Genesis Skills

Portable AI-agent workflows for science and engineering

Genesis Skills is a collection of reusable Agent Skills for scientific discovery and engineering workflows. Part of the Genesis Mission, the catalog can be used with Claude Code, Codex, Cursor, OpenCode, Gemini CLI, and LangChain Deep Agents. Individual skills may still depend on client-specific features or external tools.

Skills encode domain expertise, connect to scientific data sources, and provide reproducible patterns for computational research.

---

What Are Skills?

Skills are markdown files (SKILL.md) with YAML frontmatter that teach an AI agent how to handle domain-specific tasks. Each skill provides:

• Domain expertise — Specialized knowledge for literature, HPC, plasma simulation, scientific-data metadata, and AI-system evaluation
• Tool access — Permissions to use specific tools (Bash commands, file operations, web APIs)
• Structured workflows — Step-by-step instructions for complex scientific tasks
• Supporting resources — Scripts, templates, and reference materials the agent can use

Agents can select a skill from a natural-language request. Client-specific menus and explicit invocation shortcuts vary by agent.

---

Quick Start

There’s nothing to compile: clone the repo, then expose the skills to your agent. The repository groups skills by domain (skills///), while agent clients generally expect each skill to be a direct child of a configured skills directory. unpack.sh flattens that layout. LangChain Deep Agents receive a catalog path in code instead of discovering a user-level directory automatically.

git clone https://gitlab.osti.gov/genesis/genesis-skills.git
cd genesis-skills

Mirror repository: The repository https://gitlab.osti.gov/genesis/genesis-skills.git is a read-only mirror of the upstream development repository https://github.com/AI-ModCon/genesis-skills. Please use GitHub for development, forks, issues, and pull requests.

Method 1 — Flatten skills into place (unpack.sh)

Choose the skills directory documented by your client, then let unpack.sh flatten every skill into it:

export SKILLS_DIR="$HOME/.agents/skills"  # Codex, Cursor, OpenCode, Gemini CLI
# export SKILLS_DIR="$HOME/.claude/skills"  # Claude Code
./unpack.sh --target "$SKILLS_DIR" --mode copy

Use --mode symlink for live links back to the clone, ./unpack.sh hpc-skills for one domain, ./unpack.sh --list to preview, or ./unpack.sh --help for all options. The legacy default remains .claude/skills; an explicit --target is the client-neutral choice. Existing same-named destination entries are replaced.

Method 2 — Point the skill-search skill at the catalog

Expose just the bundled skill-search skill; it discovers the whole nested catalog on demand, either via standard progressive disclosure or a keyword search tool:

mkdir -p "$SKILLS_DIR"
ln -s "$PWD/skill-search" "$SKILLS_DIR/skill-search"

skill-search finds the catalog automatically (sibling skills/), or set SKILL_SEARCH_CENTRAL_ROOT / pass --central-root to point at a shared clone.

Use a Skill

Ask naturally, for example: “Find recent papers on machine-learning interatomic potentials” or “Draft a Slurm script for four GPUs.” Your client can select the matching skill from its description. If a newly installed skill is not noticed immediately, follow the client’s documented refresh behavior.

Discovery Helper

• skill-search — Discover and explore available skills by capability

---

Repository Structure

genesis-skills/
├── unpack.sh                  # flatten skills into an agent's skills dir (Method 1)
├── skill-search/              # the upper-level discovery skill (Method 2)
│   ├── SKILL.md
│   └── scripts/skill_search.py
├── skills/
│   ├── academy/SKILL.md
│   ├── amsc-skills/  # amsc-data-movement-api, amsc-python-client, globus-compute, i2-api, iri-api, skill-explorer (6)
│   ├── basedata-skills/  # croissant-validator, datacard-generator, hdmf-schema-builder, skill-creator, well-convert (5)
│   ├── baseeval-skills/  # card-eval-updater, configuration-creator, configuration-implementor, configuration-planner, configuration-tester, data-exploration, perlmutter-nemo-eval, perlmutter-nemo-generate (8)
│   ├── basesafe-skills/  # agentic-vulnerabilities-scan, ai-fingerprint, battleprint, fingerprinting, uncertainty-quantification, uq-metrics-evaluator (6)
│   ├── hpc-skills/  # aurora, frontier, pbs, perlmutter, slurm (5)
│   ├── literature-search/SKILL.md
│   ├── multi-agent-systems/SKILL.md
│   └── plasma-sim-skills/  # gkeyll, gs2 (2)
├── CONTRIBUTING.md
├── LICENSE
├── NOTICE                     # third-party licensing and attribution inventory
├── README.md
└── skill_spec.md              # Agent Skills format specification

Each skill is a directory containing:

• SKILL.md — Required entry point with frontmatter and instructions
• Supporting files — Optional templates, scripts, and reference docs

---

The Genesis Mission

Genesis is building the next generation of AI-powered tools for science and engineering. Our goals:

1. Democratize AI for Science — Make advanced AI capabilities accessible to researchers regardless of ML expertise
2. Accelerate Discovery — Reduce time from hypothesis to insight through intelligent automation
3. Ensure Reproducibility — Build tools that produce reliable, validated, and reproducible results
4. Foster Collaboration — Create a shared ecosystem where domain experts contribute and benefit from collective capabilities

---

Contributing

We welcome contributions from the scientific community. See the Contributing Guide for details on:

• Skill specification format
• Testing guidelines
• Review process

If you are adding a new skill, use the checklist in CONTRIBUTING.md and the pull request template in .github/pull_request_template.md before opening the PR.

Contributors

Individual skills and their directories contain specific attribution details. The catalog-level and skill contributors include the following:

• Matt Baughman
• Jean Luca Bez
• Eric Chagnon
• Jong Choi
• Shreyas Cholia
• Robin Cosbey
• David Florey
• Ian Foster
• Steven Goldenberg
• Nathan Hodas
• Natalie Isenberg
• Olivera Kotevska
• Fernando Llorente
• Kyle Parfrey
• Reilly Raab
• Emily Saldanha
• Malachi Schram
• Andrew Tritt
• Aaron Tuor
• Rohith Varikoti
• Stefan Wild
• American Science Cloud Intelligent Interfaces Team
• ModCon Base CAF Team
• ModCon Base Data Team
• ModCon Base Eval Team
• ModCon Base Safe Team

---

License

This product is licensed under the root LICENSE found at the root of this repository. Some skills under skills/ are sourced from third parties. Where a LICENSE (or LICENSE.txt) file is present in a subdirectory, that license governs the contents of that subdirectory and supersedes the root license for that subtree. The repository-level summary of individual skill and third-party licensing details is in NOTICE; see the attribution file in each identified subtree for the licensing information supplied with that content.

Acknowledgment

This work was supported by the U.S. Department of Energy (DOE), Office of Science, Office of Advanced Scientific Computing Research in alignment with DOE’s Genesis Mission.