Guide · Agent skill
How to use the TensorBoard visualization toolkit agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
This guide explains what the TensorBoard agent skill does, how to install it for Claude, and how to ask the agent for TensorBoard logging in your own training code. It is aimed at people who may never have used an agent skill before. By the end you will know how to add metric, image, histogram and graph logging to a PyTorch or Keras script and open the dashboard in your browser.
TensorBoard Visualization ToolkitOrchestra-Research/AI-Research-SKILLs · MIT Download the skill Quick start
- 1Download and unzip the skill. Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills in the Claude apps.
- 2Install the packages. Run pip install tensorboard, plus torch and torchvision for PyTorch or tensorflow for Keras.
- 3Make a training script. Have a working training script ready. The skill adds logging to existing code and does not write the model for you.
- 4Ask for logging. Describe the metrics and framework you want, for example loss and accuracy curves in PyTorch.
- 5Run training. Run your script so it writes event files into a runs or logs folder.
- 6Launch TensorBoard. Run tensorboard --logdir=runs and open http://localhost:6006 in your browser.
- 7Compare experiments. Use a separate log directory per run, then view them together on the dashboard.
What the TensorBoard skill makes the agent do
TensorBoard is Google's visualization toolkit for machine learning. It shows training metrics, histograms, model graphs, embeddings and profiling data in a local web dashboard. The skill gives the agent working knowledge of how to write the logging calls that feed that dashboard.
The skill assumes you already have a training script. The agent adds logging to it rather than building a model for you.
- Log loss and accuracy over time (scalars)
- Log images, weight and gradient histograms, and the model graph
- Project embeddings to 2D/3D with PCA, t-SNE or UMAP
- Compare hyperparameter runs in the HParams tab
- Add PR curves and text or markdown logs
Install the skill
Download the ZIP for this skill from this page. Unzip it into one of these locations: ~/.claude/skills/ for all your projects, or a .claude/skills/ folder inside a specific project.
You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
- Personal: unzip into ~/.claude/skills/
- Per project: unzip into .claude/skills/ in the project folder
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Install the Python packages it needs
The skill needs the tensorboard package. The PyTorch examples also need torch and torchvision; the Keras examples need TensorFlow, which includes TensorBoard.
pip install tensorboard
# PyTorch integration
pip install torch torchvision tensorboard
# TensorFlow integration (TensorBoard included)
pip install tensorflowAsk for TensorBoard logging in natural language
You do not need to name the skill or remember function names. Describe what you want to see and point the agent at your training script. Good requests name the framework, the script and the metrics.
Examples of requests that should trigger the skill: "Add TensorBoard logging to my PyTorch training loop in train.py so I can see train and validation loss." "Log weight histograms every epoch in my Keras model." "Set up TensorBoard so I can compare learning rates 0.001, 0.01 and 0.1."
Log scalars, images and histograms
Scalars are the most common starting point. In PyTorch, create a SummaryWriter and call add_scalar once per step or epoch. Group related values with add_scalars so train and validation loss appear on one chart.
Histograms track weight and gradient distributions. In Keras, the TensorBoard callback with histogram_freq=1 writes them each epoch.
- Use names like Loss/train and Loss/val so the dashboard groups them
- Log learning rate from optimizer.param_groups to check your schedule
- Use make_grid from torchvision to log a batch of images as one grid
writer = SummaryWriter('runs/experiment_1')
for epoch in range(10):
writer.add_scalar('Loss/train', train_loss, epoch)
writer.add_scalar('Accuracy/val', val_acc, epoch)
writer.close()Launch the dashboard and compare runs
Run tensorboard with --logdir pointing at the folder that holds your run directories, then open http://localhost:6006 in a browser. Each subfolder under the log directory appears as a separate run, so give each experiment its own writer directory to compare them.
For hyperparameter sweeps, use add_hparams and then open the HParams tab.
- Launch: tensorboard --logdir=runs
- Open: http://localhost:6006
- Use separate directories such as runs/lr0.001_bs16 for each run
tensorboard --logdir=runsProfile performance and find bottlenecks
The profiling reference covers the PyTorch Profiler, which can write traces that TensorBoard displays. It also covers GPU utilisation, memory profiling and bottleneck detection. Ask the agent to add profiling around a few training steps rather than the whole run, since full traces get large.
What people use it for
- Watching train and validation loss curves to spot overfitting
- Comparing learning-rate or batch-size settings across several runs in the HParams tab
- Checking weight and gradient histograms to find vanishing or exploding values
- Projecting image or word embeddings in the Projector tab to inspect clusters
- Profiling a few training steps to find data-loading or GPU bottlenecks
Frequently asked questions
How do I install the TensorBoard agent skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or into .claude/skills/ inside a project. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
What do I need to run it?
You need Python, the tensorboard package, and either PyTorch with torchvision or TensorFlow depending on your framework. You also need a training script of your own, because the skill adds logging to existing code.
Is TensorBoard free?
The skill is licensed under MIT. TensorBoard is a Google visualization toolkit installed locally with pip, and this page does not state any pricing or hosted service terms.
Does the skill run scripts or use the network?
According to the listing, the skill does not run scripts and does not use the network. The dashboard runs locally at http://localhost:6006 once you launch it.
Why does my dashboard show no data?
Check that the --logdir path matches the folder your writer used, and that your script ran and called writer.close() or flush so events were saved. Then refresh the browser page.
Is this better than a hosted experiment tracker?
TensorBoard runs locally and suits individual work or small comparisons. Teams that need shared, hosted experiment tracking across an organisation may find a dedicated tracking service a better fit.
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 train models in PyTorch or Keras and want the agent to add TensorBoard logging to your script and walk you through the dashboard. Skip it if you do not train models, want hosted team-wide experiment tracking, or expect the skill to analyse data without you running Python code in your own project.
Download the skill