Agentic Workflow

AMReX-Agent can be used to draft and iterate on REMORA input files with schema-aware context. This page keeps the setup close to the existing REMORA generic-inputs flow in this repository.

For MCP/Context7 setup in coding assistants, see Context7 for REMORA Workflows.

Basic Setup

  1. Clone amrex-agent and enter the repository.

    git clone --recursive https://github.com/AMReX-Codes/amrex-agent.git
cd amrex-agent

  2. Create and activate the demo environment.

    conda env create -f environment.yaml
conda activate amrex-agent-dev

  3. Export paths to local code checkouts.

    export REMORA_REPO_PATH=/path/to/REMORA
export AMREX_REPO_PATH=/path/to/amrex

  4. Build REMORA schema and indices for the agent.

    bash demo/setup_demo_database.sh --code remora --force-rebuild

    If you are testing pipeline mechanics without live embeddings/API access:

    bash demo/setup_demo_database.sh --code remora --force-rebuild --mock

  5. Run a REMORA smoke-test prompt.

    python amrex_agent.py \
  --baseline-override "REMORA/Exec/Upwelling" \
  --indexing-strategy override_static \
  --prompt "Run the REMORA Upwelling case to demonstrate wind-driven upwelling over a periodic channel."

Using inputs_generic in the Loop

Exec/Generic/inputs_generic is a generated parameter catalog that works well as a starting reference while iterating with an agent:

  1. draft case-specific inputs from inputs_generic sections,

  2. ask the agent to explain or edit targeted parameter groups,

  3. run short REMORA tests and adjust.

When REMORA ParmParse usage changes, regenerate the catalog:

bash Exec/Generic/build_schema_remora.sh
python3 Exec/Generic/render_inputs_generic.py

Review Rhythm

  • Keep generated-file updates and source changes in the same review context.

  • Treat AI edits as draft patches and verify behavior on a short run before scaling up.