Test LLM chat features offline with a terminal simulator over fake in-memory stores
To test conversation memory without deploying, add a small module CLI (python -m app.cli.chat_sim) that runs the real conversation function and real LLM adapter against fake in-memory members, tasks and completions, and keeps history with the same window and size limits as production. Inject the input and output functions and the interpreter into a run() function so tests can drive it with a scripted interpreter, and pipe printf input into it for a quick live check. Gotchas: a SimpleNamespace stand-in for settings misses fields the real code reads, so build the real Settings object instead (no network call); and a pre-edit hook that blocks only the first edit per file can leave a half-applied change, so recheck which edits actually applied.