Agent Memory Bank

A persistent memory system for storing knowledge that survives across conversations — for any domain: code, business, personal, research, creative work, client management, and more.

Universal and env-agnostic: works in any repo, any domain. How you wire it (alias, hotkey, auto-inject, parallel runner) is yours to decide — the skill only guarantees the scripts work from project root. Default use is session end via learn; orientation at session start is one command (get_mem.py).

Default location: ./memory/ (create if it doesn't exist; scripts accept an explicit path — root-agnostic, you set it)

Default structure:

• ./memory/episodic/ — active events (last 7 days, then auto-archive)
• ./memory/semantic/ — durable facts and current state
• ./memory/procedural/ — repeatable workflows and how-tos
• ./memory/pending.md — append-only inbox for micro-capture while working
• ./memory/SEED.md — hand-curated bootstrap, injected at session start
• ./memory/INDEX.md — auto-maintained living map (never hand-edit)
• ./memory/CHANGES.md — auto-generated newest-first feed (never hand-edit)
• ./memory/archive/ + summaries/ — auto-managed compression