Background
Following #7488 (AGENTS.md), I propose adding structured "skills" for AI coding agents, so that agents can help users run calculations, analyze results, and troubleshoot issues.
Design Philosophy
Skills should answer: How can an AI agent help users complete a specific computational task?
Each skill includes:
- Input file format and required parameters
- How to run the calculation
- Output file locations and formats
- How to interpret results
- Common troubleshooting
Proposed Skills
1. abacus-scf — Self-Consistent Field Calculation
Trigger: User wants to run SCF, compute energy, or study electronic structure
Content:
- INPUT file: required parameters (ecutwfc, ks_solver, basis_type)
- STRU file: structure format
- KPT file: k-point settings
- Run command:
mpirun -n N abacus
- Output:
OUT.ABACUS/ directory structure
- Results: energy, forces, stress location
2. abacus-relax — Structure Optimization
Trigger: User wants to relax a structure
Content:
- INPUT parameters: relax_method, relax_nmax, force_thr
- Workflow: SCF → force calculation → move atoms → repeat
- Output:
STRU_ION_D trajectory, final structure
- Convergence: force and energy thresholds
3. abacus-md — Molecular Dynamics
Trigger: User wants to run MD simulation
Content:
- INPUT parameters: md_type, md_nstep, md_dt
- Ensemble: NVT/NPT/NVE
- Thermostat: nhc/csvr/berendsen
- Output: trajectory, temperature, pressure
4. abacus-band — Band Structure Calculation
Trigger: User wants to compute band structure
Content:
- Two-step workflow: SCF → NSCF
- KPT file: high-symmetry path
- Output:
BANDS.dat, BANDSTRUCTURE format
- Plotting: using
abacus-plot
5. abacus-debug — Troubleshooting
Trigger: User encounters calculation issues
Content:
- SCF not converging: check mixing_beta, ecutwfc, k-points
- Out of memory: check ecutrho, parallel settings
- Slow: check ks_solver, OpenMP settings
- Wrong results: check functional, basis set, cutoff
Example Structure
skills/
├── abacus-scf/
│ └── SKILL.md
├── abacus-relax/
│ └── SKILL.md
├── abacus-md/
│ ├── SKILL.md
│ └── references/
│ ├── nhc-thermostat.md
│ └── csvr-thermostat.md
├── abacus-band/
│ └── SKILL.md
└── abacus-debug/
├── SKILL.md
└── references/
├── scf-convergence.md
└── memory-issues.md
Benefits
- Users can let agents run calculations directly
- Agents can understand output results
- Agents can help troubleshoot issues
- Lower the barrier to using ABACUS
References
If this proposal sounds good, I can submit a PR to implement it. Please assign this issue to me.
I will write based on the existing documentation.
Background
Following #7488 (AGENTS.md), I propose adding structured "skills" for AI coding agents, so that agents can help users run calculations, analyze results, and troubleshoot issues.
Design Philosophy
Skills should answer: How can an AI agent help users complete a specific computational task?
Each skill includes:
Proposed Skills
1.
abacus-scf— Self-Consistent Field CalculationTrigger: User wants to run SCF, compute energy, or study electronic structure
Content:
mpirun -n N abacusOUT.ABACUS/directory structure2.
abacus-relax— Structure OptimizationTrigger: User wants to relax a structure
Content:
STRU_ION_Dtrajectory, final structure3.
abacus-md— Molecular DynamicsTrigger: User wants to run MD simulation
Content:
4.
abacus-band— Band Structure CalculationTrigger: User wants to compute band structure
Content:
BANDS.dat,BANDSTRUCTUREformatabacus-plot5.
abacus-debug— TroubleshootingTrigger: User encounters calculation issues
Content:
Example Structure
skills/
├── abacus-scf/
│ └── SKILL.md
├── abacus-relax/
│ └── SKILL.md
├── abacus-md/
│ ├── SKILL.md
│ └── references/
│ ├── nhc-thermostat.md
│ └── csvr-thermostat.md
├── abacus-band/
│ └── SKILL.md
└── abacus-debug/
├── SKILL.md
└── references/
├── scf-convergence.md
└── memory-issues.md
Benefits
References
If this proposal sounds good, I can submit a PR to implement it. Please assign this issue to me.
I will write based on the existing documentation.