Paper Claw sends personalized daily research digests from arXiv and beyond straight to your inbox, featuring customizable categories, intelligent classification, and agent-based multilingual summaries powered by your preferred AI via private API. Designed for researchers and AI a
SaferSkills independently audited paper_claw (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Intelligent multi-source paper digest generator. Automatically fetch, classify, and summarize papers with AI-powered translations in 7 languages.
Required for email delivery:
export SMTP_HOST="smtp.qq.com"
export SMTP_PORT="465"
export SMTP_USER="[email protected]"
export SMTP_PASS="your-auth-code"Optional for AI summaries (multiple providers supported):
# Primary: Kimi AI (recommended for Chinese)
export MOONSHOT_API_KEY="sk-your-kimi-key"
# Alternatives (auto-fallback)
export OPENAI_API_KEY="sk-your-openai-key"
export ANTHROPIC_API_KEY="sk-your-claude-key"
export GOOGLE_API_KEY="your-gemini-key"
export DEEPSEEK_API_KEY="sk-your-deepseek-key"Create config/recipients.json:
{
"recipients": [
{"email": "[email protected]", "name": "Professor", "enabled": true},
{"email": "[email protected]", "name": "Student", "enabled": true}
]
}Edit config/default.json to customize sources:
{
"sources": {
"arxiv": {
"enabled": true,
"categories": [
{"id": "cs.CL", "name": "NLP", "url": "https://arxiv.org/list/cs.CL/recent"},
{"id": "cs.CV", "name": "Computer Vision", "url": "https://arxiv.org/list/cs.CV/recent"}
]
}
}
}See config/arxiv_categories.json for all 170+ available categories.
{
"language": {
"default": "zh",
"supported": ["zh", "en", "ja", "ko", "de", "fr", "es"]
}
}The fastest way to configure Paper Claw is using Presets:
from skill.example import list_presets, preview_preset, apply_preset
# Step 1: See available presets
presets = list_presets()
# Returns: [
# {"id": "speech_audio", "name": "Speech & Audio", ...},
# {"id": "nlp", "name": "NLP & LLM", ...},
# {"id": "computer_vision", "name": "Computer Vision", ...},
# {"id": "general_ai", "name": "General AI/ML", ...}
# ]
# Step 2: Preview what will be configured
preview = preview_preset("nlp")
# Shows: arXiv categories (cs.CL, cs.LG) and classification categories (LLM, RAG, etc.)
# Step 3: Apply the preset
apply_preset("nlp") # Updates config/default.json automatically| Preset ID | Research Field | ArXiv Categories | Classification |
|---|---|---|---|
speech_audio | Speech & Audio | cs.SD, eess.AS | Speech LLM, ASR, TTS, Enhancement, SLU, Paralinguistics, Audio |
nlp | NLP & LLM | cs.CL, cs.LG, cs.AI | LLM, RAG, Agents, NLP Tasks, Evaluation |
computer_vision | Computer Vision | cs.CV, cs.MM, cs.LG | Image Generation, Object Detection, Segmentation, Video Understanding, Multimodal, 3D Vision |
general_ai | General AI/ML | cs.AI, cs.LG, cs.CL, cs.CV, stat.ML | Deep Learning, RL, Generative Models, Optimization, Theory, Applications |
from skill.example import list_presets
presets = list_presets()
for p in presets:
print(f"{p['id']}: {p['name']}")
print(f" {p['description']}")from skill.example import preview_preset
# See what will be configured
preview = preview_preset("computer_vision")
print(f"ArXiv categories: {[c['id'] for c in preview['arxiv_categories']]}")
print(f"Classifications: {[c['name'] for c in preview['classification_categories']]}")from skill.example import apply_preset
# Apply NLP configuration
result = apply_preset("nlp")
if result["success"]:
print(f"Applied: {result['preset_name']}")
print(f"ArXiv: {result['arxiv_categories']}")
print(f"Categories: {result['classification_categories']}")# Fetch today's papers (default language from config)
python scripts/main.py
# Fetch with specific language
python scripts/main.py --day 2026-03-10 --language en
python scripts/main.py --day 2026-03-10 --language ja # Japanese
# Fetch date range
python scripts/main.py --start-date 2026-03-01 --end-date 2026-03-10content/posts/YYYY-MM-DD-arxiv-audio-digest.mddata/processed/YYYY-MM-DD.jsondata/raw/YYYY-MM-DD.jsonEmail is automatically sent with:
GitHub Actions: Already configured in .github/workflows/daily_digest.yml
Linux/Mac Cron:
0 1 * * * cd /path/to/paper_claw && python scripts/main.pyWindows Task Scheduler:
$Action = New-ScheduledTaskAction -Execute "python.exe" -Argument "scripts/main.py"
$Trigger = New-ScheduledTaskTrigger -Daily -At "09:00"
Register-ScheduledTask -TaskName "PaperClaw" -Action $Action -Trigger $TriggerThe system uses intelligent fallback across providers:
Kimi → OpenAI → Claude → DeepSeek → Gemini → Rule-basedEven without API keys, summaries are generated using rule-based methods.
Fetch papers from configured sources.
Parameters:
day (string, optional): Date in YYYY-MM-DD formatstart_date + end_date (string, optional): Date rangelanguage (string, optional): Output language (zh/en/ja/ko/de/fr/es)Example:
from skill.example import fetch_papers
result = fetch_papers(day="2026-03-10", language="en")Update data sources and categories.
Parameters:
sources (object): Source configuration with categoriesExample:
from skill.example import configure_sources
configure_sources({
"arxiv": {
"enabled": True,
"categories": [
{"id": "cs.AI", "name": "AI"},
{"id": "cs.LG", "name": "ML"}
]
}
})Set output language for summaries.
Parameters:
language (string): One of zh/en/ja/ko/de/fr/esExample:
from skill.example import configure_language
configure_language("ja") # Japanese outputRetrieve generated digest.
Parameters:
date (string): Date in YYYY-MM-DD formatformat (string): "markdown", "json", or "summary"Example:
from skill.example import get_digest_content
content = get_digest_content("2026-03-10", format="summary")Update email recipients.
Parameters:
recipients (array): List of {email, name, enabled}Example:
from skill.example import configure_recipients
configure_recipients([
{"email": "[email protected]", "name": "User", "enabled": True}
])Best for: Speech recognition, synthesis, audio processing researchers
ArXiv Categories:
cs.SD - Sound (Audio processing, music computing)eess.AS - Audio and Speech ProcessingClassification:
| Category | Keywords |
|---|---|
| Speech LLM | speech llm, audio llm, spoken language model |
| ASR | asr, speech recognition, speech-to-text, whisper |
| TTS | tts, text-to-speech, speech synthesis, tacotron |
| Enhancement | speech enhancement, noise reduction, beamforming |
| SLU | spoken language understanding, intent recognition |
| Paralinguistics | emotion recognition, speaker verification |
| Audio | audio classification, sound event detection |
Best for: Natural language processing, large language model researchers
ArXiv Categories:
cs.CL - Computation and Languagecs.LG - Machine Learningcs.AI - Artificial IntelligenceClassification:
| Category | Keywords |
|---|---|
| LLM | llm, gpt, transformer, prompt engineering, llama, bert |
| RAG | rag, retrieval-augmented, knowledge base, embedding |
| Agents | agent, multi-agent, tool use, function calling |
| NLP Tasks | ner, sentiment analysis, translation, summarization |
| Evaluation | benchmark, evaluation metrics, human evaluation |
Best for: Computer vision, image processing, multimodal researchers
ArXiv Categories:
cs.CV - Computer Visioncs.MM - Multimediacs.LG - Machine LearningClassification:
| Category | Keywords |
|---|---|
| Image Generation | diffusion model, gan, stable diffusion, text-to-image |
| Object Detection | yolo, rcnn, ssd, bounding box |
| Segmentation | semantic segmentation, mask, sam, u-net |
| Video Understanding | action recognition, temporal, tracking |
| Multimodal | vision-language, clip, image-text, vqa |
| 3D Vision | point cloud, depth estimation, nerf |
Best for: Broad AI/ML research covering multiple domains
ArXiv Categories:
cs.AI, cs.LG, cs.CL, cs.CV, stat.MLClassification:
| Category | Keywords |
|---|---|
| Deep Learning | neural network, optimization, gradient descent |
| Reinforcement Learning | rl, q-learning, policy gradient, actor-critic |
| Generative Models | gan, vae, diffusion, flow-based |
| Optimization | convex optimization, learning rate, adam |
| Theory | generalization, convergence, bounds, complexity |
| Applications | healthcare, finance, robotics, real-world |
After applying a preset, you can further customize:
from skill.example import configure_sources, configure_categories
# Add more arXiv categories
configure_sources({
"arxiv": {
"enabled": True,
"categories": [
{"id": "cs.IR", "name": "Information Retrieval",
"url": "https://arxiv.org/list/cs.IR/recent"}
]
}
})
# Add custom classification category
configure_categories([
{
"name": "Your Custom Category",
"labels": {"zh": "自定义分类", "en": "Custom"},
"keywords": ["keyword1", "keyword2"]
}
])| Service | Host | Port | Note |
|---|---|---|---|
| QQ Mail | smtp.qq.com | 465 | Use authorization code |
| 163 Mail | smtp.163.com | 465 | Use authorization code |
| Gmail | smtp.gmail.com | 465 | Use app password |
config/ directory.env and config/recipients.json are git-ignored for securitydata/state.json to avoid duplicate processing# Quick start - fetch and send email
python scripts/main.py --day 2026-03-10
# Multi-language examples
python scripts/main.py --day 2026-03-10 --language zh # Chinese
python scripts/main.py --day 2026-03-10 --language en # English
python scripts/main.py --day 2026-03-10 --language ja # Japanese
# View paper count
cat data/processed/2026-03-10.json | jq '.summary.total'
# View papers by category
cat data/processed/2026-03-10.json | jq '.grouped.ASR'
# Reset state and re-fetch
python scripts/reset_state.py
python scripts/main.py --day 2026-03-10skill/tools.json — Tool definitions for agent frameworksskill/example.py — Python usage examplesconfig/default.json — Source and language configurationconfig/arxiv_categories.json — Complete arXiv category listconfig/recipients.example.json — Recipient template~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.