Manage system prompts for multiple coding agents with agent-purge
This tool uses a local console or CLI to let users override, replace, or append a single system prompt (persona) across multiple coding agents like codex, claude, and grok. It supports hot-swapping and is fully reversible, with all data stored locally on your machine.
GitHubZiChuanLan921GitHub stars
agent-purgeNode.jsmagpie
Added 2026-10-31 · originally published 2026-10-30
Steps
- 1Install the tool globally by running `npm i -g agent-purge` (requires Node ≥ 20).
- 2Launch the local console for first-time use by running `agent-purge web`, view or select a persona at http://127.0.0.1:7788, and generate the local data directory `~/.agent-purge/`.
- 3Manage or switch personas by using `agent-purge ls` to list all personas, and `agent-purge use <id>` to switch the current persona.
- 4Configure the injection channel or permanent installation by running `agent-purge install <h>` to write configurations to the managed location of a specified harness, or use `agent-purge arm <h>` to enable the injection channel.
- 5Undo and restore by running `agent-purge uninstall <h>` or `agent-purge disarm <h>` to revert changes byte-for-byte if needed.
Copy-ready prompt
Act as an efficient prompt engineer and help me write a system prompt (persona) for [target Coding Agent, e.g., claude / codex]. The core goal of the prompt is: [enter the specific rules, output format, or role settings you want the AI to strictly follow]. Output the body content directly in Markdown format so that I can place it into the agent-purge personas directory.
#AI Agent#Prompt Engineering#Developer Tools#Automation
Related use cases
Summarized by AI from the original source - refer to the original post for details. Copyright belongs to the original author.