技能创建器
skill-creator
创建、改进和评估自定义技能,把专业流程转化为可重复使用的 AI 能力。
image-well【数据与研究】Search and download images from 12 APIs in parallel (Openverse, Pexels, Pixabay, Met Museum, NASA, Rijksmuseum, Wikimedia Commons, Unsplash, Smithsonian, Europeana, Iconify, Pollinations AI). Supports license filtering, presets (military, museum, stock), cached results, and bulk download with metadata sidecars. Triggers on find images, stock photos, …
独立 GitHub 快照 · 自动分类 · 以原作者文档为准
Search and download images from 12 sources through a single CLI. Four sources work with zero API keys (Openverse, Wikimedia, Met Museum, NASA). Additional sources activate when keys are set.
# Search across all no-key sources
uv run ~/.claude/skills/image-well/scripts/well.py search "ancient Minoan fresco"
# Use a preset for domain-specific searches
uv run ~/.claude/skills/image-well/scripts/well.py search "F-35 fighter jet" --preset military
# Download results with metadata sidecars
uv run ~/.claude/skills/image-well/scripts/well.py search "sunset" --format download --output ./images/
# Check which sources are available
uv run ~/.claude/skills/image-well/scripts/well.py sources
# Visual HTML preview — opens in browser, no context pollution
uv run ~/.claude/skills/image-well/scripts/well.py search "bronze statue" --format html
# Output as JSON for piping
uv run ~/.claude/skills/image-well/scripts/well.py search "cat" --format json
| Tier | Source | Key Required | License |
|---|---|---|---|
| 1 | Openverse (800M+) | No | CC variants |
| 1 | Wikimedia Commons | No | CC variants |
| 1 | Met Museum (375k) | No | CC0 |
| 1 | NASA (140k) | No | Public Domain |
| 2 | Pexels | PEXELS_API_KEY | Pexels License |
| 2 | Pixabay | PIXABAY_API_KEY | Pixabay License |
| 2 | Rijksmuseum (700k) | RIJKSMUSEUM_API_KEY | CC0 |
| 2 | Unsplash | UNSPLASH_ACCESS_KEY | Unsplash License |
| 3 | Smithsonian | No | CC0 |
| 3 | Europeana (50M+) | EUROPEANA_API_KEY | Mixed |
| 3 | Iconify (275k icons) | No | Various |
| 3 | Pollinations AI | No | Free Use (AI gen) |
Add Tier 2 keys to ~/.config/env/secrets.env for expanded coverage.
| Preset | Sources | Use Case |
|---|---|---|
military | nasa, wikimedia, smithsonian | Defense/military imagery |
museum | met, rijksmuseum, smithsonian | Art, antiquities, historical |
texture | wikimedia, pollinations | Game dev, 3D materials |
stock | pexels, pixabay, unsplash | Editorial photography |
all-free | openverse, wikimedia, met, nasa, smithsonian | All no-key sources |
--sources NAME [...] Specific sources to search
--preset NAME Use a preset group
--limit N Max results per source (default: 10)
--license LICENSE Filter: cc0, cc-by, cc-by-sa, any
--format FORMAT table, json, download, urls, html, tunnel
--output DIR Download directory (default: ./well-images/)
--no-cache Skip result cache
--timeout N Per-source timeout in seconds (default: 15)
--quiet Suppress progress output
# Search, review in table, then download the good ones
uv run well.py search "Mediterranean landscape" --preset stock --format json > results.json
# Review results.json, then download specific URLs
# Download CC0 images for a research project
uv run well.py search "Bronze Age pottery" --preset museum --license cc0 --format download --output ./corpus/
# Search, download, then resize with image-forge
uv run well.py search "sunset" --format download --output ./raw/
# Then use image-forge skill to process downloaded images
Results are cached for 24 hours at ~/.cache/image-well/.
uv run well.py cache stats # Show cache size
uv run well.py cache clear # Clear all cached results