How to use the Filestore NFS File Browser agent skill

What is an agent skill, and what does this one let the agent do?

An agent skill is a folder containing a SKILL.md file and supporting scripts and references. When it is installed, the agent can read the instructions and use the scripts to carry out a specific job. This one is for Google Cloud Filestore, a managed NFS file service.

Once installed, the agent can inspect, search and read files and POSIX metadata on Filestore instances without local NFS client packages (nfs-common), root privileges, or manual VPC network mounting on your own machine.

• Browse directory trees with file types, human-readable sizes and modification dates (tree).
• Search for filenames matching a glob pattern and grep text inside files with regex (search).
• Read text file chunks with head, tail or a line range (read).
• Inspect POSIX permissions, UID/GID, sizes and timestamps (stat).

Install the skill from the ZIP download

Download the ZIP for this skill from this page. Then choose one of the two install routes below.

For Claude Code, unzip the archive into ~/.claude/skills/ to make it available for your user, or into a project's .claude/skills/ folder to make it available only inside that project.

In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.

• 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

Set up the prerequisites before the first run

The skill calls gcloud and Python scripts that query your Google Cloud project, so the environment needs a few things in place. The skill's Quick Start asks you to enable the APIs it uses. Attribution settings are also part of the skill's commands and should not be removed.

Ensure gcloud is installed and authenticated, then enable the required services.

• gcloud CLI installed and authenticated
• Python 3 installed; the bridge needs the packages listed in requirements.txt (including fastapi)
• A Google Cloud project with Filestore, Compute and Cloud Run APIs enabled
• IAM roles depending on the engine: roles/run.invoker, roles/run.admin, or roles/iap.tunnelResourceAccessor
• Cloud Build access if you deploy the bridge container

Choose an engine: Cloud Run bridge or jump host VM

The skill reaches a Filestore share through one of two engines. The Cloud Run bridge is the primary option: a small read-only web service that the skill helps deploy, which mounts the share and answers REST calls with low latency. It scales to zero when idle. The jump host is the fallback: it uses an existing Compute Engine VM in your VPC, connected with IAP SSH, where the share is already mounted.

The jump host engine expects the share at /mnt/filestore by default, and you can pass a different mount path with the mount option.

• Cloud Run bridge: pass the bridge URL; requires roles/run.invoker.
• Jump host: pass the VM name with the jump host option; requires roles/iap.tunnelResourceAccessor.

Ask for what you need in plain language

You do not need to memorise the commands. Describe the outcome you want and include the project, instance and path if you know them. If you do not, ask the agent to list the Filestore instances first so it targets the right project.

The skill tells the agent to read in small chunks, so a request for a large log will return a slice rather than the whole file.

• "List the Filestore instances in project my-project so we can find the share."
• "Show the top-level folders on the Filestore share in instance nfs-prod, depth 1."
• "Search /app/logs for *.log files containing FATAL or Exception, max 25 results."
• "Read the last 50 lines of /backup/logs/app.log."
• "What are the permissions and size of /backup/database_dump.tar.gz?"

Run the commands yourself

You can also run the helper script directly. Each action takes the project and instance, plus the path and options for that action. The examples below use placeholders you replace with your own values.

• tree: --path, --depth and --max-entries (capped at 100 entries per response)
• search: --pattern for filenames and --grep for regex contents, with --max-results
• read: --head, --tail or --lines=START:END
• stat: shows POSIX metadata for one path

Safety notes: read-only, network and scripts

The skill is strictly read-only. It provides no write, edit, delete or truncate capabilities. It runs Python scripts and gcloud commands that query your project, and it uses the network to reach the share through the bridge or IAP. Binary files such as .tar.gz, .iso, .so and compiled binaries are protected from being dumped as text.

Because the skill runs scripts and touches cloud resources, do not use it in environments where agents must not run commands or access cloud resources. Read the bundled scripts before installing if your organisation requires a review. The skill also tells the agent to tag Filestore commands with an attribution metrics environment, which you should keep.

• Read-only: no write, edit, delete or truncate actions
• Runs scripts: yes; uses network: yes
• Excludes Cloud Storage buckets, Cloud NetApp Volumes and Persistent Disks

Tips for better results

Give the agent a specific path and a depth. Broad searches over a large share take longer and return more output. When a tree listing is truncated, the output shows a [TRUNCATED] notice, so narrow the path to a subfolder and try again.

If a read request has no range, the agent should default to a head or tail of 50 lines. Unbounded reads are capped at 100 lines automatically.

• Name the project, instance and path up front to avoid targeting the wrong environment.
• Narrow a truncated listing to a subpath rather than raising limits.
• Use search with --grep to find errors, then read the surrounding lines with --lines.