| name | get-available-resources |
| description | Detect available CPU, GPU, memory, and disk resources before compute-heavy CS work. Use when planning model training, large-scale evaluation, embedding generation, experiment sweeps, dataset processing, log analysis, or reproduction runs where hardware limits affect the workflow. |
Get Available Resources
Use this skill early in a CS workflow when hardware constraints may change the implementation plan.
Typical Triggers
- choosing between in-memory and out-of-core data processing
- checking whether a model fits on available GPU memory
- sizing parallel workers for preprocessing or evaluation
- deciding whether to run locally, distribute with Dask, or move to Modal
- estimating whether a reproduction or benchmark sweep is realistic on the current machine
Workflow
- Run
scripts/detect_resources.py to collect CPU, GPU, memory, and disk information.
- Save or inspect the generated JSON report.
- Translate the report into execution choices.
- State the constraints explicitly before proposing a heavy workflow.
CS-Focused Decisions
Map resource findings to concrete tool choices:
- fits in RAM: prefer
polars, pandas, or standard PyTorch workflows
- exceeds RAM on one machine: consider
dask or vaex
- limited local GPU but scalable remote need: consider
modal
- multi-GPU or long training runs: plan around
pytorch-lightning or distributed tooling
- low disk headroom: avoid materializing large intermediate checkpoints or duplicated datasets
What to Report
By default, summarize:
- CPU cores and whether high parallelism is realistic
- available GPUs, VRAM, and likely model-size limits
- total RAM and whether dataset / checkpoint loading is safe
- free disk space and likely pressure points
- practical recommendations for the current task
Default Output
When no exact format is requested, return:
- a concise resource summary
- the main bottlenecks
- recommended execution strategy
- any risk to training, evaluation, or large-file processing