Scientific Agent Skills
github.com/k-dense-ai/scientific-agent-skills- Category
- AI Agents
- Rank
- No. 23Tools index
- Listed in
- #3 Research with an agent
- Pricing
- Open Source
- Platform
- cli
- Type
- TOOL
- Builder
- k-dense-ai
- GitHub
- 43.9k stars
- Latest release
- v2.66.0
- Date
About
A comprehensive library of 142 ready-to-use scientific skills that transforms any AI agent into an AI scientist. Provides curated tools for bioinformatics, drug discovery, clinical research, and more, with access to 100+ scientific databases.
What it does
Scientific Agent Skills supplies structured playbooks that teach compatible coding agents how to perform specific research tasks. Agents select a skill from its description, then follow documented methods, examples, safety boundaries, and sometimes bundled helper scripts. Coverage spans scientific data access, analysis packages, laboratory systems, evidence work, modeling, visualization, and research communication.
Why it's ranked here
The case is strong because the collection combines broad domain coverage with unusually concrete workflow discipline. Examples require decision criteria, output contracts, provenance, uncertainty, negative results, checkpoints, and validation controls. The repository also documents security failures and their fixes. Still, this is guidance for agents, not an independently validated scientific execution environment.
What's good
The worked examples teach habits that matter: separate retrieval from inference, report applicability limits, preserve intermediates, and test expensive methods against trusted controls. Database guidance emphasizes deterministic queries, provenance, pagination, and rate limits. Skills containing bundled tooling require tests. Sensitive workflows add approval gates for remote writes, spending, data transfer, and physical equipment actions.
Tradeoffs
Many skills depend on external packages, public services, credentials, network access, or specialized platforms. The repository requires a recent Python runtime and installs scanner, crawling, testing, and environment-management dependencies. Skill totals also differ across supplied documentation, which weakens inventory clarity. Clinical content explicitly excludes patient-specific care, while regulatory material cannot replace certification, accreditation, or qualified release decisions.
How to use it well
It suits researchers and scientific engineers already working through a compatible coding agent. Name the intended skills, define thresholds and deliverables, demand provenance and uncertainty, save checkpoints, and validate against a known control. Review plans before any remote mutation or physical action. It does not supply patient care, regulatory approval, or automatic authority over laboratory systems.
Technical notes+
pyproject.toml defines a Python project requiring Python 3.13 or newer, with cisco-ai-skill-scanner, firecrawl-py, pytest, and python-dotenv; pytest targets tests with importlib mode. README.md presents the repository as both an Agent Skills collection and an Agent Plugins package using plugin.json plus skills/. scan_skills.py hashes complete skill directories, runs behavioral, trigger, and LLM analyzers concurrently, caches unchanged results, and emits Markdown plus JSON reports. scan_pr_skills.py scans changed skill directories, can block by severity, and skips LLM scanning when its API key is unavailable. docs/security-triage.md records verified vulnerabilities, fixes, and scanner false positives, while warning that the automatically published report receives no pre-publication plausibility check. docs/examples.md supplies composed workflows and explicit safety gates.
Observed
- License
- MIT
- Primary language
- Python
- Runtime requirement
- Python 3.13 or newer
- Packaging surface
- Python project, Agent Skills collection, and portable Agent Plugins package
- Supported agent clients
- Cursor, Claude Code, Codex, Google Antigravity, and other Agent Skills-compatible clients
- Testing structure
- Pytest targets a tests directory; bundled skill scripts are required to have tests
- Security tooling
- Repository includes full-collection and pull-request skill scanners with behavioral, trigger, and LLM analysis
Read from README.md, pyproject.toml, docs/skills.md, docs/examples.md, docs/security-triage.md, docs/open-source-sponsors.md, scan_skills.py, scan_pr_skills.py, skills/docx/scripts/__init__.py, skills/gtars/scripts/__init__.py, skills/geniml/scripts/__init__.py, skills/openpiv/scripts/__init__.py, skills/pydicom/scripts/__init__.py, skills/fluidsim/scripts/__init__.py.
What it can do
Query bioinformatics databases for genetic information
Gene names, protein IDs, or genomic coordinates → Genetic sequences, annotations, and metadata
Perform drug discovery compound screening
Molecular structures or compound databases → Drug candidate predictions and binding affinity scores
Analyze cancer genomics data
Genomic datasets and mutation profiles → Cancer biomarker identification and pathway analysis
Execute clinical research data analysis
Clinical trial data and patient records → Statistical analysis results and clinical insights
Process geospatial scientific data
Geographic coordinates and spatial datasets → Geospatial analysis results and visualizations
Access and retrieve data from scientific databases
Database queries and search parameters → Scientific data records and research datasets
Execute automated scientific workflows
Workflow parameters and input data → Processed scientific results and reports
Intel on Scientific Agent Skills
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.