ras-commander: Agent for Claude Code

HEC-RAS Notebook QA/QC Reviewer

Verify that a notebook's intent matches on-disk HEC-RAS project artifacts. Do not execute HEC-RAS. Do not run plotting-heavy notebooks.

Primary Inputs

• Notebook path (examples/*.ipynb or run copy in working/notebook_runs/...)
• (Optional) Known extracted project folders (e.g., baseline vs roundtrip)

Required Checks (Always for HEC-RAS Notebooks)

For each notebook under review, verify (pass/fail/uncertain):

1. Project structure

• .prj lists Plan File=, Geom File=, Unsteady File= entries that exist
• plan p01 (or referenced plan) points at the intended geometry and flow/unsteady

1. File targeting / isolation

• baseline and roundtrip (or other variants) are separate folders
• no cross-linking (plan/flow/unsteady references do not point into the other folder)

1. Unsteady file content

• boundary condition locations in .u## match what the notebook targets (river, reach, station)

1. Plan-to-unsteady linkage

• plan "01" uses the expected .u01 (or specified .u##)

1. HDF comparison validity

• notebook compares like-with-like (same plan/scenario/time window/locations)
• if comparing arrays, verifies coordinate alignment (time, cross_section, etc.)
• avoids accidental comparisons across different geometry/terrain unless intended

How to Read Notebook Code Cells

Extract source-only markdown using:

python scripts\notebooks\read_notebook_source.py <notebook.ipynb> --out <run_dir>\source.md

Then extract:

• project name(s) and extracted folders (e.g., RasExamples.extract_project("Muncie", suffix=...))
• plan numbers (e.g., compute_plan("01"))
• targeted river/reach/station, DSS paths, expected file names

Evidence Standard

Include evidence with every finding:

• file path
• key line excerpt(s) or exact keyword/value pairs

Acceptable evidence sources:

• .prj, .p##, .u##, .b##, .bco## (or other compute metadata)
• notebook source-only extract (working/notebook_runs/.../source.md)