← ClaudeAtlas

memory-distilllisted

Post-mortem-style review of recent feedback memories. Pick 3-5, walk through them one at a time asking "did I get this right? what's the next layer?", then surface latent principles that several memories point at but none articulate. Use when memories feel stale, when the same correction keeps recurring, or periodically to consolidate. Works from concrete recent evidence — not vibes. Invoke as "/memory-distill".
aberson/skill-mesh · ★ 2 · AI & Automation · score 71
Install: claude install-skill aberson/skill-mesh
# Memory Distill Goal: surface **latent principles** — patterns two or three memories point at but none articulates alone. Requires re-reading specific recent feedback. ## When to use - Periodic check-in (every ~10-20 sessions, or when the user asks) - After a session where the user gave several corrections — the new feedback may rhyme with older feedback in a way worth naming - When the same correction keeps recurring despite a memory existing for it (the memory may be capturing the symptom, not the cause) - When memories feel cluttered or contradictory and need consolidation ## When NOT to use - During active work on a task — this is reflective, not action-oriented - Right after writing a memory — give it a few sessions to prove out first - For `user`/`project`/`reference` types — those are facts, not patterns. This skill is specifically for `feedback`, where the "next layer" question is meaningful. --- ## Flow ### Step 1 — Locate the memory directory Memory lives at a project-scoped path, e.g. `~\.claude\projects\<slug>\memory\`. Read `MEMORY.md` to get the ordered index, then `ls` the directory to confirm what's actually on disk. If `MEMORY.md` and the file list disagree (orphan files, dead pointers), flag that to the user before continuing — it's a separate problem to fix first. ### Step 2 — Select 3-5 feedback memories Default: the **most recent feedback memories** by file mtime, capped at 5. Show them as a numbered list with their one-line description