Episodic Memory System
Track failures and decisions in structured logs so agents learn from mistakes across sessions.
Overview
AI agents write failures to memory/failures.jsonl and strategic decisions to memory/decisions.jsonl. Each entry includes root cause, prevention steps, and resolution status. New sessions load recent entries to avoid repeating mistakes.
How It Works
When a fix fails, tool behaves unexpectedly, or a pattern breaks, the agent appends a failure entry with structured fields (category, root cause, prevention, resolved boolean). Major decisions get logged with rationale, alternatives considered, and outcome. Agents load the last 10 entries from each log at session start.
Tools Used
Outcome
Logged 34 failures and 28 decisions since January 2026. Prevented 7 repeat failures by catching known issues during session initialization. Decisions log provides audit trail for strategic pivots.