Chinese deep-dive review: agent-autoloop — recoverable CLI task loops

[LAIguapi]

Hi! I wrote a detailed Chinese-language review of agent-autoloop, focusing on a small but important workflow: what happens when a command-line agent needs a human answer before it can continue.

The review covers:

• the deterministic kernel, fact log, and rebuildable projection;
• the plugin and HTTP protocol boundaries around external agents and notifications;
• the worker's durable run state, process-group handling, and recovery model;
• a local end-to-end test of install, start, try, answer, and status with a fake CLI.

The practical conclusion is deliberately narrow: the project is promising as a lightweight local scheduling and recovery layer for an existing CLI tool, while real-model behavior, real notifications, and cross-platform compatibility still need separate validation.

The full review is in Chinese and includes the terminal evidence and screenshots from that isolated test. Feel free to ask questions in English or Chinese!