Skip to content
Skills Library
DevOps & cloud · Tool integration

lepton-cli

/lepton-cli

Operate NVIDIA DGX Cloud Lepton through the globally installed `lep` CLI. Use when the user asks to inspect or manage Lepton workspaces, endpoints/deployments, dev pods, batch jobs, fine-tuning jobs, Ray clusters, Slurm clusters, Dynamo endpoints, storage, secrets, nodes, ingress, templates, logs, authentication, or workspace status — or any task that needs safe command-line access to Lepton resources. Skip when the user is operating a different cloud or a different `lep`/`lepton` tool that isn't NVIDIA DGX Cloud Lepton.

Repository
leptonai ↗
Licence
Apache-2.0
Last repository change
27.10.2026

Assessment and security

100/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content35 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
2
Scripts
0
HTML comments
0
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 27.10.2026, without references to retired models or APIs.

Licence

Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.

What it does and when to use it

Operate NVIDIA DGX Cloud Lepton through the globally installed `lep` CLI. Use when the user asks to inspect or manage Lepton workspaces, endpoints/deployments, dev pods, batch jobs, fine-tuning jobs, Ray clusters, Slurm clusters, Dynamo endpoints, storage, secrets, nodes, ingress, templates, logs, authentication, or workspace status — or any task that needs safe command-line access to Lepton resources. Skip when the user is operating a different cloud or a different `lep`/`lepton` tool that isn't NVIDIA DGX Cloud Lepton.

Original author description

Operate NVIDIA DGX Cloud Lepton through the globally installed `lep` CLI. Use when the user asks to inspect or manage Lepton workspaces, endpoints/deployments, dev pods, batch jobs, fine-tuning jobs, Ray clusters, Slurm clusters, Dynamo endpoints, storage, secrets, nodes, ingress, templates, logs, authentication, or workspace status — or any task that needs safe command-line access to Lepton resources. Skip when the user is operating a different cloud or a different `lep`/`lepton` tool that isn't NVIDIA DGX Cloud Lepton.

How it works

  1. Runner
  2. Source Of Truth
  3. Auth And Context
  4. Operating Workflow
  5. Modifying Or Deleting Existing Workloads
  6. Output Handling
  7. Known Gotchas
Original SKILL.md headings · 92 lines, 1,337 words
SKILL.md preview
---
name: lepton-cli
description: "Operate NVIDIA DGX Cloud Lepton through the globally installed `lep` CLI. Use when the user asks to inspect or manage Lepton workspaces, endpoints/deployments, dev pods, batch jobs, fine-tuning jobs, Ray clusters, Slurm clusters, Dynamo endpoints, storage, secrets, nodes, ingress, templates, logs, authentication, or workspace status — or any task that needs safe command-line access to Lepton resources. Skip when the user is operating a different cloud or a different `lep`/`lepton` tool that isn't NVIDIA DGX Cloud Lepton."
---

# Lepton CLI

Use the globally installed `lep` command to operate NVIDIA DGX Cloud Lepton resources from the terminal.

## Runner

Run commands directly:

```bash
lep <command> [args...]
```

Before doing real work, verify the CLI is available:

```bash
lep --help
```

If `lep` is missing, tell the user the global Lepton CLI is not available in the current environment and ask them to install or expose it on `PATH`. Do not attempt to download or install it yourself.

## Source Of Truth

- Use `lep --help` and `lep <group> --help` before any unfamiliar command.
- Use `lep <group> <subcommand> --help` for exact flags and required arguments.
- Read [references/workloads.md](references/workloads.md) for the workload types Lepton exposes (endpoints, dev pods, batch jobs, Ray clusters, Slurm, fine-tuning) and when each one applies. Consult it before recommending a resource shape to the user.
- Treat the installed `lep` help output as authoritative for command behavior — command availability and flags vary by version.

## Auth And Context

Lepton authentication is scoped to a single workspace. A logged-in `lep` CLI may have a default workspace, but that default must not be assumed when the user has not named a workspace.

The CLI may read local configuration and these environment variables:

- `LEP_API_URL` — workspace URL used by `lep`.
- `LEP_TOKEN` — workspace auth token.
- `LEP_WORKSPACE` — default workspace name/display name.
- `LEP_ENV` — optional environment label.

Rules:

- Before running a series of operations, identify the target workspace. If the user did not specify one, ask whether they intend to use the CLI's default logged-in workspace.
- Start read-only with `lep workspace auth-status`, `lep workspace status`, `lep workspace list`, `lep workspace url`, or `lep workspace id`.
- If the CLI is not logged in for the target workspace, fall back to provided `LEP_*` environment variables when available.
- Prefer the exact `LEP_*` variables above when using environment credentials. Do not substitute `LEPTON_WORKSPACE_*` names unless the installed `lep` help or local configuration proves this CLI version reads them.
- Do not run `lep workspace token`, print config files, echo tokens, or include tokens in final answers unless the user explicitly asks.
- A secure-default `lep endpoint create` can print a newly generated API token.
  Treat that output as a secret: do not quote it in commentary or final answers,
  do not persist the transcript, and tell the user to save and rotate it through a
  secure channel.
- Prefer existing config/env credentials. Only run `lep workspace login -i <id> -t <token>` when the user explicitly provides credentials and accepts that the token may be persisted by the CLI.
- If a read-only `lep` command fails with `FailedToOpenSocket`, DNS, or other connection errors that look like a sandboxed network, surface the error to the user and ask whether to retry with broader network permission rather than silently retrying.

## Operating Workflow

1. Identify the target workspace and resource; confirm default-workspace intent when the user has not specified a workspace.
2. Verify current context with a read-only workspace command.
3. Discover exact syntax with `--help`.
4. For reads, run the narrowest command that answers the request.
5. For mutations, inspect current state first when possible.
6. For destructive operations, require explicit user intent or ask before proceeding.
7. Summarize the result with resource names, status, and any next action; redact secrets.

Destructive or high-impact commands include `remove`, `delete`, `stop`, `stop-all`, `remove-all`, `rm`, `rmdir`, `update`, `restart`, `create`, uploads/downloads to user-sensitive paths, and interactive connections (`ssh` and `shell` commands).

## Modifying Or Deleting Existing Workloads

Workloads are endpoints/deployments, dev pods, batch jobs, fine-tuning jobs, Ray clusters, Slurm clusters, and Dynamo endpoints (see [references/workloads.md](references/workloads.md)). Any command that modifies or deletes an existing workload requires explicit confirmation before it runs — even when auto mode is active or the user has previously approved similar actions in the session. Authorization does not carry over between resources or invocations.

Before running the mutation, read the workload's current state with a narrow read-only command, then show the user: the exact `lep` command to be run, the target workload name, the workspace, and the current state. State what will change in one sentence (e.g., "About to delete endpoint `foo` in workspace `bar` — this is irreversible") and ask the user to confirm. Only execute the mutating command after an explicit yes. A waiver applies only to the single named workload — do not generalize it to other workloads or to a later command.

## Output Handling

- Prefer table output for human inspection.
- Use JSON/YAML only when a command advertises `--output` or when the command naturally emits structured JSON.
- For logs or streaming commands, scope the request with available flags such as replica/name/time/tail when the command supports them.
- If a command writes files through options like `--path`, `download`, or `upload`, verify the source and destination before running.

## Known Gotchas

- `deployment` may be available as an alias for `endpoint`.
- Commands must be spelled in full; abbreviations such as `lep depl` or `lep pod sh` are rejected.
- Dynamo endpoints are a separate resource: use `lep dynamo ...`, not `lep endpoint ...`. `lep dynamo replica log` returns a snapshot of a replica's recent lines rather than a stream.
- Some command groups may be hidden from top-level help but still invokable. If the user asks for a known Lepton resource, try `lep <group> --help` before concluding it is unsupported.
- Slurm clusters are not self-serve — they are provisioned on request by the Lepton team. See [references/workloads.md](references/workloads.md).
- Slurm is exposed as nested commands with flag-based selectors (no positional resource arguments). Start with `lep slurm cluster list` and `lep slurm job list`; both lists accept repeatable `--status` filters, while cluster list also accepts repeatable `--name` substring filters. Use `lep slurm cluster get --name <name>` (or `--id <namespace/name>`), `lep slurm job get --id <job-id>` (or `--name <job-name>`), `lep slurm job attempts ...`, and `lep slurm job log ...` for detail. Add `--output json` where advertised when structured output is needed.
- Interactive shells: `lep pod shell`, `lep endpoint shell`, `lep job shell`, and `lep raycluster shell` open the dashboard's web-terminal shell through the workspace API (no SSH key or Teleport client) and apply the dashboard's availability rules. Pass `--replica` when an endpoint or job has several running replicas; a pod shell uses its newest replica and a Ray cluster shell defaults to the head node. Archived jobs, platform-managed tuning jobs, and Dynamo endpoints have no shell.
- Teleport SSH (local `tsh` v18+): `lep pod ssh --transport teleport`, `lep job ssh`, `lep node ssh`, `lep slurm cluster ssh`, `lep slurm job ssh`, and `lep slurm devpod ssh --transport teleport`. Pods on the new DevPod API and Jobs do not publish their Teleport proxy, so they use the active `tsh` profile or `--teleport-proxy`. The Slurm and node commands require a personal API token whose owner matches the Teleport sign-in.
- Slurm job submission and cancellation still happen through native Slurm tools (`sbatch`, `squeue`, `scancel`) after connecting to the cluster. The CLI's Slurm job API is intentionally read-only. `lep slurm cluster shell --name <name>` opens an interactive login-node shell through the workspace API, and `lep slurm cluster ssh --name <name>` connects to a login node through Teleport; treat both as high-impact under the confirmation rules above.
- A Slurm Dev Pod belongs to the current user and can be listed, inspected, created, removed, or connected to under `lep slurm devpod`. `create` takes `--cluster <name-or-namespace/name>`. For `get`, `remove`, `shell`, and `ssh`, `--name` and `--id` are mutually exclusive while `--cluster` is an optional scope that can also select the cluster's Dev Pod by itself. `get` displays configuration, identity, connection details, and a web link by default (`--output json` preserves the API record). `ssh` executes the platform-reported bastion command, or connects through Teleport with `--transport teleport`; `shell` uses the workspace API. Treat `create`, `remove`, and interactive connection commands as high-impact operations under the confirmation rules above.
- Help output is authoritative for the installed CLI version; command availability and flags can vary by version.
Files in this package (2)
  • SKILL.md 9 KB
  • references/workloads.md 6 KB

Versions

The database has no other skills named lepton-cli.

Database snapshot: 13.11.2026 · package version 6fe030432b7d (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 90dgx-station
    Inspect and guide NVIDIA DGX Station GB300 development using the local dgx-assist CLI and pinned NVIDIA playbooks. Use for general Station platform questions, Software 1.0 or 2.0 compatibility, GB300 or RTX GPU selection, UUID ordering, mixed ATS/HMM coherency, CDMM, general…
  • 100dynamo-troubleshoot
    Diagnose failed or unhealthy Dynamo deployments. Use when pods, model-cache jobs, PVCs, workers, frontend/router health, endpoints, or benchmark jobs fail; use recipe-runner/router-starter before this for normal bring-up.
  • 90dgx-station
    Inspect and guide NVIDIA DGX Station GB300 development using the local dgx-assist CLI and pinned NVIDIA playbooks. Use for general Station platform questions, Software 1.0 or 2.0 compatibility, GB300 or RTX GPU selection, UUID ordering, mixed ATS/HMM coherency, CDMM, general…
  • 100d1v
    Operate d1v.ai projects and container workspaces with the d1v CLI. Use when a task involves D1V authentication, project binding, workspace synchronization, container commands, environment metadata, preview deployments, deployment status, or a user-confirmed production release.
  • 100databricks-cli
    Expert guidance for using the Databricks CLI to manage Databricks workspaces, clusters, jobs, pipelines, Unity Catalog, SQL warehouses, serving endpoints, secrets, bundles, and all other Databricks resources. Use this skill when running databricks commands, managing Databricks…
  • 100spark-memory-thermal-ops
    Manage unified memory and thermals during long-running ML jobs on NVIDIA DGX Spark. Use when planning memory headroom for a training run on GB10, when a job OOMs on unified memory, or when monitoring temperature and power during multi-hour training.

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.