Cooking with GStack today. I just dropped 10 PR bug fixes from the community plus a big refactor of E2E CI tests, which should help stability overall.

/autoplan is suddenly much more usable. The more automated something is, the more likely having extra eyes on it from Codex will help you avert bugs and make the right choice. Also when the two co-authors disagree, the user can adjudicate and that's a good outcome.

Some serious reduction in context in this PR T1 skills: ~40% reduction (~105 lines / ~1200 tokens saved). These skip AskUserQuestion format, Completeness Principle, Repo Mode, and Search Before Building — none of which they use. T2 skills: ~15% reduction (~30 lines / ~400 tokens saved). These skip Repo Mode and Search Before Building.

Also updated better Windows support