Memory System

The @saas agent team uses a two-tier memory system to persist knowledge across sessions.

Tiers

Tier 0 (global):  ~/.claude/template-memory/    cross-project wisdom (15 files)
Tier 1 (project): docs/context/                 this project's state (16 files)

Tier 0 — Cross-Project Wisdom

Tier 0 contains generalizable lessons extracted from all projects. It's stored in ~/.claude/template-memory/ so it's available in any Claude Code session.

/athena:promote     # extract Tier 1 lessons → Tier 0
/athena:forget      # archive weak Tier 0 lessons (Ebbinghaus brake)

Tier 1 — Project State

Tier 1 is this project's specific context, stored in docs/context/:

File	Purpose
session-summary.md	Last session's state — read at session start
epic-progress.md	Lightweight state file (IDs + status only)
agent-memory.md	Each agent's accumulated knowledge
migration-review/	SQL artifacts for the migration review gate

Commands

/athena:save    # checkpoint all agents → write context docs → commit
/athena:load    # load project context → summarize state (session start)
/athena:learn   # refresh MEMORY.md accuracy → detect drift → suggest promote

Memory-Aware Planning

/athena:plan brainstorm reads Tier 0 lessons scored by relevance and injects them as design considerations before the brainstorm dialogue begins.

/athena:plan brainstorm "notifications system"
# → loads relevant Tier 0 lessons
# → 7-step Q&A dialogue
# → emits enriched epic file