NabilEkoWahyudi/eco-agent — reverse-engineered prompt
Reverse engineered prompt
Build me a lightweight terminal AI agent called Eco Agent that feels good to use in a real coding workflow. I want a CLI with an interactive chat mode and a one shot mode, where the agent can plan, take actions, observe results, and keep looping until a task is done. It should work with cloud and local models, especially Groq, OpenRouter, Ollama, plus a mock option for testing, with a simple first run setup wizard for picking a provider and saving the API key.
Make it able to read and edit files, run shell commands, search the project, and use a web search plugin. Before it changes or deletes anything, show me a clear diff and ask for approval. I also want project scanning with an init command, saved sessions I can resume, smart handling of large pasted context, and handy chat commands like commit message generation, PR text generation, debug a failing command, plan mode, act mode, and a simple swarm mode for parallel agents. If needed, look up current docs online and wire in basic MCP support too.
Want more depth? Deep Reverse