What We Shipped

• Workout Quality Scoring — New scoring system that evaluates completed workouts against their targets. Trends show up on the dashboard so athletes can see consistency over time. Also fixed marathon target time parsing (H:MM format) and auto-applies goal-derived threshold pace — but only when the user hasn't manually set one.

• Admin Dashboard — Consolidated /admin page for managing the platform. One place instead of scattered routes.

• Calendar UX Improvements — Press T to jump to today (small thing, big quality of life). Completed workouts now have a delete button in the activity modal.

• Sleep Card Enhancements — Bedtime and wake time now display on the Sleep Hours card, pulling from Garmin data.

Behind the Scenes

• Sub-Agent Skill System — Inspired by OpenAI's "Shell + Skills + Compaction" article, we restructured our agent architecture. Five workspace-level skills (kanban management, build logs, bug triage, code review dispatch, Discord support) now have proper routing descriptions with negative examples. Sub-agent SOUL.md files got "use me when / don't use me when" sections to reduce misfires.

• Sub-Agent Spawning Issues — Attempted to dispatch backlog tasks to Coder agent. Researcher delivered (PR #323 merged in 4 minutes), but Coder sessions silently failed — created but never processed. Root cause TBD.

What's Next

• Investigate coder agent spawn failures
• Continue backlog triage (29 items and counting)
• Agent API buildout