CLI Reference¶
Global usage¶
| Option | Description |
|---|---|
--config PATH |
Use a specific config file |
--ui plain|rich|auto |
Override UI mode |
--nointeractive |
Disable prompt fallback |
--version |
Show version |
Running slurmkit with no command opens the interactive command picker on interactive terminals.
Top-level commands¶
slurmkit init¶
Create .slurmkit/config.yaml.
slurmkit install-skill¶
Install the slurmkit Codex skill via npx skills.
Example:
slurmkit migrate¶
Rewrite old local state into the current layout.
Migrates:
.slurm-kit/->.slurmkit/.job-collections/->.slurmkit/collections/- old spec
output_dir/logs_dir->job_subdir
slurmkit generate¶
Generate job scripts from a spec and bind them to a collection.
Examples:
slurmkit generate experiments/exp1/slurmkit/job_spec.yaml --into exp1
slurmkit generate experiments/exp1/slurmkit/job_spec.yaml --into exp1 --dry-run
slurmkit spec-template¶
Write a starter job_spec.yaml scaffold.
Examples:
slurmkit spec-template
slurmkit spec-template --output experiments/exp1/slurmkit/job_spec.yaml
slurmkit spec-template --force
slurmkit submit¶
Submit jobs from a collection.
Examples:
slurmkit resubmit¶
Retry jobs from a collection, or target one logical job by tracked SLURM job ID.
slurmkit resubmit [COLLECTION] [--job-id JOB_ID] [--filter pending|running|completed|failed|unknown|preempted|timeout|cancelled|node_fail|out_of_memory|oom|all] [--dry-run] [-y]
Job-ID targeting rules:
- With
--job-id, collection is inferred automatically unless[COLLECTION]is provided to scope lookup. - If no collection matches the job ID, the command errors.
- If multiple collections match the job ID, the command errors and asks for
[COLLECTION]. - In
--job-idmode,--filteris allowed and enforced against the targeted job's latest effective state. - In
--job-idmode,--select-file/--select-functionare not supported.
Advanced regeneration controls are also available:
--template--extra-params--extra-params-file--extra-params-function--select-file--select-function--submission-group--regenerate / --no-regenerate
Examples:
slurmkit resubmit exp1 --filter failed --dry-run
slurmkit resubmit exp1 --filter failed --submission-group retry_after_fix
slurmkit resubmit --job-id 123456 --no-regenerate -y
slurmkit resubmit exp1 --job-id 123456 --no-regenerate -y
slurmkit status¶
Show the compact live status view for one collection.
collections commands¶
Available commands:
listshowanalyzerefreshcanceldelete
Examples:
slurmkit collections list
slurmkit collections show exp1
slurmkit collections analyze exp1 --param learning_rate --param batch_size
slurmkit collections refresh exp1
slurmkit collections refresh --all
slurmkit collections cancel exp1 --dry-run
slurmkit collections delete exp1 -y
Structured output is JSON-only where supported:
slurmkit collections show exp1 --json
slurmkit collections analyze exp1 --json
slurmkit status exp1 --json
slurmkit collections show --json includes effective_raw_state for each job,
which exposes row-level SLURM diagnostics used to derive canonical state.
notify commands¶
Available commands:
testjobcollection-final
Examples:
slurmkit notify test --dry-run
slurmkit notify test --route team_slack --dry-run
slurmkit notify job --job-id 123456 --exit-code 1 --dry-run
slurmkit notify job --collection exp1 --job-id 123456 --exit-code 1 --route team_slack --dry-run
slurmkit notify collection-final --collection exp1 --job-id 123456 --dry-run
slurmkit notify collection-final --collection exp1 --job-id 123456 --force --dry-run
sync¶
Write the per-host sync snapshot, optionally for one collection.
clean commands¶
slurmkit clean outputs [COLLECTION] [--threshold SECONDS] [--min-age DAYS] [--dry-run] [-y]
slurmkit clean wandb [--project NAME] [--entity NAME] [--threshold SECONDS] [--min-age DAYS] [--dry-run] [-y]
config¶
home¶
Open the interactive command picker explicitly: