Speaker
Description
This talk explores practical applications of agentic AI within the Python in Heliophysics Community (PyHC). We show how customizing off-the-shelf agentic coding harnesses can deliver 90% of the benefits of a purpose-built AI agent with 10% of the effort of creating one from scratch. We do this using instruction files, reusable skills, subagents, and MCP servers to adapt tools such as Claude Code, Codex, Copilot CLI, and Gemini CLI to PyHC-specific workflows. These customizations are embedded directly in project repositories, so users can immediately work with the agents in whatever compatible harness they already use or prefer.
The talk demonstrates three agent-powered tools developed for PyHC: (1) the “PyHC Standards Evaluator,” an agent that automates the grading of packages against PyHC's development standards; (2) the “HSSI Metadata Extractor,” which parses software repositories to extract the metadata needed for submission to the Heliophysics Software Search Interface (HSSI); and (3) “PyHC-Chat,” an agent designed to answer questions about the PyHC ecosystem and its roughly 100 packages, help users identify and install the right packages for specific scientific tasks, write code with them, draft (executable) papers, and much more.
As a broader takeaway, the approach behind PyHC-Chat—pairing a curated collection of resources with “lay of the land” instructions—is a reusable pattern attendees can apply to create customized agents for whatever organization, community, business, or specialized field matters to them.