CP2K input generator#

Generate a practical CP2K draft, not a fake “final validated setup”. Prefer explicit assumptions, conservative defaults, and warnings over false certainty.

Follow this workflow#

1. Read references/design-rules.md.
2. Read references/job-spec-schema.md to understand the normalized contract.
3. Read only the references needed for the current task:
   • references/cp2k-task-map.md for task routing
   • references/cp2k-defaults.md for safe defaults
   • references/cp2k-kinds.md for basis/potential choices
   • references/ambiguity-policy.md for under-specified requests
   • references/structure-sources.md when the user needs a structure source rather than already having one
   • references/conversion-rules.md when the user wants Gaussian/VASP/ORCA/Quantum ESPRESSO drafts derived from the CP2K draft or normalized spec
4. Normalize the request before drafting the CP2K input.
5. Generate three deliverables whenever possible:
   • normalized.json
   • job.inp
   • report.md

Output discipline#

Always make these clear:

• interpreted task
• detected system type
• periodicity assumption
• major defaults applied
• warnings / review-required items

Do not silently invent scientifically decisive inputs such as nontrivial charge, spin state for uncertain open-shell systems, or detailed periodic settings for an xyz file pretending to be a crystal.

Use the bundled scripts when helpful#

These scripts are local helper utilities for deterministic text generation inside the skill folder. They are intended for transparent, offline preprocessing/rendering of CP2K draft files, not hidden network actions or privileged system changes.

• scripts/normalize-request.py: normalize a raw request JSON into the standard job spec
• scripts/render-cp2k-input.py: render a CP2K input draft from the normalized spec and a structure file
• scripts/generate_cp2k_bundle.py: one-shot pipeline that writes normalized.json, job.inp, and report.md
• scripts/convert-cp2k-input.py: convert a generated CP2K draft into a conservative Gaussian, VASP, ORCA, or Quantum ESPRESSO draft

Current supported draft space#

Support is strongest for:

• molecular xyz jobs
• single-point energy
• geometry optimization
• cell optimization for periodic systems
• short test MD inputs
• vibrational analysis drafts
• conservative periodic-material drafts with heuristic k-points
• draft conversion from the generated CP2K input/spec into Gaussian, VASP, ORCA, and Quantum ESPRESSO input sets

For harder cases like transition states, NEB, unusual excited-state methods, or poorly defined transition-metal spin states, generate a conservative draft only if the workflow is still interpretable and mark the risky fields for manual review.

Do not pretend that every structure format is parsed equally deeply. If the renderer cannot deterministically recover element-wise KIND data from the provided structure file, keep the draft conservative, surface the limitation in report.md, and require manual review instead of fabricating details.

If converting a CP2K input into another code, prefer CP2K drafts generated by this skill itself. Treat arbitrary hand-written CP2K files as only partially supported unless they stay close to the emitted block structure. file, keep the draft conservative, surface the limitation in report.md, and require manual review instead of fabricating details.

If converting a CP2K input into another code, prefer CP2K drafts generated by this skill itself. Treat arbitrary hand-written CP2K files as only partially supported unless they stay close to the emitted block structure.