One persona spans coding agents

A coding agent’s system prompt helps set how it behaves. But different tools read that prompt from different places, which can make it awkward to keep the same instructions when switching agents. The GitHub project agent-purge offers a local console to replace or add prompts across a range of coding-agent tools. Its README describes four modes, including replacing the prompt, appending a persona, and turning changes off. The project says changes can be reversed and that its state is stored locally, not uploaded.

For an operator, the practical idea is consistency. If a team uses more than one coding agent, a shared persona could make it easier to carry the same working instructions between them. That doesn’t mean every tool will behave identically, and the project notes that support varies by agent. Some prompt changes are rougher than others, and tools that keep prompts on a server may not support injection. The project reports 874 stars and a pace of 192 stars per day, though stars alone don’t show how many people use it in real work. The project’s README lays out the supported tools and limits.

For businesses, the useful question is whether shared instructions reduce the effort of managing multiple agents without creating extra maintenance. Teams can assess whether a shared persona fits their own workflows.