memo-session-skill
Memo Session Skill
Persistent AI agent memory for coding agents — decisions, gotchas, workarounds, and open loops survive context window resets because they live in git-tracked files, not volatile chat.
The context window is working memory: it clears when the session ends. Cross-session persistence needs a write path — memory consolidation and memory routing into typed layers. Memo Session Skill runs preflight → consolidate → classify → route → conflict gate so knowledge lands in MEMORY.md, memory/ (HOT/WARM/COLD), project wiki, AGENTS.md, skills, or optional portfolio memory — with temperature limits and no duplicate paragraphs across channels.
Standalone skill — no vector database, no managed platform, no network calls during sessions. Optional goal-mode checkpoint hooks: references/goal-mode-integration.md.
How this differs
| This skill | Not this |
|---|---|
| Git-tracked persistent memory for coding sessions | Mem0, Zep, Letta — managed vector/graph memory platforms |
| Write-path memory consolidation from live sessions | RAG — read-heavy static knowledge corpora |
Typed memory architecture (HOT/WARM/COLD in memory/ + wiki) |
Embedding stores, hybrid retrieval pipelines |
| Cross-session continuity and session handoff | Enterprise customer-memory or contact-center products |
Complements RAG and managed memory layers; does not replace them.