cold-recoverylisted

# /cold-recovery > **Continuity gate.** Reconstruct session state before first response. Lighter than `/boot` — no fleet hard-stop, no persona picker, no sentinel. > > **When to run:** > - User says "keep working", "pick up where we left off", or pastes the cold-recovery protocol > - New session after runtime switch (claude.ai ↔ Claude Code ↔ Cursor) > - Agent detects cold start and continuity is the bottleneck (not infrastructure) > > **When not to run:** > - User explicitly skips context ("sandbox", "load without context", "no startup", "don't run boot") — run cold-recovery only; do **not** run `/boot` > - Infrastructure is the question — use `/boot` or `/startup` instead > - Physical or personal emergency — respond first, recover after --- ## Relationship to boot | | **Cold recovery** | **Boot (`/boot`)** | |---|---|---| | Steps | 5 | 14 | | Postgres gate | No hard stop | Hard stop if down | | Grove / ledger / SOIL stack | Skip unless a gap requires it | Full pass | | Persona picker | Skip unless user asks | Render picker | | Primary source | Conversation + indexes | Infrastructure + handoffs | | Latency target | One parallel batch | Full fleet orientation | **Rule:** Cold recovery answers *"where were we?"* Boot answers *"is the fleet healthy and wired?"* Use cold recovery when continuity is the bottleneck. Use boot when infrastructure is the bottleneck. They are not replacements for each other. --- ## Step 0 — Detect runtime Identify which index layer exists. **