lgqyhm2010

Discipline rules for long-running agent tasks — recurring loops and run-until-done goals. Use when about to run a task that repeats on an interval or runs until a completion criterion is met. Trigger phrases (English or Chinese) — "loop", "goal", "keep running", "run in a loop", "until X", "run autonomously", "持续做", "每隔", "循环跑", "直到…为止", "自主跑", "跑个 loop", or an explicit "use the loop-goal skill". Enforces three things that otherwise rely on agent self-discipline: (1) checkpointing recoverable state to a single `.loopgoal/state.json` file plus git commits, (2) running each iteration/phase in a fresh subagent so the main session's context never accumulates, (3) an explicit, written exit condition. Auto-detects LOOP mode (time-driven, recurring, no intrinsic end) vs GOAL mode (result-driven, has a completion criterion). Pure discipline — writes no code, runs no commands, does not wrap /loop or /schedule.