Objective

Build a routine that sweeps the SAIL Knowledge Base on a schedule, classifies each active handoff by evidence (git log, referenced files, Notion activity, status flags), and surfaces the ones that actually need Sam's attention. Per earlier approval set (Q1A–Q6A):

• Hybrid GitHub + Notion storage
• Twice-weekly GitHub Actions schedule (Tue + Fri 07:00 ET)
• One Claude subagent per handoff for deep reasoning (scaffolded, not yet wired)
• Auto-execute PARTIAL handoffs under Gate-Only-at-Go-Live rule (opt-in via --commit)
• Cowork artifact + Notion view for the dashboard
• Scaffold first, Notion read-only dry-run, review before writeback goes live

What got built

• routines/github/handoff_auditor/run.py — query, evidence check, verdict, writeback (gated)
• routines/github/handoff_auditor/CLAUDE.md — runtime/schedule/inputs/outputs/state/severity docs
• .github/workflows/twice-weekly-handoff-audit.yml — Tue + Fri 12:00 UTC cron, workflow_dispatch for manual with --commit and --limit flags
• outputs/2026-04-23/handoff-auditor.md — full dry-run report
• outputs/2026-04-23/handoff-auditor.json — structured bucket + row data for dashboard consumption
• outputs/handoff_dashboard.html — self-contained filter/sort dashboard (opens in any browser)
• state/handoff-auditor.json — counts + page ID sets per verdict

Dry-run results (all 118 active/blocked/follow-up handoffs)

• DONE: 9 — git commits match topic, files exist or strongly imply completion. Medium confidence across the board, so auto-close stays OFF.
• PARTIAL: 1 — git activity present but referenced files missing.
• NEEDS_EXECUTION: 8 — handoffs describe next steps with no matching git work. Top of Sam's action list.