agentleFS
brevdevbrevdev/workshop-build-an-agent/.claude/skills/nvwb-project/SKILL.md

nvwb-project

Skill156 starsChanged 8 days agoApache-2.0View on GitHub

Apply when working in a directory containing .project/spec.yaml — provides NVIDIA AI Workbench project awareness for in-container development. This skill provides context about project structure, environment configuration, and rebuild/restart requirements.

  • Reads credentials
  • Installs packages

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 2 are distinctive, and 7 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If .project/spec.yaml exists in the project root, this is a Workbench project.

    line 13
  2. Distinctive:

    After editing a build-time or runtime file, always tell the user what action is needed.

    line 46
  3. Very specific:

    If compose.yaml or docker-compose.yaml exists, multi-container services may be running alongside the project container:

    line 54
  4. Very specific:

    If /project/agent-bridge/ exists, this project is set up for communication between inner Claude (here, in the container) and outer Claude …

    line 62
  5. Very specific:

    Read common-context/context.md for current task state

    line 65
  6. Very specific:

    Write a clear request to /project/agent-bridge/inner.md

    line 66
  7. Distinctive:

    Wait for the user to relay it to outer Claude on their local machine

    line 67
  8. Very specific:

    Read the response from /project/agent-bridge/outer.md

    line 68
  9. Very specific:

    Rules: Do not copy files into or out of agent-bridge/ — it is a communication channel, not a file staging area.

    line 70

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

NVIDIA AI Workbench — In-Container Project Awareness

This skill activates when Claude is working inside an AI Workbench project container (detected by .project/spec.yaml existing in or above the current working directory).

Detection

If .project/spec.yaml exists in the project root, this is a Workbench project. Read it first to understand the project's layout, apps, mounts, and resources.

Key Facts

  • The project root is bind-mounted at /project/ inside the container
  • The nvwb CLI is NOT available inside the container — all nvwb commands must run on the user's local machine (the control plane)
  • This container may be running locally or on a remote — it does not matter for in-container work; the project files are always at /project/
  • Code in layout directories is live-mounted — edits take effect immediately
  • The environment variable AI_WORKBENCH_FLAG=true may be set to detect Workbench context
  • Compose services are reachable by service name (e.g. curl http://api:8080/) via the shared workbench network

What's Safe to Edit

Freely editable — no action needed:

  • Any files in layout directories (code/, data/, models/ or custom paths from spec.yaml)
  • Application code, scripts, notebooks

Requires rebuild — inform user to run nvwb build on the host:

  • apt.txt — system packages
  • requirements.txt — pip packages
  • preBuild.bash — pre-install build script
  • postBuild.bash — post-install build script

Requires restart — inform user to run nvwb close then nvwb open on the host:

  • variables.env — runtime environment variables
  • Mount configuration changes in .project/spec.yaml
  • App definitions in .project/spec.yaml

Edit carefully — validate format:

  • .project/spec.yaml — do not break YAML structure; suggest nvwb validate project-spec after changes

When Informing the User

After editing a build-time or runtime file, always tell the user what action is needed. Examples:

I've updated requirements.txt to add the transformers package. You'll need to run nvwb build on the host to rebuild the container with this change.

I've added CUDA_VISIBLE_DEVICES=0 to variables.env. This will take effect after a container restart — run nvwb close then nvwb open on the host.

Compose Awareness

If compose.yaml or docker-compose.yaml exists, multi-container services may be running alongside the project container:

  • Services communicate over a shared Docker network — use service names as hostnames (e.g. http://api:8080/)
  • /nvwb-shared-volume/ is shared between containers
  • Changes to compose files require nvwb compose down + nvwb compose up on the host — request this via the bridge if running as inner Claude
  • Compose services exposed through the Workbench proxy use NVWB_TRIM_PREFIX=true as an environment variable in the service definition (not in spec.yaml)

Host↔Container Bridge

If /project/agent-bridge/ exists, this project is set up for communication between inner Claude (here, in the container) and outer Claude (on the user's local machine, where nvwb is installed).

When you need a local-machine action (compose restart, rebuild, nvwb command):

  1. Read common-context/context.md for current task state
  2. Write a clear request to /project/agent-bridge/inner.md
  3. Wait for the user to relay it to outer Claude on their local machine
  4. Read the response from /project/agent-bridge/outer.md

Rules: Do not copy files into or out of agent-bridge/ — it is a communication channel, not a file staging area. The content of inner.md and outer.md is unrestricted — include logs, snippets, and diagnostics as needed.

Sandbox: If hooks are configured, your actions are enforced — nvwb commands are blocked (use the bridge), sensitive environment variables are scrubbed from your shell commands, writes inside agent-bridge/ are restricted to your designated files, and reads of sensitive files and hook configuration are blocked. Do not attempt to modify hook configuration or settings files.

Note: if this project is running in a remote context, the bridge files live on the remote machine. The context name (from nvwb list contexts) is also the SSH alias — Workbench adds a matching entry to ~/.ssh/config. Outer Claude can SSH directly using the context name to read and write bridge files without involving the user as a relay.

Requests can be rich prose — include what you changed, what you need done, and what you want reported back (logs, curl output, etc.).

See references/agent-bridge.md for templates and the full format guide.

Runtime Constraints

sudo is NOT available in the running container — the container runs as the workbench user without elevated privileges at runtime. If a task requires system-level changes, it must go in preBuild.bash or postBuild.bash.

Build Script Constraints

Build scripts (preBuild.bash, postBuild.bash) run during container image build:

  • They CANNOT reference /project/ — the project directory is not mounted during build
  • Available variables: $NVWB_UID, $NVWB_GID for file ownership
  • Both scripts run as the workbench user (not root) — any write to a system directory requires sudo. This includes mkdir, npm install -g, writing to /usr/local/, etc. Shell redirections (>, >>) also run as workbench, so use sudo tee instead of sudo command > /path.

References

See references/config-files.md for detailed information about each configuration file.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.