Episodic, Semantic, and Procedural Stores

PROJECT CONTEXT:
All my agent's memories live in one pile, so session stories, project facts, and how-to knowledge interfere with each other. Memory surveys and taxonomies consistently separate episodic, semantic, and procedural stores with different schemas and lifecycles.

GOAL:
Split memory into three typed stores - episodic (events), semantic (facts), procedural (skills) - with type-appropriate schemas, recall, and lifecycles.

REQUIREMENTS:
- Three stores with distinct schemas: episodes, facts, and procedures
- A classification step routing new memories to the right store
- Type-appropriate recall: episodes by time/task, facts by topic, procedures by goal
- Distinct lifecycles: episodes decay fastest, facts need confirmation, procedures update on failure
- Cross-links: facts and procedures cite the episodes that produced them

CONSTRAINTS:
- One entry, one store - content that spans types is split, not duplicated
- Procedures must be actionable as written: steps, preconditions, verification

DEFINITION OF DONE:
- New memories route to the correct store in a 12-case classification test
- Recall for 'how do I deploy?' returns a procedure, not a war story
- An episode where a procedure failed triggers a procedure update
- Every fact and procedure traces back to at least one source episode

COMMON FAILURES TO AVOID:
- Facts phrased as stories, so recall returns anecdotes when the agent needs truth
- Procedures without verification steps - recipes that cannot be trusted
- Episodes hoarded forever at full fidelity, bloating the store
- No links between layers, so a failed episode never corrects the stale procedure