wraplisted

# wrap End-of-session consolidation. The friend runs `/ren:wrap`; this skill writes the session's L1 narrative summary (always — quarantined as unreviewed LLM-auto content, never treated as instruction) and gates any candidate durable items through the classifier, auto-applying them (revertible) unless held for a contradiction or surfaced as a promotion suggestion. Per spec §3.1's discipline, most sessions should produce **zero** durable candidates — the classifier biases hard toward "not durable." ## When to use this skill - Friend invokes `/ren:wrap` at (or near) the end of a session. - Friend says something like "let's wrap up", "consolidate this session", "what should we remember from today." ## When NOT to use this skill - Friend wants to pin one specific fact mid-session → `/ren:pin`, not `/ren:wrap`. - Friend wants to look something up → `/ren:recall`. - No session content exists yet (this is the very first turn) → nothing to wrap. ## Behavior 1. **Compose the L1 narrative.** The live session writes this ITSELF — never a subagent (`execution_tier: judgment`; the exception to worker delegation: only the main model holds the conversation being summarized in context). A short narrative markdown summary of what happened this session — what was done, what's open, what changed. Lead with outcomes. Target ≤1,000 tokens. This is data, not doctrine; it always gets quarantined on write (queue Task 2.4 wiring), so nothing here needs to hedge its own confidence. 2. **Extrac