paper-discovery — Claude Code skill: what it does, how to install
paper-discovery runs a literature search across academic databases and keeps the results in one collection you can filter. Useful when you need a reading list, not a single search hit.
Apache-2.0 1 installs ★ 15 updated 2026-03-04
npx skills add acertainKnight/project-thoth --skill paper-discovery -a claude-codeWhat paper-discovery does
This skill searches academic databases and keeps what it finds in one collection. You name a topic, it builds a query, runs it, then shows the hits. Six tools carry the work: list_available_sources lists the databases on offer, create_research_question defines keywords, run_discovery_for_question executes the search, list_articles and search_articles browse and filter the output, collection_stats reports collection size.
Sources include arXiv, PubMed, biorxiv, Semantic Scholar, OpenAlex and Crossref. The default pair is semantic_scholar plus openalex, so for fresh CS preprints you add arxiv to the request yourself. Relevance threshold starts at 0.7. Lower it for wider coverage, raise it past 0.8 to keep only close matches.
When a journal or repository isn't covered, the skill points to custom-source-setup, which wires up scraping for a given site. Deep reading is out of scope: papers go to a Research Analyst agent through send_message_to_agent. And the tools must exist in your setup — the skill only calls them, so a missing tool means an empty result.
What it helps with
- build a reading list on a new topic in one query
- filter results by relevance threshold
- swap in
arxivorpubmedinstead of the default pair - check how many papers the collection already holds
- hand the found papers to a subagent for analysis
Example prompts for the agent
- Find recent papers on sparse attention in vision transformers; use
arxivandsemantic_scholar. - Show me the available sources, then run a discovery query on protein structure prediction.
- Search the collection for AlphaFold and print the stats.
How to install paper-discovery
1Skills CLI (Vercel)
npx skills add acertainKnight/project-thoth --skill paper-discovery -a claude-code2Manually: put the files in place
~/.claude/skills/paper-discovery/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/paper-discovery/
1Skills CLI (Vercel)
npx skills add acertainKnight/project-thoth --skill paper-discovery -a codex2Manually: put the files in place
.agents/skills/paper-discovery/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add acertainKnight/project-thoth --skill paper-discovery -a cursor2Manually: put the files in place
.cursor/skills/paper-discovery/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install acertainKnight/project-thoth src/thoth/.skills/paper-discovery2Manually: put the files in place
.github/skills/paper-discovery/SKILL.md1Skills CLI (Vercel)
npx skills add acertainKnight/project-thoth --skill paper-discovery -a gemini-cli2Platform CLI
gemini skills install https://github.com/acertainKnight/project-thoth --path src/thoth/.skills/paper-discoveryManually: put the files in place
.devin/skills/paper-discovery/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add acertainKnight/project-thoth --skill paper-discovery -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about paper-discovery
How do I install paper-discovery in Claude Code?
Run npx skills add acertainKnight/project-thoth --skill paper-discovery -a claude-code. The skill file lands at ~/.claude/skills/paper-discovery/SKILL.md.
What does paper-discovery do?
It searches academic databases for papers, stores the results in a collection and lets you filter that collection by keyword and relevance.
Can it search a journal that isn't in the source list?
Yes, via the custom-source-setup skill, which sets up scraping for a specific site.
Source text
SKILL.md · Apache-2.0
---
name: paper-discovery
description: Find and curate research papers from academic sources. Use when user
asks to find papers, search for research, discover articles, or explore a new topic.
tools:
- list_available_sources
- create_research_question
- run_discovery_for_question
- list_articles
- search_articles
- collection_stats
---
# Paper Discovery
Find and curate research papers across academic sources (arXiv, PubMed, Semantic Scholar, OpenAlex, etc.).
## Tools to Use
For discovery tasks, use ONLY these tools:
| Tool | Purpose |
|------|---------|
| `list_available_sources` | See available search sources |
| `create_research_question` | Create a new search query |
| `run_discovery_for_question` | Execute the search |
| `list_articles` | Browse results |
| `search_articles` | Filter/search within results |
| `collection_stats` | Check collection size |
## Quick Discovery (5 min)
For a quick search on a topic:
```
Step 1: Create the query
create_research_question(
title="User's topic in 1-2 sentences",
keywords=["keyword1", "keyword2", "keyword3"],
sources=["semantic_scholar", "openalex"],
max_papers=25,
relevance_threshold=0.7
)
Step 2: Run discovery
run_discovery_for_question(question_id="[from step 1]")
Step 3: Review results
list_articles(limit=20, sort_by="relevance")
```
## Source Selection Guide
| Research Area | Recommended Sources |
|---------------|---------------------|
| CS/ML/AI | `arxiv`, `semantic_scholar` |
| Medical/Bio | `pubmed`, `biorxiv` |
| General Science | `openalex`, `crossref` |
| Cross-disciplinary | `semantic_scholar`, `openalex` |
**Default**: Use `semantic_scholar` + `openalex` for broad coverage.
### Need a Source Not Listed?
If the user wants papers from a website/journal not in the built-in sources list:
**Load the `custom-source-setup` skill** to set up auto-detected scrapers for any website. This allows adding sources like:
- Specific journal websites (Nature, Science, PLOS ONE, etc.)
- Conference proceedings pages (ACL Anthology, NeurIPS, etc.)
- Institutional repositories (NBER, SSRN, arXiv mirrors, etc.)
- Any website with article listings
Example trigger phrases:
- "Can you get papers from NBER?"
- "Add Nature Neuroscience as a source"
- "Scrape articles from this URL: https://..."
## Keyword Extraction
Extract keywords from user's request:
1. **Core nouns**: Main concepts (e.g., "transformers", "attention")
2. **Technical terms**: Field-specific language (e.g., "multi-head", "self-attention")
3. **Modifiers**: Scope limiters (e.g., "efficient", "sparse", "2024")
**Example**:
- User: "Find papers on efficient attention mechanisms in vision transformers"
- Keywords: `["vision transformer", "efficient attention", "ViT", "sparse attention"]`
## Relevance Threshold Guide
| Threshold | Use When |
|-----------|----------|
| 0.8+ | User wants only highly relevant papers |
| 0.7 | Default - good balance |
| 0.6 | Comprehensive search, broader coverage |
| 0.5 | Exploratory, casting a wide net |
## When to Delegate to Research Analyst
Delegate using `send_message_to_agent` when user needs:
- Deep analysis of discovered papers
- Quality assessment of results
- Literature synthesis across papers
- Citation network exploration
**Example delegation**:
```
send_message_to_agent(
agent_name="Research Analyst",
message="Analyze these 10 papers on sparse attention and summarize key approaches: [paper IDs]"
)
```
## Workflow Examples
### Example 1: Specific Topic Search
**User**: "Find recent papers on mixture of experts in LLMs"
```
1. create_research_question(
title="Mixture of Experts in Large Language Models",
keywords=["mixture of experts", "MoE", "sparse MoE", "LLM"],
sources=["arxiv", "semantic_scholar"],
max_papers=30,
relevance_threshold=0.75
)
2. run_discovery_for_question(question_id="...")
3. list_articles(limit=15, sort_by="date")
4. Report: "Found X papers on MoE in LLMs. Top 5: [list].
Would you like me to analyze any of these in depth?"
```
### Example 2: Broad Exploration
**User**: "I want to explore what's happening in protein folding research"
```
1. list_available_sources() # Show user options
2. create_research_question(
title="Recent advances in protein structure prediction",
keywords=["protein folding", "AlphaFold", "protein structure prediction"],
sources=["biorxiv", "pubmed", "semantic_scholar"],
max_papers=50,
relevance_threshold=0.65
)
3. run_discovery_for_question(question_id="...")
4. collection_stats() # Show what was found
5. Report summary of results by sub-topic
```
## Error Handling
| Error | Solution |
|-------|----------|
| No results | Lower threshold, broaden keywords, add sources |
| Too many results | Raise threshold, add specific keywords |
| Wrong domain papers | Add negative keywords, change sources |
| Timeout | Reduce sources, lower max_papers |
## Response Template
After discovery, report:
```
## Discovery Results: [Topic]
**Sources searched**: [list]
**Papers found**: [count]
**Relevance threshold**: [value]
### Top Papers:
1. [Title] - [Authors] - [Year]
Brief: [1 sentence description]
2. ...
### Next Steps:
- Would you like me to analyze any of these papers in depth?
- Should I set this up as a recurring search?
- Want me to adjust the search parameters?
```
© acertainKnight · Apache-2.0 · acertainKnight/project-thoth