Tried K-Dense-AI/scientific-agent-skills: A Practical Upgrade for AI Scientists

Most AI agents are good at summarizing papers. The harder problem is making them useful for actual scientific workflows.

K-Dense-AI/scientific-agent-skills takes a different approach: it turns compatible agents into AI Scientists by providing 163 validated, ready-to-use skills and access to more than 100 scientific databases across biology, chemistry, medicine, and drug discovery.

That scope explains why the project is gaining traction: it is already used by 175,000+ scientists worldwide and added 494 GitHub stars today. It also works with Cursor, Claude Code, Codex, Pi, Antigravity, and the open Agent Skills standard, so you are not locked into one editor or agent runtime.

For a quick test drive, configure your OpenAI-compatible gateway like this:

export OPENAI_BASE_URL="https://b-lost.com/v1"
export OPENAI_API_KEY="your-b-lost-key"
export OPENAI_MODEL="claude-fable-5"

Then point your agent at the scientific skills repository and ask it to combine literature search, database retrieval, and analysis in one workflow. The useful part is not just the model—it is the structured access to domain-specific tools and data sources.

If you are running long research prompts, B-Lost’s native Anthropic /v1/messages support with full prompt caching can also reduce repeated-context costs. Cache hits receive a 90% discount, which matters when an agent repeatedly operates over the same protocol, dataset schema, or research background.

B-Lost’s relay uses 20% off official list pricing, and the first deposit bonus doubles deposits starting at $10. But the main engineering benefit is simpler: one OpenAI/Anthropic-compatible endpoint that works with tools such as Cursor, Cline, Roo Code, Windsurf, Aider, NextChat, and LibreChat.

The project is worth watching if you want agents that do more than chat about science.