claude-mem-persistent-memorylisted

# claude-mem — Persistent Memory for Claude Code Brain-multiplier skill. Compresses past sessions into structured context and re-injects only the relevant chunks into future sessions. The reported token savings: **~75% per session**. Effect on the Octopus: every arm's runtime budget multiplies, every long engagement gets cross-session memory automatically. ## When to use - A client engagement spans many sessions and context drift is hurting quality - Token spend is the bottleneck on a high-frequency arm - You want session-to-session continuity that survives `/clear`, compaction, and machine switches - You're evaluating whether to standardize cross-session memory across all arms ## When NOT to use - One-shot tasks (overhead not worth it) - Arms whose deliverables cannot carry a third-party dependency at all. The copyleft objection that used to sit here was about AGPL and does not apply: upstream has been Apache-2.0 since 2026-05-08. See "License caveat" below for what to check instead - Engagements where the brain's existing auto-memory at `~/.claude/projects/<arm>/memory/` already covers the need (it's lighter, no compression, no ChromaDB) ## Source of truth - Repository: `github.com/thedotmack/claude-mem` (76.5k+ stars at time of writing — verify current state) - License: **Apache-2.0** upstream since 2026-05-08 (36b0929f), before this skill entered on 2026-05-18. It was AGPL-3.0 until then, which is what an older reading of this repo would tell you, so check the LICE