Scientific Agent Skills
Turn any AI agent into an AI Scientist: 159 ready-to-use skills and 100+ scientific databases covering biology, chemistry, medicine, and drug discovery.
Evidence: README explicitly warns skills can execute code, advises review and selective installation; SECURITY.md defines vulnerability reporting and scanning; source attribution exists (e.g., Anthropic's document skills). Deductions: no concrete least-privilege implementation, user confirmation mechanism not explicit, data flow transparency lacking, sensitive data handling only mentions avoiding credential leaks, dependency security only mentions scanning without lockfiles, external effects not clearly documented, rollback mechanism entirely absent.
Evidence: test contracts (tests/_contract) define structure, CLI, and shared file tests ensuring consistency; failure messages well-designed, e.g., CLI tests report timeouts and missing dependencies. Deductions: dependency availability not fully verified, only mentioned tests run in isolated environment; no dependency version pinning or compatibility matrix.
Evidence: README clearly defines target audience (scientists, engineers) and multiple use cases (drug discovery, single-cell analysis); capability boundaries stated, e.g., medical skills research-only; environment fit good, supports multiple agent hosts and OS. Deductions: trigger precision insufficient, no explanation of how skills are triggered beyond manual invocation; no detailed mechanism for skill selection or context management.
Evidence: clear information architecture with TOC and categories; detailed install instructions with multiple methods; naming stable, skill names consistent; examples and FAQ abundant; known limitations mentioned in security disclaimer; MIT license explicit; versioning and changelog auto-generated via release.yml; maintenance responsibility clear, maintained by K-Dense team. Deductions: naming stability not explicitly promised, but version control exists.
Evidence: high output usability with code examples and docs; high marginal value saving API documentation research; reasonable cost-benefit, open-source free. Deductions: no actual run results or performance data, only static assessment.
Evidence: some claims in README have links (e.g., security scan badge), but no concrete evidence; cross-source corroboration insufficient, relies on single repo; fact-inference separation unclear, e.g., claim '170,000+ scientists' without source. Deductions: weak claim traceability, lack of specific data or citations.
- Skills can execute arbitrary code; review SKILL.md and scripts before installation.
- No rollback mechanism provided; difficult to undo after installation.
- Dependency security relies only on scanning, no lockfiles, supply chain risk exists.
- Claims like '170,000+ scientists' lack verifiable sources.
What does this agent do, and when should you use it?
Scientific Agent Skills is an MIT-licensed open-source skill library by K-Dense that provides 159 ready-to-use Agent Skills and unified access to 100+ scientific databases. These skills follow the open Agent Skills standard and are compatible with major coding agents such as Cursor, Claude Code, Codex, and Google Antigravity, enabling agents to execute complex multi-step scientific workflows. The skills span bioinformatics (e.g., single-cell RNA-seq, genomics), cheminformatics (e.g., molecular property prediction, virtual screening), proteomics, clinical research, medical imaging, machine learning, materials science, and more. Each skill includes comprehensive documentation, code examples, best practices, and test suites, with regular security scanning. The library is community-maintained and includes specialized tools like Paperclip for full-text literature retrieval, database-lookup, and PK/PD modeling.
The library provides predefined skills, each containing a SKILL.md, code examples, and tests. After installation, agents can discover and invoke skills as needed. For instance, the database-lookup skill provides deterministic REST API access to 78 public databases including PubChem, ChEMBL, and UniProt. Other skills like scanpy for single-cell analysis, rdkit for molecular manipulation, and paper-lookup for literature search are included. Skills typically use Python and the uv package manager for dependencies, and may interact with external APIs. Outputs include visualizations, research reports, virtual screening results, and more.
- Drug discovery researchers: query ChEMBL for EGFR inhibitors, perform SAR analysis with RDKit, and virtual screening with DiffDock.
- Bioinformaticians: analyze single-cell RNA-seq data, perform differential expression and gene regulatory network inference.
- Clinical researchers: retrieve clinical trials and drug safety data, perform PK/PD modeling and dose regimen evaluation.
- Materials scientists: use pymatgen and Materials Project for crystal structure analysis and property prediction.
- Research writers: use scikit-learn and statsmodels for statistical analysis and generate scientific figures.
- Lab automation engineers: design and simulate liquid handling protocols with Opentrons skill.
What are this agent's strengths and limitations?
- Broad coverage: 159 skills and 100+ databases across multiple scientific domains.
- Standardized: follows the open Agent Skills standard, compatible with multiple coding agents.
- High quality: each skill includes documentation, examples, and tests, with security scanning.
- Community-driven: maintained by K-Dense with active community contributions.
- Context overhead: installing all skills may consume significant context, so install subsets as needed.
- Dependency complexity: skills depend on numerous Python packages and require uv, adding environment setup overhead.
- Security risk: community-contributed skills may not be deeply reviewed; users must evaluate themselves.
- Learning curve: requires understanding Agent Skills standard and skill installation configuration.
How do you install or deploy this agent?
Recommended: use npx: npx skills add K-Dense-AI/scientific-agent-skills. Alternatively, use GitHub CLI (v2.90.0+): gh skill install K-Dense-AI/scientific-agent-skills. For other hosts supporting Agent Skills, manually clone the repo to ~/.agents/skills/ or .agents/skills/ directory. Requires Python 3.13+ and uv package manager.
How do you use this agent?
After installation, agents can discover skills from configured paths. Users can manually invoke a skill by mentioning its name in the prompt, e.g., 'Use the scanpy skill to analyze this single-cell data'. Skills automatically install dependencies and execute workflows. It is recommended to read the relevant SKILL.md for specific usage and caveats.