Guide · Agent skill
How to use the Cloud Storage FUSE Mounts agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
This guide explains what the google-cloud-storage-fuse agent skill does, how to install it into Claude, and how to ask for help with it in plain language. It is aimed at people who have not used agent skills before. By the end you will know whether your workload should use a gcsfuse mount at all, how to get the agent to help you set one up, and how to ask it to diagnose a slow or expensive mount.
Cloud Storage FUSE Mountsgoogle/skills · Apache-2.0 Download the skill Quick start
- 1Download the ZIP. Download the skill as a ZIP file from this page. The archive contains SKILL.md and a references folder.
- 2Unzip into a skills folder. Unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder to limit it to that project.
- 3Or upload it in the Claude apps. If you use the Claude apps, open Settings, go to Capabilities, then Skills, and upload the ZIP.
- 4Prepare your Google Cloud setup. Have a Cloud Storage bucket, the gcloud and gcsfuse CLIs, and a project where you hold appropriate IAM permissions. For GKE deployments, the skill also expects Workload Identity.
- 5Describe your goal in plain language. Tell the agent what your code does and where the data lives, for example that your training code uses hardcoded file paths and reads a dataset from a bucket.
- 6Answer the fit questions. The agent may ask whether your code can read gs:// paths, and how it reads and writes files. Your answers decide whether a mount is the right choice.
What the skill makes the agent do
An agent skill is a folder of instructions and reference files that an AI agent loads when your request matches the skill's purpose. This one covers Cloud Storage FUSE, which mounts a Google Cloud Storage bucket so that programs can read and write it like a normal folder, using the gcsfuse tool.
The skill does not start with a mount. It first checks whether a mount is the right choice, since native gs:// paths or Filestore are often better. If a mount fits, the agent helps configure, deploy and tune it, and warns about write patterns that are unsafe on object storage. It can also help diagnose slow training or unexpected Cloud Storage bills using gcsfuse metrics.
- Decides whether FUSE, native gs:// reads, or Filestore/Managed Lustre fits a workload
- Helps deploy tuned mounts on GKE, Compute Engine, or Cloud Run
- Covers checkpoint and write safety, such as rename behaviour and close-time finalization
- Walks through performance and cost diagnosis using gcsfuse metrics
Install the skill in Claude
Download the skill as a ZIP file from the skill's page on this directory. Then choose one of two ways to install it.
For Claude Code, unzip the file into ~/.claude/skills/ to make it available in every project, or into a project's .claude/skills/ folder to limit it to that project. The skill's folder should contain SKILL.md and the references folder.
In the Claude apps, you can upload the ZIP instead. Open Settings, go to Capabilities, then Skills, and add the skill there.
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into the project's .claude/skills/ folder
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Trigger the skill with natural requests
You do not need to name the skill. Describe what you want in ordinary words that mention Cloud Storage, a bucket, or a mount. The skill's description covers requests about deploying mounts, cache and mount flags, ML checkpointing, and slow training or bill spikes.
The skill sorts requests into three groups: provisioning a mount, checking whether a write pattern is safe, and diagnosing a regression. If your request is vague, the skill asks one question first, usually whether your code can read gs:// paths directly.
- Provision: 'Mount my bucket so my training pods can read the dataset'
- Safety: 'Is it safe to write checkpoints by renaming a directory on this mount?'
- Safety: 'Can two writers share this gcsfuse mount?'
- Diagnosis: 'Training through the mount got slow after last week, where do I start?'
- Diagnosis: 'The Cloud Storage bill spiked after we mounted the bucket'
What the agent does with your request
For a provisioning request, the agent works from the GKE training deployment reference. It chooses between a profile StorageClass and a static PersistentVolume depending on the GKE version, and it covers Workload Identity bindings, file cache sizing on Local SSD, and sidecar annotations. It also covers Compute Engine with the aiml-training profile and Cloud Run volume mounts.
For a diagnosis request, the agent starts with telemetry. If gcsfuse metrics are not enabled on the mount, enabling them is the first step. The skill explains that the default mount is tuned for coherency rather than performance, so the default settings are often the cause of slow re-reads.
Run the gcloud commands with attribution
The skill asks that every gcloud command it runs or gives you carries a metrics environment variable that identifies the skill and its version. The variable is set inline on each command, not stored with gcloud config set, so it does not carry over into unrelated work.
You can copy this prefix from the skill's instructions. Replace the placeholder with the gcloud subcommand you need.
CLOUDSDK_METRICS_ENVIRONMENT="gcs-skills gcs-skills/1.0 (skill:google-cloud-storage-fuse)" \
gcloud <command> [flags]Safety notes before you use it
This skill runs scripts and uses the network. Read what the agent proposes to run before you approve it, especially commands that create resources or change IAM bindings in your Google Cloud project.
Mounting is not always the right answer. The skill says not to use FUSE for databases or applications that need file locking (flock or fcntl) on shared storage, and to use Filestore or Managed Lustre instead. A mount also adds cost and complexity, so it is not worth it for workloads that can read gs:// paths natively.
- Runs scripts: yes
- Uses network: yes
- Requires a Google Cloud project with appropriate IAM permissions
- Not for bucket administration without a mount; use the google-cloud-storage-basics skill for that
What people use it for
- Giving a GKE training job a folder view of a dataset stored in a Cloud Storage bucket
- Mounting a bucket on a Compute Engine VM with the aiml-training profile for a training run
- Checking whether a checkpoint publish step that renames a directory is safe on your bucket type
- Diagnosing why re-reads of training data are slow on an existing gcsfuse mount
- Working out why the Cloud Storage bill rose after a mount was added
Frequently asked questions
Is the google-cloud-storage-fuse skill free to use?
The skill is published under the Apache-2.0 licence. Google Cloud resources you create, such as buckets, requests and GKE clusters, are billed by Google Cloud under your own account. This page does not list prices.
What do I need installed to use this skill?
The skill's requirements are the gcloud CLI, the gcsfuse CLI, a Cloud Storage bucket, and a Google Cloud project where you have appropriate IAM permissions. For GKE deployments it also expects Workload Identity, and Local SSD is optional for file cache sizing.
Should I mount a bucket or read gs:// paths directly?
If your code or library accepts gs:// URIs natively, such as pandas with gcsfs or TensorFlow with tf.io.gfile, the skill says to read the paths directly and skip the mount. A mount is for code that requires POSIX file paths and cannot read gs:// URIs.
What are the alternatives to Cloud Storage FUSE?
The skill names Filestore (NFS with POSIX locking) and Managed Lustre for shared, mutable workloads that need file locking. For workloads that can read gs:// paths natively, the skill says no mount is needed.
Why is my gcsfuse mount slow or expensive?
The skill explains that the default mount is tuned for coherency, not performance, so the file cache is off and the metadata TTL is 60 seconds. Enable gcsfuse metrics first, then ask the agent to diagnose cache-hit and request-mix patterns, and check Class A and Class B billing lines.
Can I use the skill on Claude Code and in the Claude apps?
Yes. For Claude Code, unzip the ZIP into ~/.claude/skills/ or a project's .claude/skills/ folder. In the Claude apps, upload the ZIP under Settings → Capabilities → Skills.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you are an ML engineer or developer whose existing code needs POSIX file paths and must read or write a Cloud Storage bucket, or if you need to diagnose a slow or costly gcsfuse mount. Pick something else if your code can read gs:// paths natively, if you need file locking on shared storage, or if you only need bucket administration, since a mount then adds cost and complexity you do not need.
Download the skill