v1.4.0 — Self-Improving Agent + Bug Fixes
v1.4.0 — Self-Improving Agent + Bug Fixes
- New local skill:
self-improving-agent— autoresearch-inspired loop that analyzes feedback issues, runs eval scenarios, proposes skill improvements on a dedicated branch, and keeps/discards based on score comparison - New eval system: Test scenario runner with 3 baseline scenarios
- Fix: Completion checkpoint now inlines full packager/pruner scripts
- Fix: Context fields resolve correctly via env vars with .active-workflow fallback
- Fix: Duration calculation added to metrics packager
- Fix: Step tracking fires in all skills with .active-workflow step updates
- New: Incremental metrics packaging at brainstorming and writing-plans handoffs
See RELEASE-NOTES.md for full details.