Eric Provencher

Provencher earned a BCSc at Concordia in 2016 with the computer games option. A university profile describes him as lead developer of Radiant Crusade, which his team adapted into a virtual-reality game after a university competition. His GitHub biography records former staff-engineer work on XR at Unity. He now identifies his work as Codex developer experience at OpenAI.[1][2][6]

Separate gathering context from reasoning about it

Repo Prompt assembles code and documents for analysis. Its Context Builder locates relevant material and creates a handoff for a separate reasoning model, with code maps and slices controlling how much detail enters context. OpenAI’s developer case study describes this separation explicitly: one agent curates the inputs, then another model concentrates on the analysis. The current product also coordinates coding agents through a native macOS interface.[3][5]

Release the tool’s active development branch

RepoPrompt’s website identifies Community Edition as the open-source home for ongoing development and Classic as an archived version of the original manual workflow. Community Edition combines context preparation, MCP and command-line integrations, planning and orchestration. Its public repository provides the application source and contribution path.[3][7]

In his Astra article, Provencher argues for revisiting accumulated skill descriptions, repository instructions and task prompts as models change. Short descriptions help an agent select the right skill, while supporting references let it load detail only when useful. He also recommends making completion and stopping boundaries explicit, so instructions do not force unnecessary work or cause an agent to stop before the requested result is ready.[4]