name: alfred-search-papers description: > Search arXiv and Google Scholar for AI/ML research papers. Use when: "check arXiv", "new AI papers", "research papers today", "what's new in AI research", "search papers". allowed-tools: Read, Write, Edit, Glob, Grep, WebFetch, WebSearch, mcpclaude-in-chrome*

Alfred Search: Papers

Search academic sources for AI/ML research papers relevant to your interests. Focus on breakthrough papers, not incremental work.

Why This Matters

The frontier of AI moves through papers before it reaches products. Staying current on research gives you:

Early signal on what's coming
Technical depth for better judgment
Ideas for HIHQ and investment theses

Config

Read from: agents/agent-alfred/context/RESEARCH_CONFIG.md → Topics of Interest

Sources

Primary (Always Search)

Source	URL	Focus
arXiv cs.AI	arxiv.org/list/cs.AI/recent	AI general
arXiv cs.LG	arxiv.org/list/cs.LG/recent	Machine learning
arXiv cs.CL	arxiv.org/list/cs.CL/recent	NLP, language models

Secondary (When Relevant)

Source	URL	Focus
Google Scholar Alerts	scholar.google.com	Specific topics
Semantic Scholar	semanticscholar.org	Citation tracking
Papers With Code	paperswithcode.com	Implementation-focused

Filtering Criteria

Include papers that:

Introduce new architectures or techniques
Show significant benchmark improvements (>5%)
Come from top labs (OpenAI, Anthropic, DeepMind, Meta AI, Google Brain)
Have practical implications for products
Challenge existing assumptions

Skip papers that:

Are incremental improvements (<2% gains)
Focus on narrow domains outside interests
Are primarily theoretical with no practical path
Rehash existing work without novelty

Execution

1. WebFetch arXiv recent submissions for cs.AI, cs.LG, cs.CL
2. Search titles and abstracts for relevance
3. For promising papers:
   - Read abstract fully
   - Check authors/institution
   - Note key claims and methods
   - Assess practical implications
4. WebSearch for paper discussions (Twitter, HN, Reddit)
5. Output to research report

Output

Write to agents/agent-alfred/outputs/YYYY-MM-DD-HHmm-search-papers.md:

# Paper Search — YYYY-MM-DD

## Signal Summary

| Signal | Type | Relevance | Tags |
|--------|------|-----------|------|
| [Paper title] | BREAKTHROUGH | HIGH | #ai #inference #[ticker] |
| [Paper title] | NOTABLE | MEDIUM | #llm #agents |

## Breakthrough Papers

#### [Paper Title]
**Authors:** Names (Institution)
**arXiv:** <a href="https://arxiv.org/abs/XXXX.XXXXX" target="_blank">arXiv:XXXX.XXXXX</a>

> One-sentence summary of key contribution

**Key Claims:**
- Claim 1
- Claim 2

**Method:** Brief description of approach

**Results:** Key benchmarks/improvements

**Why It Matters:** Connection to your interests

**Implications:** What this enables or changes

**Discussion:** <a href="url" target="_blank">Twitter/HN thread</a> (if found)

**Researcher Profiles:**
- <a href="scholar-url" target="_blank">Name — Google Scholar</a>
- <a href="twitter-url" target="_blank">@handle — X</a>

---

[Repeat for each significant paper]

### Quick Mentions

Papers worth noting but not deep-diving:

- <a href="url" target="_blank">Title</a> — one-liner
- <a href="url" target="_blank">Title</a> — one-liner

### Sources Searched

- [x] arXiv cs.AI (X papers)
- [x] arXiv cs.LG (X papers)
- [x] arXiv cs.CL (X papers)

Benchmark

Metric	Target
arXiv categories searched	3 (cs.AI, cs.LG, cs.CL)
Breakthrough papers identified	1-3
Quality over quantity	Yes

Success Criteria

arXiv cs.AI, cs.LG, cs.CL searched
Filtering applied (not just listing everything)
Breakthrough papers identified with full analysis
Practical implications stated
Author profiles found for key papers
Output formatted correctly

Notes

Quality over quantity — 2-3 great papers beat 20 mediocre ones
Practical lens — Always ask "what does this enable?"
Follow the labs — OpenAI, Anthropic, DeepMind papers get priority
Check discussions — Twitter threads often have better explanations than abstracts

Enter Passcode