research-watch

name	research-watch
description	Monitor a research topic for new papers, trends, and breakthroughs. Set up recurring watches on arXiv, Semantic Scholar, or specific authors/venues. Use when tracking a fast-moving field, waiting for follow-up work, or maintaining a living literature review. Inspired by Feynman's /watch workflow.
metadata	{"tags":["research","phd","monitoring","alerts","literature","trends"]}

Research Watch

Stay current without drowning in papers.

Set up intelligent monitoring for research topics, authors, venues, or keywords. Get notified when relevant new work appears, with automatic relevance scoring and synthesis.

"The literature moves fast. Your knowledge should move faster." — Anonymous Researcher

---

1. Watch Types

1.1 Topic Watch

Monitor a research topic for new papers:

Topic: "mechanistic interpretability"
Keywords: ["mechanistic interpretability", "circuit tracing", "activation patching"]
Venues: [NeurIPS, ICML, ICLR, arXiv cs.LG]
Frequency: weekly

Relevance scoring:

Score	Criteria	Action
90-100	Directly addresses your research question	Read immediately
70-89	Related method or application	Read within 2 days
50-69	Same field, tangential topic	Skim abstract
< 50	Same broad area	Skip, but log for completeness

1.2 Author Watch

Track specific researchers:

Authors: ["Yann LeCun", "Yoshua Bengio", "Geoffrey Hinton"]
Rationale: "Foundational figures in deep learning"
Alert on: new papers, preprints, blog posts

1.3 Venue Watch

Monitor specific conferences/journals:

Venues: ["NeurIPS", "ICML", "ICLR", "JMLR"]
Filter: papers matching [your keywords]
Alert: when proceedings published

1.4 Citation Watch

Track who cites your work or key papers:

Target papers: [your-paper-id, foundational-paper-id]
Alert on: new citations
Include: citation context (how they cite you)

---

2. Setup Protocol

Step 1 — Define Watch Scope

Watch Name: [descriptive name]
Type: [topic / author / venue / citation]
Query: [keywords, authors, or paper IDs]
Sources: [arXiv / Semantic Scholar / Google Scholar / PubMed]
Frequency: [daily / weekly / monthly]

Step 2 — Configure Filters

Exclude noise:

• Negative keywords: ["survey", "review", "tutorial"] (if you want only original research)
• Minimum citations: [0 for new work, 10 for established work]
• Date range: [last 30 days / last 90 days / all time]

Include priority:

• CCF-A venues: always alert
• arXiv from known authors: always alert
• Preprints with >50 citations in first month: flag as potentially important

Step 3 — Set Up Alerts

arXiv RSS/API:

# arXiv API query
http://export.arxiv.org/api/query?search_query=all:mechanistic+interpretability&sortBy=submittedDate&sortOrder=descending&max_results=10

Semantic Scholar Alerts:

# Semantic Scholar search with date filter
curl "https://api.semanticscholar.org/graph/v1/paper/search?query=mechanistic+interpretability&fields=title,authors,year,abstract,citationCount&publicationDateOrYear=2024:2025"

Google Scholar Alerts:

• Create alert at https://scholar.google.com/scholar_alerts
• Use exact phrase matching: "mechanistic interpretability"

Step 4 — Synthesis Template

When new papers arrive, produce a brief synthesis:

## Research Watch Report: [Topic] — [Date Range]

### New Papers (N found)

#### 🔴 High Relevance
1. **[Title]** by [Authors] ([Venue], [Year])
   - **Why relevant:** [1 sentence]
   - **Key claim:** [1 sentence]
   - **Relation to your work:** [how it connects]
   - **Action:** [read / skim / save for later]

#### 🟡 Medium Relevance
2. ...

#### 🟢 Low Relevance
3. ...

### Trend Analysis
- **Emerging direction:** [what's new this period]
- **Confirming trend:** [what continues from before]
- **Contradicting result:** [paper that challenges consensus]

### Recommended Actions
1. [Read paper X immediately — directly relevant]
2. [Add paper Y to reading list — related method]
3. [Update SOTA matrix with paper Z — new baseline]

---

3. Watch Management

Active Watches Log

Keep a log of all active watches:

## Active Research Watches

| Watch Name | Type | Query | Frequency | Last Check | Papers Found |
|---|---|---|---|---|---|
| MechInterp-2024 | Topic | "mechanistic interpretability" | Weekly | 2024-01-15 | 12 |
| Transformer-Scaling | Topic | "scaling laws" + "transformer" | Weekly | 2024-01-15 | 8 |
| LeCun-Papers | Author | "Yann LeCun" | Daily | 2024-01-15 | 1 |
| NeurIPS-2024 | Venue | NeurIPS 2024 proceedings | Monthly | 2024-01-15 | 0 |

Watch Review Cycle

Monthly:

• Review all active watches
• Deactivate watches that are no longer relevant
• Merge overlapping watches
• Update keywords based on evolving research focus

Quarterly:

• Assess watch effectiveness (signal-to-noise ratio)
• Adjust relevance thresholds
• Add new watches for emerging sub-topics

---

4. Integration with Research Workflow

Watch → SOTA Survey

Research Watch (ongoing)
    ↓ (monthly synthesis)
SOTA Survey Update (sota-survey)
    ↓
Updated Literature Matrix
    ↓
Research Question Refinement (research-question)

Watch → Experiment

Research Watch detects new baseline
    ↓
Paper Audit (paper-audit) — verify claims
    ↓
Add to experiment tracking (experiment-tracking)
    ↓
Run comparison experiment

Watch → Paper

Research Watch detects related work published
    ↓
Update Related Work section (paper-writing)
    ↓
Add citation + differentiate your work

---

5. Tools for Research Watching

Manual (Free)

• arXiv RSS: Subscribe to category feeds
• Google Scholar Alerts: Email alerts for queries
• Semantic Scholar: Weekly digest for followed authors
• Twitter/X: Follow researchers (noisy but fast)

Semi-Automated

• Zotero + RSS: Auto-import papers from feeds
• Notion/Airtable: Track papers with custom views
• Obsidian: Link papers in knowledge graph

Automated (Future)

• LLM-powered filtering: Auto-score relevance, generate summaries
• Slack/Discord bots: Push alerts to team channels
• Auto-synthesis: Generate weekly reports automatically

---

6. Anti-Patterns

Anti-pattern	Problem	Fix
Too many watches	Information overload	Limit to 3-5 active watches
Too broad queries	90% irrelevant papers	Use exact phrases, add negative keywords
Set-and-forget	Watches become stale	Monthly review and cleanup
No synthesis	Papers accumulate unread	Weekly synthesis, not just alerts
Ignoring low-relevance	Miss paradigm shifts	Monthly review of "skipped" papers

---

Integration with Other Skills

This skill provides	Related skill	For deeper dive
New paper discovery	sota-survey	Deep analysis of discovered papers
Paper verification	paper-audit	Check if new papers are trustworthy
Literature update	paper-writing	Update Related Work sections
Trend detection	research-question	Identify new research directions

---

References

• Feynman /watch — Recurring research watch
• arXiv API — Programmatic paper access
• Semantic Scholar API — Academic graph API
• Google Scholar Alerts — Email alerts