Prev-Talk | Next-Talk | All-Talks | Talks-Sorted | MOOS-DAWG'26
Talk-19: Agentic Coding and MOOS-IvP
Charles Benjamin
McGill University, Centre for Intelligent Machines
MOOS-IvP Skills is a plugin for agentic coding tools like Claude Code and Codex that gives them the instructions and reference material needed to work effectively with MOOS-IvP. The project has been in development since spring 2026 and consists of an interwoven set of 'skills' that guide tasks such as app and behavior design, mission construction, Monte Carlo test-harness development, documentation lookup, post-mission analysis, and more.
The project has already supported development of several tools, including moos-map, a PyPI-distributed TIFF map generator accessible through a GUI, CLI, or the skills plugin; a substantially improved MOOS-IvP Visual Studio Code extension with enhanced highlighting, hover definitions, and parameter validation; and a NotebookLM notebook for new MOOS-IvP students. Benchmarking and feedback from a June mini-course for Greek military personnel indicate that the notebook improves conceptual learning while reducing the need for human TA support. CI/CD infrastructure and related testing practices will be discussed in the following presentation.
Overall, applications have ranged from classroom instruction to core feature development and completed master's thesis work. Additional use cases for the broader MOOS-IvP community will be presented during the talk. Early access requests may be directed to cbenj@mit.edu.
Categories:
- Generative AI