dft-cp2k

name	dft-cp2k
description	Route CP2K requests to task-specific subskills based on user intent. Use when the user asks for CP2K workflows and you must decide between static, relaxation, molecular dynamics, or electronic-analysis preparation. This orchestration skill dispatches to the correct CP2K subskill and enforces consistent handoff to submission skills.
compatibility	Requires a runnable CP2K environment and suitable basis/pseudopotential data for target elements.
license	GPL-3.0-only
metadata	{"author":"qqgu","version":"0.1.0","repository":"https://github.com/cp2k/cp2k"}

CP2K Task Router

Use this skill as the top-level CP2K orchestration layer.

Purpose

This skill routes requests to one task-specific CP2K subskill path:

• dft-cp2k/static
• dft-cp2k/relax
• dft-cp2k/md
• dft-cp2k/electronic

Scope

This router skill should:

• require user-provided structure or prerequisite context
• classify request intent into one CP2K task type
• collect minimal shared context before dispatch
• delegate detailed parameter handling to selected subskill
• enforce consistent output/handoff policy across subskills

This router skill should not:

• own detailed templates for all CP2K tasks
• execute or submit calculations
• bypass subskill-specific guardrails

Hard requirement

The user must provide enough starting context:

• structure input for static / relax / md
• prerequisite converged context for electronic

If prerequisites are missing, stop and ask for them.

Routing rules

1. If user requests single-point SCF/energy: route to dft-cp2k/static.
2. If user requests geometry optimization: route to dft-cp2k/relax.
3. If user requests molecular dynamics: route to dft-cp2k/md.
4. If user requests electronic analysis workflow: route to dft-cp2k/electronic.
5. If intent is ambiguous, ask one focused clarification question before routing.

Shared policy for all subskills

• do not invent missing basis/potential files
• expose assumptions and unresolved scientific choices explicitly
• return handoff-ready task layout
• if execution is requested, hand off to dpdisp-submit

GPUMD / NEP integration

CP2K is a common DFT backend for generating NEP training labels, especially for large systems where linear-scaling methods are advantageous.

• NEP labeling: use dft-cp2k/static for single-point energy+force+stress. Recommended: EPS_SCF 1e-7 or tighter.
• Conversion: convert CP2K output to NEP extxyz via GPUMD/tools/Format_Conversion/cp2k2xyz or dpdata-cli:

  uvx dpdata cp2k_output -i cp2k/output -O train.xyz -o extxyz
  

• Stress convention: CP2K reports stress in GPa by default. Verify virial sign and units after conversion.

Output from router

Provide:

1. selected subskill path
2. why it was selected
3. minimal missing inputs (if any)
4. explicit next step (invoke selected subskill)