Image: GitHub
Why it mattersA developer running two or three coding agents stops editing the same system prompt in two or three places. One file sets the voice and the rules everywhere.
A developer who runs Claude Code in one terminal and Codex in another knows the small daily tax. Each agent keeps its own system prompt in its own file. Editing one voice means editing three places and keeping them in sync by hand. agent-purge is a free command-line tool that puts one shared persona across each agent on the machine. The repository was created on 25 October. It has 777 stars three days later.
The tool is written in JavaScript and runs on Node.js 20 or newer. It has zero dependencies. The licence is MIT. The author goes by ZiChuanLan.
What it does
A persona is a Markdown file in ~/.agent-purge/personas/. The top of the file sets a few fields in YAML, like a description and which agent it is for. The body of the file is the system prompt itself. One command then applies that persona across every supported agent at once.
The README lists several supported agents, including Claude Code, Codex, OpenCode and Grok. Each agent reads its system prompt from a different place on disk. agent-purge writes to each place in the format that agent expects.
A single /purge on, /purge off or /purge status command inside a running agent switches modes without leaving the session. The web console at http://127.0.0.1:7788 shows the same state in a browser tab.
Four modes
The project calls out four modes. "Replace" swaps the agent's own system prompt for the custom persona. "Strict" sends only the custom persona and drops the agent's tool descriptions. "Append" puts the custom persona before the agent's own prompt, so both are in play. "Off" deactivates every change and restores the original files.
The README says each change is reversible. The uninstall command "restore[s] files byte-for-byte". No agent binaries are touched and no authentication files are changed. The project does no network calls on its own.
Where it fits
A team wants one voice across its code style, review rules and release notes. Today it writes the rules once and copies them into each agent's own prompt file. agent-purge replaces that copy work with one file read from one place. Three agents on one laptop go from three edits to one.
The star count is the one hard signal. The 777 stars came from the agent's own community in three days, not from a review process. Early forks are high at 579. Both numbers go into the gate for a "breakout" project. Neither says how many of the stars are from active users.
Who it is not for
A team that only uses one coding agent gets no benefit. The gain comes from more than one tool reading the same prompt. A team on a shared build server also has to install the tool on each machine. Every change is local to the user's own files.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
