Add nvidia-alchemi agent (NVIDIA ALCHEMI Toolkit) 1.1.2

[Andre Assis (anogassis)]

Summary

Adds the nvidia-alchemi agent: a prompt agent with one GPU tool built on the NVIDIA ALCHEMI Toolkit for batched atomistic simulation. It evaluates single-point energies and forces, relaxes geometries (FIRE/FIRE2) and runs molecular dynamics (NVE, NVT, NPT, NPH) with MACE and AIMNet2 potentials, which can be combined with Lennard-Jones, DFT-D3 dispersion and Ewald/PME electrostatics. It also supports fused multi-stage pipelines, inflight batching for screening, Zarr trajectories, MLIP fine-tuning and multi-GPU runs. The tool image builds from the public nvidia/cuda:13.0.1-devel-ubuntu24.04 base (pinned by digest) with exact package pins, and downloads the MACE and AIMNet2 checkpoints at build time with SHA-256 verification, so jobs run offline.

Type of change

• New agent (under agents/<name>/)
• Update to an existing agent
• New starter kit (under starter-kits/<name>/)
• Update to an existing starter kit
• Schema change (docs/schemas/**)
• Workflow / script change (.github/**)
• Documentation only
• Other (describe below)

Related issue / tracking

n/a

Schema impact

• No schema changes
• Schema changes are backward compatible
• Schema changes are breaking — add the breaking-change label and link the rollout plan below

Validation checklist

• Documentation (README.md, schemas) is updated to match the change.
• No hand-edits to .auto-registry/** (it is generated post-merge).
• No OS / editor artefacts committed (.DS_Store, .env, *.swp, .idea/, .vs/, etc.).
• If model-weight files are added, they are Git-LFS tracked, ≤5 GB each, and in an allowed format. (No weights are committed; checkpoints are downloaded and hash-verified at image build time.)
• No secrets, customer data, or internal-only URLs are committed.
• All Markdown links resolve.

Reviewer notes (optional)

• Changes are confined to agents/nvidia-alchemi/ (23 files).
• Ran locally on Linux before opening: .github/scripts/validate_pr.py passes with no findings, and .github/tests/test_schema_security.py passes (37 tests).
• spellcheck_prose.py reports two advisory SPELL-001 warnings, both false positives: "Lond" is in the citation Proc. R. Soc. Lond. A (THIRD_PARTY_NOTICES.md), and "COO" is the coordinate-format neighbor list listed for MACE in the potentials table in agent.yaml.
• Tool tests in tools/nvidia-alchemi/tests/ (32 tests) need the built image and a GPU, so they are not run by CI. They pass inside the image, and release 1.1.2 was also validated end to end on one NVIDIA H100 NVL: the smoke suite, offline single-point checks of all three bundled models, and a 100-step MACE NVT run with an independent kinetic-energy check. Multi-GPU execution and full training are documented capabilities but are outside that validation (see the README's "Verification status").
• Third-party licenses are listed in THIRD_PARTY_NOTICES.md and copied into the image under /app/licenses/.

[Andre Assis (anogassis)]

@microsoft-github-policy-service agree company="Microsoft"