Given a paper, find its related works (cited references and successor/citing papers) and produce a structured map.
Given an academic paper, build a comprehensive related-works map covering both predecessors (papers it cites) and successors (papers that cite it or build on it).
Parse the paper from user input (arXiv URL, ID, title, PDF path).
Use AlphaXiv skill to read the paper (always try this first):
WebFetch: https://alphaxiv.org/overview/{PAPER_ID}.md
WebFetch: https://alphaxiv.org/abs/{PAPER_ID}.md
Extract the full paper content for the walkthrough and related works analysis.
Write a structured walkthrough of the paper. This goes at the top of the final report.
| Method | Benchmark | Metric | Value |
|--------|-----------|--------|-------|
| This paper | ... | ... | ... |
| Best baseline | ... | ... | ... |
From the paper's references and related work section, identify papers grouped by role:
| Role | Description | Example |
|---|---|---|
| Foundation | Core method this paper builds on | "We build on DP3 [Ze et al.]" |
| Baseline | Methods compared against in experiments | "We compare with ACT, DP..." |
| Component | Borrowed modules (encoder, loss, etc.) | "We use PTv3 as backbone" |
| Concurrent | Independent parallel work on same problem | "Concurrent to ours, X also..." |
| Problem | Papers defining the problem or benchmark | "On the RLBench benchmark [James et al.]" |
For each cited paper, extract:
Search for papers that cite or build on the target paper using multiple sources:
4a. Semantic Scholar API (primary)
# Get paper details + citations
curl -s "https://api.semanticscholar.org/graph/v1/paper/ArXiv:{PAPER_ID}?fields=title,year,citationCount,citations.title,citations.year,citations.authors,citations.externalIds,citations.citationCount"
# If not found by arXiv ID, search by title
curl -s "https://api.semanticscholar.org/graph/v1/paper/search?query=<URL_ENCODED_TITLE>&limit=5&fields=title,year,externalIds,citations.title,citations.year,citations.authors,citations.externalIds,citations.citationCount"
4b. Local Vec-db semantic search (find related top-venue work)
cd /home/vla-reasoning/proj/litian-research/vec-db
npx tsx src/cli.ts search "<paper's key concepts>" --top 10
This finds top-venue papers with similar themes that may not directly cite the target but are closely related. Especially useful for discovering parallel/concurrent work.
4c. Web search (supplementary)
WebSearch: "<paper_title>" improvements OR "builds on" OR "extends" site:arxiv.org
WebSearch: "<paper_title>" cite OR citing 2024 2025 2026
4d. AlphaXiv (if available)
Check the AlphaXiv overview for mentions of follow-up or concurrent work.
4d. Connected Papers / Papers With Code
WebSearch: "<paper_title>" site:paperswithcode.com
WebSearch: "<paper_title>" site:connectedpapers.com
For each successor found, classify:
| Relationship | Description |
|---|---|
| Direct extension | Explicitly builds on this paper's method |
| Application | Applies the method to a new domain/task |
| Improvement | Proposes fixes or enhancements |
| Comparison | Uses as baseline in experiments |
| Integration | Combines with other methods |
Rank by relevance: direct extensions > improvements > applications > comparisons.
ASK USER before generating: "I found N predecessors and M successors. Want me to:
Output format:
# Paper Walkthrough & Related Works: <Paper Title>
**Paper:** <title> (<authors>, <year>)
**arXiv:** <link>
**Core contribution:** <1 sentence>
---
## 1. Paper Walkthrough
### Background & Problem
<What problem, why it matters, what gap exists>
### Motivation & Key Insight
<Core insight, what's different from prior work>
### Method
<High-level approach, architecture, key design choices>
### Results
| Method | Benchmark | Metric | Value |
|--------|-----------|--------|-------|
| This paper | ... | ... | ... |
| Best baseline | ... | ... | ... |
<Key ablation findings, qualitative highlights>
### Limitations & Future Work
- <Author-stated limitations>
- <Observed limitations>
- <Proposed future directions>
---
## 2. Predecessors (Papers This Work Builds On)
### Foundations
| Paper | Year | Relationship | Code |
|-------|------|-------------|------|
| [Title](https://arxiv.org/abs/...) | ... | ... | [repo](url) |
### Baselines
| Paper | Year | Relationship | Code |
|-------|------|-------------|------|
### Components
| Paper | Year | Relationship | Code |
|-------|------|-------------|------|
### Concurrent Work
| Paper | Year | Relationship | Code |
|-------|------|-------------|------|
## 3. Successors (Papers That Build on This Work)
### Direct Extensions
| Paper | Year | What They Add | Citations | Code |
|-------|------|--------------|-----------|------|
| [Title](https://arxiv.org/abs/...) | ... | ... | ... | [repo](url) |
### Improvements
| Paper | Year | What They Fix/Improve | Citations | Code |
|-------|------|----------------------|-----------|------|
### Applications
| Paper | Year | Domain/Task | Citations | Code |
|-------|------|------------|-----------|------|
## 4. Research Lineage
<ASCII diagram showing the main line of research>
<target_paper>
├── built on: <foundation_1> → <foundation_2> → ...
├── extended by: <successor_1>, <successor_2>
└── applied to: <application_1>
## 5. Suggested Reading Order
1. <paper> — <why read this first>
2. <paper> — <builds on #1>
...
Always save the report to a file. Ask user for preferred location, default:
mkdir -p doc/related_works
Save as doc/related_works/<paper_short_name>_related.md.
Every paper entry MUST include a clickable link. Use this format for all tables:
| [Paper Title](https://arxiv.org/abs/XXXX.XXXXX) | 2025 | Relationship | `XXXX.XXXXX` |
If no arXiv ID, use Semantic Scholar link: https://www.semanticscholar.org/paper/<S2_ID>
For papers with code, add repo link:
| [Paper Title](https://arxiv.org/abs/XXXX.XXXXX) | 2025 | Description | [code](https://github.com/...) |
After presenting the map, ask:
/create_skill_with_paper)"/topic_survey)"| Issue | Recovery |
|---|---|
| AlphaXiv 404 | Fall back to reading PDF directly |
| Semantic Scholar rate limit | Wait 60s and retry, or switch to web search |
| Paper not on arXiv | Search by title on Semantic Scholar, Google Scholar |
| Too many successors | Filter: citations > 10, or last 2 years only |
| No successors found | Paper may be very recent; note this and rely on web search |