reflectlisted

# Reflect Mine the current conversation for durable learnings, then route each one into a skill edit. The output is edits to skills, not notes. ## When this fires - The user said "reflect". - The user corrected the approach mid-task — especially a correction they have made before. - A complex task (5+ tool calls) just landed cleanly and the recipe is worth keeping. - Dead ends were hit, the working path was found, and the path generalizes. - A non-trivial workflow emerged that isn't captured anywhere. Skip when the conversation is trivial, off-topic, or already covered by a skill that was followed correctly. One-offs are not learnings. ## The highest-value finding shape A correction the user has already given in previous sessions, and is giving again, is worth more than any novel insight in the transcript. Repetition is the signal: it means the existing encoding failed. The specific failure mode to watch for is **a rule that already lives in `~/.claude/CLAUDE.md` and still gets violated.** Adding a sixth restatement of it to a seventh file is not a fix; it is the thing that already didn't work. When a finding lands on an already-documented rule, do not accept a prose edit. Route it to a mechanism — a hook, a skill description that actually fires, a verification skill, a lint rule, a script — or send it to Backlog. This is what the **principle-encode-lessons-in-structure** skill is for, and step 4 enforces it. ## Process ### 1. Locate the active transcript Claude Cod