🌍 Geoscience Skills

Portable geoscience skills for Codex, Claude Code, GitHub Copilot, Gemini CLI, Windsurf, OpenCode, Cline, Roo Code, OpenClaw, and other coding agents using the Agent Skills format.

39 domain skills + 8 workflows + 1 discovery skill. The same skill files are shared across agents; platform commands, hooks, and subagents are optional. See the verification results and compatibility matrix.

---

🎯 Capabilities at a Glance

---

⚡ Quick Examples

"Read this SEG-Y file and show the first 10 traces"
"Calculate water saturation from these well logs using Archie equation"
"Build a 3D geological model with two faulted horizons"
"Invert this ERT survey and plot the resistivity section"
"Compute variogram and run kriging on this spatial data"
"Load this NetCDF climate file and compute monthly anomalies"
"Create a chondrite-normalized REE spider diagram"

---

📦 Installation

Upstream skills CLI (recommended)

Run from the project where you want to use the skills. The installer supports multiple coding agents and manages their destination directories.

# Inspect all domain, workflow, and routing skills
npx skills add SteadfastAsArt/geoscience-skills --full-depth --list

# Select skills and target agents interactively
npx skills add SteadfastAsArt/geoscience-skills --full-depth

# Install a starting set for multiple agents
npx skills add SteadfastAsArt/geoscience-skills --full-depth \
  --skill using-geoscience-skills lasio segyio --agent codex claude-code

# Install the whole collection for one agent
npx skills add SteadfastAsArt/geoscience-skills --full-depth \
  --skill '*' --agent gemini-cli

# Install a workflow and the domain skills needed for it
npx skills add SteadfastAsArt/geoscience-skills --full-depth \
  --skill well-log-evaluation lasio dlisio welly petropy striplog pyvista \
  --agent github-copilot

Keep --full-depth to include workflows nested under workflows/. Add --global for user scope; use --copy if you prefer copies to symlinks. Choosing a workflow does not automatically install its complementary skills or Python dependencies. Other targets include windsurf, opencode, cline, roo, and openclaw; consult the upstream supported-agent list.

npx skills list
# Explicitly update installed skills when desired
npx skills update

From a local checkout

git clone https://github.com/SteadfastAsArt/geoscience-skills.git
# Run this in your target project; replace the path with your checkout location.
npx skills add /path/to/geoscience-skills --full-depth --agent codex

For manual installation, copy each selected skill directory to a directory supported by your agent. Preserve its SKILL.md, references/, and scripts/. For a workflow, copy workflows// as / alongside the domain skills. Do not copy the whole repository into an agent’s skill directory. Platform configuration and optional role guides are separate from skill installation. See platform notes and the manual reading fallback.

Python dependencies

Install the packages needed for your task in its Python environment. For example:

# LAS inspection and conversion
python -m pip install lasio pandas

# SEG-Y inspection and subsetting
python -m pip install segyio numpy

Skill metadata.dependencies records library requirements; installing a skill does not install or validate those packages. See SKILLS.md for additional domain-specific package lists.

---

🧠 39 Domain Skills

By Domain

Seismic & Seismology     → obspy, segyio, segysak, disba
Well Logs & Petrophysics → lasio, welly, dlisio, striplog, petropy
3D Geological Modelling  → gempy, loopstructural, gemgis, geolime (licensed)
Geophysical Inversion    → simpeg, devito, pylops, pygimli, discretize
Potential Fields         → harmonica, boule
Rock Physics             → bruges
Geostatistics            → verde, geostatspy, scikit-gstat, gnnwr
Hydrology                → pastas, flopy
Surface Processes        → landlab
Structural Geology       → mplstereonet
Geochemistry             → pyrolite
Near-surface Geophysics  → gprpy, mtpy
Data Formats             → xarray (NetCDF/HDF5/Zarr), pyleoclim
Visualization            → pyvista, pygmt
Datasets                 → pooch, ensaio, rockhound (legacy)

📋 Full details: SKILLS.md

---

🔧 Usage

Describe the task naturally, or select an installed skill using your agent’s skill interface:

• “Use lasio to inspect the headers and curves in this LAS file.”
• “Use well-log-evaluation to evaluate these logs.”
• “Use using-geoscience-skills to choose the tools for this task.”

The router selects domain and workflow skills by name. Workflows can run within one agent session and do not require a particular delegation tool. Only the references and workflow stages needed for the task should be loaded.

Claude Code users working in this checkout can also use the optional aliases in .claude/commands/: /seismic-workflow, /well-analysis, /model-3d, /inversion-workflow, and /spatial-gridding. These aliases and the local SessionStart hook are not installed by the generic skills CLI.

---

📊 Coverage

---

Verification results

Updated 2026-09-15. PR #3 and the subsequent P1–P3 delivery in PR #5 are merged. The workflow follow-up extends their scientific evidence; native Agent runs retain their original recorded revisions.

Claude runtime testing is excluded from this round. Other client availability and optional-adapter decisions are in the platform assessment. Scientific execution is verified on Linux. GeoLime requires a licensed vendor runtime; RockHound is maintained here for legacy migration, with its failed upstream PREM URL documented. The new collection report distinguishes the executed workflow cases from remaining field-calibration, sampling-metadata, joint-inversion and forecasting limits.

---

🔄 Workflow Skills

Multi-step workflows that chain domain skills together:

The audited examples have separate scientific regression checks using generated LAS/SEG-Y files, elastic logs, geological/inversion models, and published field data. These check numerical outputs, coordinates and uncertainty assumptions. Recorded Codex task evaluations separately check skill selection and produced files.

🤖 Optional Role Guides

The files in agents/ provide optional role guidance. A generic skill installation does not register them as subagents; use them explicitly when available.

---

🤝 Contributing

PRs welcome! See CONTRIBUTING.md for the full guide, including:

• Step-by-step instructions for adding a new skill
• Portable frontmatter and project metadata conventions
• Task boundaries, resource loading, and scientific validation guidance
• Recursive skill validation, generated platform manifests, and installation smoke tests

See docs/ROADMAP.md for completed priorities and remaining validation boundaries.

Dependency maintenance describes the weekly registry report and the checks required before adopting new library or installer versions.

---

📚 Resources

• Source: awesome-open-geoscience
• Skills Spec: Agent Skills specification
• Community: Software Underground

---

📄 License

MIT © 2024

Bundled data retain their own licenses and attribution:

• Alps GNSS: original CC BY 3.0 and curated CC BY 4.0.
• Published well logs: CC BY 4.0.
• GEOFON waveform and response: CC BY 4.0.
• Crescentino ERT: CC BY 4.0.

Project-generated audit fixtures are explicitly synthetic and do not represent field observations.