scientify 1.0.0

package/skills/arxiv/SKILL.md

@@ -0,0 +1,138 @@
+---
+name: arxiv
+description: "Search arXiv.org for academic papers using the built-in arxiv tool. Use for literature search, finding related work, and downloading paper sources."
+metadata:
+  {
+    "openclaw":
+      {
+        "emoji": "📚",
+      },
+  }
+---
+
+# ArXiv Search
+
+Use the built-in `arxiv` tool to search for academic papers on arXiv.org.
+
+## Basic Search
+
+```
+arxiv query:"graph neural network" max_results:10
+```
+
+## Parameters
+
+| Parameter | Type | Description |
+|-----------|------|-------------|
+| `query` | string | Search query (required) |
+| `max_results` | number | Max papers to return (default: 10, max: 50) |
+| `sort_by` | string | Sort by: "relevance", "lastUpdatedDate", "submittedDate" |
+| `date_from` | string | Filter papers after this date (YYYY-MM-DD) |
+| `download` | boolean | Download .tex source files (default: false) |
+| `output_dir` | string | Directory for downloads (default: ~/.openclaw/workspace/papers/) |
+
+## Examples
+
+Search for recent transformer papers:
+```
+arxiv query:"transformer attention mechanism" sort_by:"submittedDate" max_results:5
+```
+
+Search with date filter:
+```
+arxiv query:"diffusion models" date_from:"2024-01-01"
+```
+
+Search and download .tex sources:
+```
+arxiv query:"transformer attention" max_results:5 download:true
+```
+
+## Output
+
+Returns JSON with:
+- `query`: The search query
+- `total`: Number of results
+- `papers`: Array of paper objects with:
+  - `title`: Paper title
+  - `authors`: List of authors
+  - `abstract`: Paper abstract
+  - `arxiv_id`: ArXiv ID (e.g., "2401.12345")
+  - `pdf_url`: Direct PDF link
+  - `published`: Publication date
+  - `categories`: ArXiv categories (e.g., "cs.LG", "cs.AI")
+
+When `download: true`:
+- `downloads`: Array of download results with:
+  - `arxiv_id`: Paper ID
+  - `format`: "tex" or "pdf" (fallback)
+  - `files`: List of downloaded files
+  - `error`: Error message if download failed
+- `output_dir`: Directory where files were saved
+
+---
+
+## Downloading Paper Source (.tex)
+
+**IMPORTANT**: Prefer downloading .tex source over PDF for better AI readability.
+
+### Download .tex Source (Recommended - Use Tool)
+
+The easiest way is to use the `arxiv` tool with `download: true`:
+
+```
+arxiv query:"your search" max_results:5 download:true output_dir:"~/.openclaw/workspace/papers"
+```
+
+This automatically:
+1. Downloads .tex source from `https://arxiv.org/src/{arxiv_id}`
+2. Extracts tar.gz archives
+3. Falls back to PDF if .tex unavailable
+
+### Manual Download (Bash)
+
+If you need to download specific papers manually:
+
+```bash
+mkdir -p ~/.openclaw/workspace/papers/{arxiv_id}
+cd ~/.openclaw/workspace/papers/{arxiv_id}
+curl -L "https://arxiv.org/src/{arxiv_id}" -o source.tar.gz
+tar -xzf source.tar.gz
+```
+
+### Why .tex over PDF?
+
+| Format | AI Readability | Formulas | Structure |
+|--------|---------------|----------|-----------|
+| **.tex** | Excellent | Full LaTeX | Preserved |
+| .pdf | Poor (needs OCR) | Lost/garbled | Lost |
+
+### Fallback to PDF
+
+If .tex source is unavailable (404 error), fall back to PDF:
+```bash
+curl -L "https://arxiv.org/pdf/{arxiv_id}.pdf" -o ~/.openclaw/workspace/papers/{arxiv_id}.pdf
+```
+
+---
+
+## Workspace Integration
+
+**Project-based workspace**: When using with `idea-generation` or `research-pipeline` skills, papers are stored per-project:
+
+```
+~/.openclaw/workspace/projects/
+├── .active                   # Current project ID
+├── {project-id}/             # e.g., nlp-summarization, cv-segmentation
+│   ├── project.json          # Project metadata
+│   ├── papers/               # Downloaded papers for THIS project
+│   │   ├── 2401.12345/       # Extracted .tex source
+│   │   │   ├── main.tex
+│   │   │   └── ...
+│   │   └── 2402.67890.pdf    # PDF fallback
+│   └── ...
+```
+
+**When called from idea-generation**: Use `output_dir: "$WORKSPACE/papers"` where `$WORKSPACE` is the active project directory.
+
+**Standalone usage**: Default `output_dir` is `~/.openclaw/workspace/papers/` (flat structure).