Research Genealogy
unumbrela/research-genealogy
Genealogia de pesquisa verificável: veja como ideias se cruzaram e evoluíram
Ver no GitHubSkill do Claude Code que transforma uma direção de pesquisa em uma genealogia não-linear verificável de seus artigos-chave. Integra metadados reais do OpenAlex, valida citações e gera árvores visuais em ASCII, Mermaid ou DrawIO para mostrar como um campo evoluiu: trabalho fundador → problema abordado → quem construiu sobre ele → linhas paralelas → supersessões → fronteira atual.
Instalação
npx convez-ai unumbrela/research-genealogy1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Mapear a evolução histórica de um campo de pesquisa (ex: detecção de imagens geradas)
- Identificar trabalhos fundadores, hubs influentes e fronteiras recentes em uma área
- Entender relações não-lineares entre papers (construções, inspirações, linhas paralelas)
Veredito

Ferramenta sólida e diferenciada para pesquisadores que precisam entender genealogias intelectuais, não apenas listas de papers. A verificação de citações reais via OpenAlex e a representação não-linear com marcação de relações (`builds-on`, `parallel`, `supersedes`) a distinguem de alternativas genéricas. Recomendado para quem estuda evolução de campos em profundidade.
README
A tradução para o português deste README ainda está na fila.
research-genealogy
Input a research direction → get its development genealogy, not just a paper list.
A Claude Code skill that researches the literature for a field (e.g. "generated image detection", "对比学习", "neural machine translation") and lays out how it evolved: the founding work → the problem it tackled → who built on it → which lines ran in parallel → what got superseded → today's frontier.
Output is a terminal ASCII genealogy tree (drawn with -- connectors) plus a
short narrative — the relationships are non-linear (branches & parallel
lines), not a flat timeline.
(real run; colored in a terminal — see examples/generated-image-detection.md)
╭───────────────────────────────────────────────────────────╮
│ 生成图像检测 (Generated / AI-Synthesized Image Detection) │
│ 12 papers · 2018 → 2026 │
╰───────────────────────────────────────────────────────────╯
│
2018 │ ● Marra et al. ██████░ 389 ← 奠基:生成图像残留可检测指纹
│ │
2019 │ ├── ◉ Ning Yu et al. ██████░ 425 · GAN指纹归因
│ │
2019 │ ├┈┈ ○ Durall et al. █████░░ 176 ← 频域分支起点
│ │ └── ○ Frank et al. █████░░ 209 ∥ parallel: Wang
│ │ └── ○ Corvi et al. ██████░ 236 (扩散时代取证) ∥ DIRE/Ojha
│ │
2020 │ └── ◉ Sheng-Yu Wang et al. ███████ 994 ← CNNDetection 主干
│ 单一ResNet+强增广跨GAN泛化,最强基线
│ │
2023 │ ├── ◉ Utkarsh Ojha et al. ██████░ 269 · 冻结CLIP特征通用检测
│ │ ├── ★ Cozzolino et al. ✦NEW █████░░ 85 (2024, CLIP SOTA)
│ │ └┈┈ ★ Marco Willi et al. ✦NEW ······· 0 (2026, 机制剖析)
│ ├── ○ Zhendong Wang et al. █████░░ 222 · DIRE 重建误差判据
│ │ └── ★ Ricker et al. ✦NEW ████░░░ 38 (2024, training-free)
│ └── ★ Huan Liu et al. ✦NEW ████░░░ 64 (2024, FatFormer)
│
● founder ◉ hub ★ frontier · ├── builds-on ├┈┈ inspired-by ∥ parallel
A left year axis, role markers (● founder / ◉ hub / ★ frontier), relation-coded branches (
├──builds-on,├┈┈inspired-by), citation bars, and∥ parallelcross-links. The frontier pass guarantees recent (2024+) work is included, not just the classics.
Why this is different
| Existing tools | What they give | What's missing |
|---|---|---|
| Survey generators (SurveyForge, AutoSurvey…) | a survey organized by theme | not who built on whom |
| ResearchRabbit | a citation graph to read yourself | no narrative |
| Paper search (Semantic Scholar…) | a list | no lineage |
research-genealogy gives you the lineage: a readable genealogy of ideas
with explicit builds-on / parallel / supersedes edges.
No hallucinated papers — and verifiable edges
Every node comes from real metadata fetched from OpenAlex
(or Semantic Scholar) — Claude organizes and narrates, it never recalls papers
from memory. Node summaries are written from the papers' real abstracts
(--abstract).
And the lineage itself is checkable: scripts/verify.py confirms that every
builds-on edge is a real citation in the data, marking each ✓ verified,
⚠ unverified, ↺ reversed, or ‼ cross-cite. The genealogy shows the marks
inline — so you can trust the arrows, not just the boxes.
2019 │ ├── ◉ Ning Yu et al. ✓ ██████░ 425 ← edge verified as a real citation
2020 │ │ └── ○ Frank et al. ⚠ █████░░ 209 ← citation not found in OpenAlex; flagged honestly
…
citations: ✓ 8 verified ⚠ 7 to review (run verify.py)
Stdlib-only scripts, no pip install, no API key required.
Install
npx skills add unumbrela/research-genealogy -g -a claude-code
Or drop this folder into ~/.claude/skills/research-genealogy/.
Use
In Claude Code, just ask:
帮我梳理「生成图像检测」这个方向的发展历程
Claude will refine a draft into a verified genealogy (see SKILL.md).
One command (fast path)
python3 scripts/genealogy.py "generated image detection" --nodes 12 --render
Runs every search pass, selects load-bearing nodes spanning founders → frontier,
and derives the builds-on edges from the real citation graph — so the draft
is citation-grounded out of the box (most edges verify ✓ immediately). Claude then
refines summaries, prunes off-topic nodes, and relabels relations.
Output formats
python3 scripts/render_tree.py lineage.json # colored terminal tree (default)
python3 scripts/render_tree.py lineage.json --format mermaid # GitHub-renderable graph
python3 scripts/render_tree.py lineage.json --format markdown # report: tree + table + edges
python3 scripts/render_tree.py lineage.json --format bibtex # cite every node
python3 scripts/render_tree.py lineage.json --format drawio # editable draw.io / diagrams.net diagram
Manual passes (full control)
# anchor search (relevance); add --precise if results look off-topic
python3 scripts/papers.py search "generated image detection" --limit 25
# landmark lookup by exact title -> real id/metadata (+abstract to ground summary)
python3 scripts/papers.py search "CNN-Generated Images Are Surprisingly Easy to Spot" --abstract --limit 1
# frontier pass (REQUIRED): recent work that citation-sort would bury
python3 scripts/papers.py search "AI-generated image detection" --precise --from-year 2024 --sort citations --limit 15
# expand a hub to ground edges, verify, render
python3 scripts/papers.py expand W3034577585 --limit 25 --abstract
python3 scripts/verify.py lineage.json --write
python3 scripts/render_tree.py lineage.json
Search flags: --precise (every term in title/abstract — cuts off-topic
giants), --from-year / --to-year (time window), --sort relevance|citations|recent,
--abstract (ground summaries). Backends: --source openalex (default, keyless)
or --source s2 (set S2_API_KEY). Set OPENALEX_MAILTO=you@example.com for
OpenAlex's faster pool.
lineage.json schema
{
"field": "生成图像检测",
"nodes": [
{"id":"wang2020","title":"...","authors":"Wang et al.","year":2020,
"venue":"CVPR","citations":1500,
"problem":"<一句话问题>","contribution":"<一句话方法>","url":"https://..."}
],
"edges": [
{"from":"marra2018","to":"wang2020","relation":"builds-on"}
]
}
relation ∈ builds-on | inspired-by | parallel | supersedes.
A node may have several parents — that's the point.
License
MIT