Hey everyone 👋 I'm Adam. I built AgentBubble after repeatedly running into the same problem while using Claude Code and Codex on real projects. The agents were often capable of solving the problem, but they would sometimes touch files outside the intended scope, miss important project constraints, or make changes that weren't part of the original task. AgentBubble is my attempt to solve that. It creates a local workspace where you define the current ticket, scope, architecture notes, risky systems, and acceptance criteria before the agent starts working. Afterwards, it can audit changed files against the original task. It's intentionally simple: * Local-only * No telemetry * No hosted service * No agent runtime I'd love feedback from anyone actively using Claude Code, Codex, Cursor, or similar coding agents.

AgentBubble

Stop coding agents from wandering outside the ticket.

Stop coding agents from wandering outside the ticket.