Add skills for getting arxiv papers and wechat articles about VLA

nanobot/skills/arxiv-vla-papers/SKILL.md

@@ -0,0 +1,225 @@
+---
+name: arxiv-vla-papers
+description: Query and summarize latest Vision-Language-Action (VLA) papers from arXiv. Use when user asks for recent VLA papers, robotic manipulation papers, vision-language-action research, or wants to track latest developments in embodied AI. Supports filtering by date range (today, 2 days, 1 week) and categories (AI, CV, HCI, ML, Robotics). Automatically summarizes papers focusing on innovations, methods, and differences from prior work.
+metadata: {"nanobot":{"emoji":"📄","requires":{"bins":["python3","jq"]}}}
+---
+
+# arXiv VLA Papers
+
+Query and summarize the latest Vision-Language-Action (VLA) papers from arXiv, focusing on robotic manipulation and navigation.
+
+## Overview
+
+This skill queries arXiv for recent Vision-Language-Action papers, filters by relevant categories and keywords, and provides concise summaries highlighting innovations, methods, and differences from prior work.
+
+## Quick Start
+
+### Query recent papers
+
+The script is located at `scripts/arxiv_query.py` relative to this skill directory. Use the full path to the script:
+
+```bash
+python3 nanobot/skills/arxiv-vla-papers/scripts/arxiv_query.py --days 7 --max-results 10
+```
+
+Or if the skill is in workspace skills, use:
+```bash
+python3 skills/arxiv-vla-papers/scripts/arxiv_query.py --days 7 --max-results 10
+```
+
+**Note**: The agent can determine the correct path by reading this SKILL.md file location and constructing the path to `scripts/arxiv_query.py`.
+
+### Get JSON output for processing
+
+```bash
+python3 nanobot/skills/arxiv-vla-papers/scripts/arxiv_query.py --days 2 --json > papers.json
+```
+
+## Workflow
+
+### 1. Query arXiv for papers
+
+Use the script to fetch papers matching VLA criteria:
+
+```bash
+# Last 7 days (default)
+python3 nanobot/skills/arxiv-vla-papers/scripts/arxiv_query.py --days 7
+
+# Last 2 days
+python3 nanobot/skills/arxiv-vla-papers/scripts/arxiv_query.py --days 2
+
+# Today only (1 day)
+python3 nanobot/skills/arxiv-vla-papers/scripts/arxiv_query.py --days 1
+
+# More results
+python3 nanobot/skills/arxiv-vla-papers/scripts/arxiv_query.py --days 7 --max-results 30
+```
+
+The script searches for papers containing:
+- Keywords: vision-language-action, robotic manipulation, language-conditioned, instruction following, embodied AI
+- Categories: cs.AI, cs.CV, cs.HC, cs.LG, cs.RO
+
+### 2. Fetch paper details
+
+For each paper, use `web_fetch` to get the abstract page:
+
+```json
+{
+  "url": "https://arxiv.org/abs/2401.12345",
+  "extractMode": "markdown",
+  "maxChars": 5000
+}
+```
+
+Or fetch the PDF if full text is needed:
+
+```json
+{
+  "url": "https://arxiv.org/pdf/2401.12345.pdf",
+  "extractMode": "text",
+  "maxChars": 20000
+}
+```
+
+### 3. Summarize papers
+
+Use the `summarize` skill to extract key insights:
+
+```bash
+summarize "https://arxiv.org/abs/2401.12345" --model google/gemini-3-flash-preview --length medium
+```
+
+Or use LLM directly with a focused prompt:
+
+```
+Summarize this VLA paper focusing on:
+1. Main innovation/contribution
+2. Key methods and approach
+3. How it differs from prior work
+4. Experimental results (if mentioned in abstract)
+```
+
+## Customization
+
+### Adjust date range
+
+```bash
+# Last week
+python3 nanobot/skills/arxiv-vla-papers/scripts/arxiv_query.py --days 7
+
+# Last 2 days
+python3 nanobot/skills/arxiv-vla-papers/scripts/arxiv_query.py --days 2
+
+# Today only
+python3 nanobot/skills/arxiv-vla-papers/scripts/arxiv_query.py --days 1
+
+# No date filter (all papers)
+python3 nanobot/skills/arxiv-vla-papers/scripts/arxiv_query.py --no-date-filter
+```
+
+### Custom categories
+
+```bash
+python3 nanobot/skills/arxiv-vla-papers/scripts/arxiv_query.py --categories cs.RO cs.CV
+```
+
+### Additional keywords
+
+```bash
+python3 nanobot/skills/arxiv-vla-papers/scripts/arxiv_query.py --keywords "mobile manipulation" "dexterous manipulation"
+```
+
+## Output Format
+
+The script returns papers with:
+- Title
+- Authors
+- Publication date
+- Abstract
+- arXiv URL (abs and PDF)
+- Categories
+
+## Integration with Other Skills
+
+### Use with summarize
+
+After fetching paper URLs, use `summarize` for detailed analysis:
+
+```bash
+# Get paper list
+python3 nanobot/skills/arxiv-vla-papers/scripts/arxiv_query.py --days 7 --json > papers.json
+
+# Summarize each paper
+for url in $(jq -r '.[].abs_url' papers.json); do
+  summarize "$url" --model google/gemini-3-flash-preview --length medium
+done
+```
+
+### Use with web_fetch
+
+Fetch full abstracts or PDFs for deeper analysis:
+
+```bash
+# Get paper URLs
+python3 nanobot/skills/arxiv-vla-papers/scripts/arxiv_query.py --days 2 --json | jq -r '.[].abs_url' | head -5
+```
+
+## Focus Areas
+
+When summarizing papers, emphasize:
+
+1. **Innovation**: What's new? What problem does it solve?
+2. **Method**: How does it work? Architecture, training, inference?
+3. **Differences**: How does it differ from prior work? What's the key improvement?
+4. **Results**: Performance metrics, benchmarks, real-world validation (if available in abstract)
+
+## Example Workflow
+
+### Complete workflow: Query → Fetch → Summarize
+
+```bash
+# 1. Query recent papers (last 7 days)
+python3 nanobot/skills/arxiv-vla-papers/scripts/arxiv_query.py --days 7 --max-results 10 --json > papers.json
+
+# 2. Extract top 5 papers
+jq '.[:5]' papers.json > top5.json
+
+# 3. For each paper, fetch and summarize
+for url in $(jq -r '.[].abs_url' top5.json); do
+  echo "=== $url ==="
+  summarize "$url" --model google/gemini-3-flash-preview --length medium
+  echo ""
+done
+```
+
+### Quick summary of today's papers
+
+```bash
+# Get today's papers and summarize each
+python3 nanobot/skills/arxiv-vla-papers/scripts/arxiv_query.py --days 1 --json | \
+  jq -r '.[].abs_url' | \
+  while read url; do
+    echo "Summarizing: $url"
+    summarize "$url" --model google/gemini-3-flash-preview --length short
+    echo "---"
+  done
+```
+
+### Focused analysis with custom prompt
+
+For each paper URL, use `web_fetch` to get the abstract, then ask LLM to summarize with focus on innovations:
+
+1. Fetch abstract: `web_fetch` with `url: "https://arxiv.org/abs/XXXXX"`
+2. Summarize with prompt: "Focus on: (1) main innovation, (2) key methods, (3) differences from prior work"
+
+## References
+
+- `references/arxiv_api.md` - arXiv API documentation and query syntax
+
+## Notes
+
+- arXiv API has no authentication but rate limit ~1 request per 3 seconds
+- Abstract pages are usually sufficient for summarization
+- PDFs can be fetched for full text analysis but are larger
+- Categories are automatically filtered to VLA-relevant areas
+- Keywords focus on robotic manipulation and navigation contexts

nanobot/skills/arxiv-vla-papers/references/arxiv_api.md

@@ -0,0 +1,74 @@
+# arXiv API Reference
+
+## API Endpoint
+
+Base URL: `http://export.arxiv.org/api/query`
+
+## Query Parameters
+
+- `search_query`: Search query using arXiv query syntax
+- `start`: Starting index (0-based)
+- `max_results`: Maximum number of results (default: 10, max: 2000)
+- `sortBy`: Sort field (`relevance`, `lastUpdatedDate`, `submittedDate`)
+- `sortOrder`: `ascending` or `descending`
+- `submittedDate`: Date range filter (format: `YYYYMMDDHHMMSS*YYYYMMDDHHMMSS`)
+
+## Query Syntax
+
+### Field Queries
+
+- `all:term` - Search in all fields
+- `ti:term` - Title
+- `au:term` - Author
+- `abs:term` - Abstract
+- `cat:term` - Category
+- `co:term` - Comment
+
+### Boolean Operators
+
+- `AND` - Both conditions must match
+- `OR` - Either condition matches
+- `ANDNOT` - First matches, second doesn't
+
+### Category Codes
+
+- `cs.AI` - Artificial Intelligence
+- `cs.CV` - Computer Vision and Pattern Recognition
+- `cs.HC` - Human-Computer Interaction
+- `cs.LG` - Machine Learning
+- `cs.RO` - Robotics
+
+## Example Queries
+
+```
+# Vision-language-action papers
+all:"vision language action" AND (cat:cs.AI OR cat:cs.CV OR cat:cs.RO)
+
+# Robotic manipulation
+all:"robotic manipulation" AND cat:cs.RO
+
+# Recent papers (last 7 days)
+all:"VLA" AND submittedDate:[20250101*20250108]
+```
+
+## Response Format
+
+Returns Atom XML with entries containing:
+- `<id>` - arXiv ID
+- `<title>` - Paper title
+- `<author>` - Author names
+- `<summary>` - Abstract
+- `<published>` - Publication date
+- `<category>` - Subject categories
+- `<link>` - PDF and abstract URLs
+
+## Rate Limits
+
+- No authentication required
+- Rate limit: ~1 request per 3 seconds recommended
+- Max results per query: 2000
+
+## References
+
+- [arXiv API User Manual](https://arxiv.org/help/api/user-manual)
+- [arXiv Query Syntax](https://arxiv.org/help/api/user-manual#query_details)