))
meta_lines = [
f"{labels['level']}: {level} | {labels['domain']}: {domain}",
f"{labels['date']}: {timestamp}",
labels["method"],
]
for line in meta_lines:
story.append(Paragraph(line, STYLES["subtitle"]))
story.append(PageBreak())
# --- Table of Contents (lightweight) ---
story.append(Paragraph(labels["contents"], STYLES["h1"]))
story.append(Spacer(1, 4*mm))
toc_items = [
f"1. {labels['takeaways']} (TL;DR)",
f"2. {labels['main_report']}",
f"3. {labels['fact_check']}",
f"4. {labels['critic_review']}",
f"5. {labels['open_questions']}",
]
for item in toc_items:
story.append(Paragraph(item, ParagraphStyle(
"TOC", fontName="Arial", fontSize=11, leading=18,
textColor=ACCENT, leftIndent=10*mm, spaceAfter=2*mm,
)))
story.append(PageBreak())
# --- 1. TL;DR ---
synthesis = read_file(os.path.join(output_dir, "synthesis.md"))
tldr_text = ""
if synthesis:
# Extract TL;DR section
tldr_pattern = "|".join(labels["tldr_headings"])
tldr_match = re.search(
rf'(?:^|\n)##?\s*(?:\d+\.\s*)?(?:{tldr_pattern})[^\n]*\n(.*?)(?=\n##?\s|\Z)',
synthesis, re.DOTALL | re.IGNORECASE
)
if tldr_match:
tldr_text = tldr_match.group(1).strip()
story.append(Paragraph(f"1. {labels['takeaways']}", STYLES["h1"]))
if tldr_text:
story.extend(parse_markdown_to_flowables(tldr_text))
else:
story.append(Paragraph(
f"<i>{labels['tldr_missing']}</i>", STYLES["body"]
))
story.append(PageBreak())
# --- 2. Main Report ---
story.append(Paragraph(f"2. {labels['main_report']}", STYLES["h1"]))
if synthesis:
# Remove TL;DR section to avoid duplication, and frontmatter
body = re.sub(
rf'(?:^|\n)##?\s*(?:\d+\.\s*)?(?:{tldr_pattern})[^\n]*\n.*?(?=\n##?\s|\Z)',
'', synthesis, flags=re.DOTALL | re.IGNORECASE
)
# Remove YAML frontmatter
body = re.sub(r'^---\n.*?\n---\n', '', body, flags=re.DOTALL)
# Remove top-level title (first # line)
body = re.sub(r'^#\s+[^\n]+\n', '', body.strip())
story.extend(parse_markdown_to_flowables(body.strip()))
else:
story.append(Paragraph(
f"<i>{labels['synthesis_missing']}</i>", STYLES["body"]
))
story.append(PageBreak())
# --- 3. Fact Check ---
fact_check = read_file(os.path.join(output_dir, "reviews", "_fact_check.md"))
story.append(Paragraph(f"3. {labels['fact_check']}", STYLES["h1"]))
if fact_check:
# Remove top-level title
fact_check = re.sub(r'^#\s+[^\n]+\n', '', fact_check.strip())
story.extend(parse_markdown_to_flowables(fact_check.strip()))
else:
story.append(Paragraph(
f"<i>{labels['fact_check_missing']}</i>", STYLES["body"]
))
story.append(PageBreak())
# --- 4. Critic Review ---
critic = read_file(os.path.join(output_dir, "reviews", "_critic_review.md"))
story.append(Paragraph(f"4. {labels['critic_review']}", STYLES["h1"]))
if critic:
critic = re.sub(r'^#\s+[^\n]+\n', '', critic.strip())
story.extend(parse_markdown_to_flowables(critic.strip()))
else:
story.append(Paragraph(
f"<i>{labels['critic_missing']}</i>", STYLES["body"]
))
story.append(PageBreak())
# --- 5. Open Questions ---
unknowns = read_file(os.path.join(output_dir, "unknowns_and_next.md"))
story.append(Paragraph(f"5. {labels['open_questions']}", STYLES["h1"]))
if unknowns:
unknowns = re.sub(r'^#\s+[^\n]+\n', '', unknowns.strip())
story.extend(parse_markdown_to_flowables(unknowns.strip()))
else:
story.append(Paragraph(
f"<i>{labels['unknowns_missing']}</i>", STYLES["body"]
))
# --- Footer on every page ---
def add_page_number(canvas_obj, doc_obj):
canvas_obj.saveState()
canvas_obj.setFont("Arial", 8)
canvas_obj.setFillColor(MUTED)
page_num = canvas_obj.getPageNumber()
canvas_obj.drawCentredString(
A4[0] / 2, 12*mm,
f"{labels['footer']} — {topic} — p. {page_num}"
)
canvas_obj.restoreState()
try:
doc.build(story, onFirstPage=add_page_number, onLaterPages=add_page_number)
except Exception as e:
sys.stderr.write(f"PDF build failed: {e}\n")
sys.exit(1)
return os.path.abspath(pdf_path)
if __name__ == "__main__":
if len(sys.argv) < 5:
print(f"Usage: {sys.argv[0]} <output_dir> <topic> <level> <domain> [language]")
sys.exit(1)
output_dir = sys.argv[1]
topic = sys.argv[2]
level = sys.argv[3]
domain = sys.argv[4]
language = sys.argv[5] if len(sys.argv) > 5 else "English"
pdf_path = build_pdf(output_dir, topic, level, domain, language)
print(pdf_path)
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |