simlisted

# Sim Run one simulator issue loop with the fewest moving parts that still produces evidence. Default loop: 1. `launcher` gets the daemon and app into a known-ready state. 2. `tester` runs one concrete scenario and captures logs. 3. The main agent reads the evidence, forms the hypothesis, and adds instrumentation or code changes locally. 4. `launcher` rebuilds if needed. 5. `tester` reruns the same scenario. 6. The main agent compares before and after and decides whether the issue is fixed. Prefer observation over inference. If the logs do not show the behavior clearly, add instrumentation and rerun instead of debating the code. ## Default Behavior - Do not create a plan file unless the work is going to span multiple rounds or the user explicitly asks for one. - Do not require `planner`, `analyst`, `solver`, or `reviewer` by default. - Keep the scenario fixed across before and after runs unless the current scenario is clearly the wrong probe. - Prefer one rich scenario over many small ones. For chat rendering and streaming issues, prefer `mixed-markdown-multi-tool.txt`. Use `smoke-hello.txt` only for launcher and connectivity checks. ## Optional Agents Use extra agents only when they materially reduce mistakes, and only when the user wants subagents or delegation. - `analyst`: use when the tester logs are ambiguous and you want a second pass on the evidence. - `reviewer`: use when the fix is risky or the before and after comparison is subtle. - `solver`: use for a bo