autoresearchlisted

# Autoresearch Autonomous experiment loop: try ideas, measure, keep what works, discard what doesn't, never stop. ## Startup 1. **Check for session files.** Look for `autoresearch.md` in the repo root. - If missing: tell the user to run `/autoresearch-init` first. Stop. - If present: read it along with `autoresearch.sh`, `autoresearch.ideas.md` (if exists), and `results.jsonl` (if exists). 2. **Identify the trunk branch.** Read the `## Branch` section of `autoresearch.md` — it contains the trunk branch name (e.g., `autoresearch/sharpe-2026-03-14`). Switch to it: `git checkout <trunk>`. 3. **Understand the state.** Check `git log --oneline -20` and `results.jsonl` to see where things left off. How many runs? What was tried? What's the current best? What branches exist under `autoresearch/exp/*`? 4. **Read the source files in scope.** Every file listed in autoresearch.md's "Files in Scope." Read them fully — deep understanding beats random mutations. 5. **If `$ARGUMENTS` contains a specific idea**, try that first. Otherwise, check `autoresearch.ideas.md` for the highest-priority untried idea. If nothing there either, form your own hypothesis from the results history and source code. 6. **Start looping immediately.** Do not ask permission. ## The Loop **LOOP FOREVER.** Never ask "should I continue?" — the user expects autonomous work. **NEVER STOP ON YOUR OWN. The user may be sleeping. Keep going until interrupted.** ``` REPEAT: 1. Think: review results.jsonl