AI Agent Integration#

gpCAM ships with a set of Claude Code skills that guide an AI assistant through designing autonomous experiments — custom kernels, acquisition functions, noise and cost models, the full ask / tell / train loop, and large-scale or multi-task setups.

Experimentalists who want smart, autonomous data acquisition without deep knowledge of GP math or the gpCAM API can let an AI assistant translate a plain-English description of their measurement into a working gpCAM script.

Installing the marketplace in Claude Code#

The gpCAM repository is published as a Claude Code plugin marketplace. Inside any Claude Code session, run:

/plugin marketplace add lbl-camera/gpCAM
/plugin install gpcam@gpcam

The first command registers this repo as a marketplace and pulls its marketplace.json manifest. The second installs the gpcam plugin from that marketplace — a single plugin that bundles every skill listed below. Once installed, the skills are available to Claude in any project on your machine; you do not need to clone gpCAM locally.

Useful follow-up commands:

Command	What it does
`/plugin marketplace list`	Show the marketplaces you’ve added.
`/plugin marketplace update gpcam`	Pull the latest skill versions.
`/plugin list`	Show installed plugins.
`/plugin uninstall gpcam@gpcam`	Remove the plugin.

Using the skills#

Once the plugin is installed, Claude will activate the appropriate skill automatically when you describe an experiment-design problem. For example:

“I want to map the photoluminescence of a thin film over a 2-inch wafer. Each measurement takes 30 seconds, motor moves are slow in the X direction, and I have 4 hours of beam time.”

…will trigger the experiment-designer skill (and pull in cost-functions for the asymmetric motor cost). You can also invoke a skill explicitly:

“Use the kernel-designer skill to build a periodic + Matérn kernel for a temperature-dependent diffraction scan.”

Available skills#

Skill	Description
experiment-designer	End-to-end autonomous experiment design. Translates a scientist’s description of their measurement into a complete, runnable gpCAM script.
kernel-designer	Design and compose custom kernel functions that encode domain knowledge — smoothness, periodicity, symmetry, anisotropy, non-Euclidean inputs.
acquisition-functions	Write custom acquisition functions that encode experimental priorities: exploration vs exploitation, multi-objective targets, constraints, cost-aware acquisition, UCB / LCB, probability of improvement.
prior-mean-functions	Encode known physics or expected trends as prior mean functions, so the GP regresses against a baseline rather than a flat zero prior.
noise-functions	Model position-dependent, heteroscedastic, or count-rate-dependent noise from detector characteristics.
cost-functions	Account for motor travel time, settling, directional costs, sample damage, beam time, and zone-based penalties.
gp2scale-advanced	Large-scale experiments (>10k points up to millions) using sparse, compactly-supported kernels and Dask distributed computing.
multi-task-advanced	Multi-output, vector-valued, or function-valued experiments using `fvGPOptimizer`.

Using the skills outside Claude Code#

The skills are plain Markdown SKILL.md files and are compatible with any agentic harness that can read them — for example OpenClaw. To use them outside Claude Code:

Clone the gpCAM repository:

git clone https://github.com/lbl-camera/gpCAM.git

Point your assistant at the skills/ directory at the repo root.

When the cloned repo is present in your working directory, Claude Code also picks up the root CLAUDE.md and skills/ directory automatically, so the marketplace install is only needed for use outside the repo.

Source#

The marketplace manifest lives at .claude-plugin/marketplace.json and the per-skill content under skills/. Contributions and new skills are welcome — open a pull request.