Idea
Runs converge on their own (width × depth default to ∞), so a deep tree can take a while. Two ergonomics improvements:
- Progress summary — a periodic one-line status (nodes derived / rounds / open frontier / framings exercised) so the user can see it's making progress toward convergence rather than stalling.
--resume polish — re-invoking with the same --out <dir> already resumes from the incremental write (§7.1); make the resume path explicit and ergonomic (detect the latest run dir, report what's being resumed, skip completed leaves cleanly).
What's involved
Mostly in the engine spec / skill orchestration (docs/ENGINE.md §7–§8) and how tree.json state is read back. Worth a short design note in the issue thread before implementation.
Idea
Runs converge on their own (width × depth default to ∞), so a deep tree can take a while. Two ergonomics improvements:
--resumepolish — re-invoking with the same--out <dir>already resumes from the incremental write (§7.1); make the resume path explicit and ergonomic (detect the latest run dir, report what's being resumed, skip completed leaves cleanly).What's involved
Mostly in the engine spec / skill orchestration (
docs/ENGINE.md§7–§8) and howtree.jsonstate is read back. Worth a short design note in the issue thread before implementation.