autoresearch-initlisted

# Autoresearch Init Collaborative setup for an autonomous experiment loop. The goal is to produce session files that a loop agent (via `/autoresearch`) can pick up and run indefinitely without further guidance. ## Before You Start Check if `autoresearch.md` already exists in the repo root. If it does, ask the user: are we starting fresh (overwrite) or modifying the existing session? If modifying, read the existing files first and adjust rather than rewriting from scratch. ## Step 1: Understand the Goal Have a conversation. Don't rush to write files. Gather these through discussion (or infer from `$ARGUMENTS` and context): - **What are we optimizing?** (perf, bundle size, accuracy, Sharpe ratio, val_bpb, test speed, etc.) - **How do we measure it?** (benchmark script, test suite, backtest, build command, etc.) - **What files can the agent modify?** Ideally 1-3 files. More = harder to reason about. - **What must NOT change?** (tests, API contracts, data formats, date ranges, etc.) - **How long does one run take?** This determines how many experiments fit in a session. ## Step 2: Design the Evaluation Strategy This is the most important part. Get it right here and the loop runs itself. Get it wrong and the loop thrashes. ### Think about metrics carefully The obvious metric isn't always the right one. Push the user: - "You say you want to optimize speed — but speed of what exactly? Cold start? P99? Average?" - "Is there a metric that captures what you actually care abo