Skip to content

SKILL.md

Latest commit

 

History

History
110 lines (78 loc) · 4.64 KB

SKILL.md

File metadata and controls

110 lines (78 loc) · 4.64 KB
name abinit-workflows
description Use when the task involves ABINIT electronic-structure workflows, including .abi input preparation, gs, relax, nscf, DOS, and band setups, pseudopotential and dataset organization, convergence review, restart handling with DEN or WFK files, output diagnosis, and scheduler scripts.

ABINIT Workflows

This skill handles practical ABINIT workflows for setup, review, and recovery. Use it when the request is specifically about ABINIT, .abi inputs, multi-dataset runs, pseudopotential handling, convergence choices, or reading .abo-style outputs.

When to use

Use this skill when the user mentions or implies:

  • ABINIT, .abi, .abo, abinit, abiinput, or multi-dataset inputs
  • ground-state, geometry optimization, DOS, band structure, or restart workflows in ABINIT
  • ecut, ngkpt, shiftk, nshiftk, tolvrs, toldfe, ndtset, getwfk, irdwfk, or irdden
  • pseudos, scheduler scripts, walltime planning, or restart files such as DEN, WFK, and HIST

Operating stance

Prioritize missing information in this order:

  1. target task: gs, relax, dos, band, or restart
  2. pseudopotential family and exchange-correlation consistency
  3. basis and k-point controls: ecut, ngkpt, shiftk, and stopping criteria
  4. parent-child file flow: which stage produces DEN, WFK, or geometry for the next stage
  5. scheduler and MPI layout

Never silently invent:

  • pseudopotential filenames or versions
  • whether the calculation should be metallic, spin-polarized, or +U
  • a high-symmetry path without a real source
  • whether an old WFK or DEN file is scientifically safe to reuse

Workflow

1. Classify the request

  • Setup: generate one stage or a staged directory layout.
  • Review: inspect an existing ABINIT directory and summarize status.
  • Recovery: explain the likely failure mode and recommend the smallest safe restart.

2. Gather the minimum viable context

Before making code-specific recommendations, establish:

  • structure source and whether the cell is bulk, slab, or molecule-in-box
  • pseudopotential family and intended functional
  • whether the system is metallic, magnetic, or unknown
  • target observable: relaxed geometry, total energy, DOS, or band plot
  • scheduler environment and expected walltime

3. Use the bundled helpers

  • scripts/make_abinit_inputs.py Generate conservative gs, relax, nscf, dos, or band workflow skeletons.
  • scripts/check_abinit_job.py Check an ABINIT directory for missing inputs, restart dependencies, and obvious pseudo-path issues.
  • scripts/summarize_abinit_run.py Summarize an ABINIT directory or .abo output using simple, auditable heuristics.
  • scripts/recommend_abinit_recovery.py Turn incomplete or broken ABINIT stages into concrete restart and parent-child recovery guidance.
  • scripts/export_status_report.py Export a shareable markdown status report from an ABINIT run or staged workflow.
  • scripts/export_input_suggestions.py Export conservative ABINIT input snippets based on detected recovery patterns.

4. Load focused references only when needed

  • ABINIT-specific workflow guidance: references/abinit.md
  • convergence planning: references/convergence.md
  • failures and restarts: references/failure-modes.md
  • scheduler considerations: references/schedulers.md

5. Deliver an auditable answer

Whenever you recommend settings or restart changes, include:

  • the task and parent-child stage assumption
  • unresolved physics choices the user still needs to confirm
  • exact files that were created or edited
  • what output or artifact must exist before the next stage can run

Task guardrails

Ground-state and relaxation

  • Keep exploratory relaxations numerically sane but not over-tight.
  • Separate geometry convergence from production-quality total energies.
  • Do not compare final energies from a loose relaxation against a tighter static setup without rerunning consistently.

DOS

  • Treat DOS as a child workflow that depends on a trustworthy parent state.
  • Use denser k-point sampling for DOS than for a cheap exploratory relaxation.
  • Be explicit about whether the stage reads DEN, WFK, or both.

Band structure

  • Require a converged parent stage.
  • Do not fabricate a band path; use a symmetry tool or literature reference.
  • Keep the band stage read-only with respect to density whenever possible.

Quality bar

  • Prefer conservative defaults over flashy guesses.
  • Distinguish ABINIT-specific syntax advice from material-specific physics advice.
  • Point out when a recommendation is only a heuristic.
  • If the output is incomplete or the pseudo policy is unclear, say so directly.