"Use SkyPilot to run, manage, and scale AI workloads across clouds,
Scanned 9/9/2026
Install to Claude Code
npx -y skills add VectorSpaceLab/AREX-Skill --skill skypilot --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Skypilot?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/vectorspacelab-skypilot)More formats (shields.io, HTML) on the badges page.
---
name: skypilot
description: "Use SkyPilot to run, manage, and scale AI workloads across clouds,
Kubernetes, Slurm, SSH node pools, managed jobs, SkyServe services, storage
mounts, and the Python SDK."
disable-model-invocation: true
metadata:
disco-role: operating
license: Apache 2.0
---
# SkyPilot
SkyPilot is a unified interface for launching, managing, and scaling AI workloads across cloud VMs, Kubernetes, Slurm, SSH node pools, and managed service deployments. Use this repo skill when a user asks for SkyPilot task YAMLs, `sky` CLI commands, Python SDK flows, managed jobs, SkyServe, storage mounts, cloud/resource selection, or SkyPilot source-code changes.
## Bootstrap
1. Confirm SkyPilot is installed and the CLI is reachable:
```bash
python scripts/check_skypilot_env.py --check-cli
```
2. For live SkyPilot use, check API server connectivity:
```bash
sky api info
```
3. For cloud/provider readiness, check credentials only when the user intends to launch or query resources:
```bash
sky check -o json
```
4. Read `references/troubleshooting.md` if install, import, API server, credentials, optional extras, or command-output issues appear.
## Route By Task
| User goal | Read |
| --- | --- |
| Write, review, convert, or validate SkyPilot task YAMLs, service YAML task fields, resources, mounts, envs, secrets, storage, or `sky.Task`/`sky.Resources` config. | `sub-skills/task-yaml/SKILL.md` |
| Launch/reuse interactive clusters, run `sky exec`, inspect status/logs/queues, manage autostop, stop/start/down, or plan safe cluster CLI commands. | `sub-skills/cluster-operations/SKILL.md` |
| Submit long-running training or batch workloads with `sky jobs`, managed spot recovery, job logs/queue/cancel, job groups, or job pools. | `sub-skills/managed-jobs/SKILL.md` |
| Deploy or debug SkyServe services, service YAMLs, readiness probes, replicas, autoscaling, load balancing, updates, logs, or LLM serving recipes. | `sub-skills/serving/SKILL.md` |
| Configure clouds, Kubernetes, Slurm, SSH node pools, GPU/resource selection, `sky check`, `sky gpus`, workspaces, object storage, file mounts, or volumes. | `sub-skills/infrastructure-storage/SKILL.md` |
| Use the Python SDK, async SDK, request IDs, `sky.get()`, API server lifecycle, remote server login, dashboard/API deployment, or compatibility checks. | `sub-skills/sdk-api-server/SKILL.md` |
| Modify the SkyPilot repository, choose focused tests, run formatting/linting, regenerate protobufs, rebuild dashboard assets, or prepare PR `Tested:` notes. | `sub-skills/repo-development/SKILL.md` |
## Core Decision Rules
- Prefer SkyPilot's optimizer: specify resources and constraints, then let SkyPilot choose cloud/region/zone unless the user explicitly pins infrastructure.
- Use `sky launch --dryrun` before expensive interactive launches; use parser-only validators for YAML checks when cloud access is unnecessary.
- Use `sky jobs launch` for unattended long-running work that should survive spot preemptions and capacity failures.
- Use SkyServe for service endpoints and replica management; test the underlying HTTP server with a simpler task or cluster workflow when startup is uncertain.
- Use structured output (`-o json`) for automation-friendly `status`, `queue`, `jobs queue`, `check`, and related inspection commands when supported.
- Keep secrets in `secrets`, `--secret`, or secret files; do not embed credentials in shared YAML, prompts, logs, or scripts.
- Treat smoke tests, live launches, serving deployments, pool creation, and provider checks as resource/credential-bound unless the user explicitly authorizes them.
## Bundled Helpers
- `scripts/check_skypilot_env.py` verifies package importability, version, optional CLI help, and common missing-install symptoms without launching resources.
- `scripts/inspect_cli_groups.py` prints safe help summaries for selected `sky` command groups without running launches, jobs, or services.
- Sub-skill scripts provide parser-only validators, command builders, SDK inspectors, resource checklists, and repo-maintenance sanity checks.
## Evidence And Staleness
- Read `references/repo-provenance.md` before deciding whether this skill matches a current SkyPilot checkout.
- Read `references/evidence-map.md` to understand the source areas distilled into this self-contained skill.
- If the current checkout commit, major public APIs, CLI flags, YAML schema, docs, examples, or dependencies changed, refresh this skill before relying on detailed guidance.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!