PromptBridge: AI Coding Agents you can Trigger from Chat and Trust

Abstract

AI coding agents such as Claude Code, Cursor, Codex, and OpenCode have matured into capable multi-step software engineers, yet nearly all of them still assume a developer is sitting at a terminal. PromptBridge removes that assumption by connecting six messaging platforms, Telegram, Discord, Slack, Microsoft Teams, GitHub, and Email, to four coding agents through a single dispatcher. This paper presents the complete, matured architecture of PromptBridge: the four-layer design, the session and git-aware change tracking model, the security and configuration layer, and the packaged Electron desktop application. Beyond describing the current system, this paper evaluates it against related agent frameworks and bot toolkits, sets out a formal threat model, examines the structural checkpoints that keep a human in the approval loop for every agent-initiated change, and lays out a concrete future scope covering sandboxed execution, richer output formatting, session summarization, and controlled user studies. The goal is to give a single, self-contained account of what PromptBridge is today and where it is going next.