Claude Skill: Build a DeepDriveWE Example

This project ships a Claude Code skill that walks Claude through creating a new DeepDriveWE example end-to-end. Install it once, then ask Claude something like "build a new DeepDriveWE example for ligand binding with AMBER" -- Claude automatically loads the skill and follows the project's conventions (directory layout, agent subclassing, binner/resampler choices, Parsl compute configs, verification steps).

What the Skill Covers

• Required project layout (main.py, workflow.py, config.yaml, common_files/, inputs/bstates/).
• Why Parsl-serializable classes must live in workflow.py, not main.py.
• How to subclass SimulationAgent and WestpaAgent.
• Selection guide for binners, recyclers, resamplers, and compute configs.
• Checkpoint/resume pattern and SIGTERM handling.
• Verification steps before scaling up.

Where to Install It

Claude Code loads skills from two locations:

User-level (recommended)Project-level

~/.claude/skills/deepdrivewe-example/SKILL.md

Available in every project you work on. Install this way if you plan to build DeepDriveWE examples across multiple checkouts or downstream projects.

.claude/skills/deepdrivewe-example/SKILL.md

Committed to this repo and automatically available whenever you open it in Claude Code. No install step needed -- it's already here.

Install (User-Level)

curlFrom a local checkoutwget

mkdir -p ~/.claude/skills/deepdrivewe-example
curl -fsSL \
  https://raw.githubusercontent.com/ramanathanlab/deepdrivewe-academy/main/.claude/skills/deepdrivewe-example/SKILL.md \
  -o ~/.claude/skills/deepdrivewe-example/SKILL.md

mkdir -p ~/.claude/skills/deepdrivewe-example
cp .claude/skills/deepdrivewe-example/SKILL.md \
   ~/.claude/skills/deepdrivewe-example/SKILL.md

mkdir -p ~/.claude/skills/deepdrivewe-example
wget -O ~/.claude/skills/deepdrivewe-example/SKILL.md \
  https://raw.githubusercontent.com/ramanathanlab/deepdrivewe-academy/main/.claude/skills/deepdrivewe-example/SKILL.md

Once installed, Claude Code will see deepdrivewe-example in its skill list on the next session start.

How to Trigger It

Just ask in natural language. Claude matches the user request against the skill's description and loads the skill automatically.

Example prompts:

"Create a new DeepDriveWE example for NTL9 folding using AMBER instead of OpenMM."

"Build a new weighted ensemble workflow to study ligand unbinding from a kinase. Use a distance-based progress coordinate."

"Set up a minimal DDWE example that runs on a single laptop CPU with a synthetic simulation."

Contents Preview

The skill is ~200 lines of structured guidance. A condensed outline:

1. Before You Start -- requirements gathering (system, MD engine, progress coordinate, target direction, compute, starting point).
2. Project Layout -- required files and the critical rule about where serializable classes must live.
3. Step-by-Step (6 steps) -- config models, agent subclassing, main.py pattern, config.yaml pattern, verification.
4. Component Selection Guide -- tables for picking the right binner, recycler, resampler, and compute config.
5. Common Extension Points -- custom pcoords, data products, resamplers, ML-driven workflows.
6. Gotchas -- project conventions and easy-to-miss pitfalls.

View the full source at .claude/skills/deepdrivewe-example/SKILL.md.

Updating the Skill

If you improve the skill locally, please open a PR against develop so everyone benefits. See the Contributing guide.