auto-memory

name	auto-memory
description	Capture learnings, patterns, and corrections discovered during this session into persistent project memory. This runs automatically at the end of `session-handoff` and can be invoked standalone.
metadata	{"author":"cascadian-gamers","version":"1.0"}

Auto Memory

Capture learnings, patterns, and corrections discovered during this session into persistent project memory. This runs automatically at the end of session-handoff and can be invoked standalone.

When to Run

• End of every session (chained from session-handoff)
• After a bug is found and fixed (capture the pattern)
• After a code review reveals a recurring issue
• When the user says "remember this" or "don't forget"

Memory File

.kiro/steering/memory.md — loaded every session via steering, survives across sessions.

Process

Phase 1: Scan Session for Learnings

Review the conversation for:

1. Bugs found — what broke, why, how it was caught (e.g., "renderedHtml undefined in loadSession")
2. API contract surprises — values that didn't match docs (e.g., "ENABLED" vs "enabled")
3. Pattern corrections — things the AI got wrong and had to be corrected (e.g., wrong payload format for Lambda invoke)
4. Stakeholder preferences — new preferences expressed this session
5. Tool/SDK gotchas — things that work differently than expected
6. Review findings that recur — if the same finding keeps coming up, it's a pattern to remember

Phase 2: Deduplicate

Read existing .kiro/steering/memory.md. Don't add entries that are already captured.

Phase 3: Append

Add new entries to the appropriate section. Each entry is one line with a date stamp:

# Project Memory

## Bug Patterns
- [2026-02-25] loadSession() must set renderedHtml for assistant messages — template uses [innerHTML] not {{ content }}
- [2026-02-25] Lambda test invocations: payload is raw JSON, NOT wrapped in {"body": "..."} — .NET service sends direct

## API Contracts
- [2026-02-25] AgentCore Memory list_events returns events in reverse chronological order
- [2026-02-25] Bedrock reasoningConfig uses "ENABLED" not "enabled" — case-sensitive enum

## Stakeholder Preferences
- [2026-02-25] No image generation in AI chat
- [2026-02-25] No social media content generation in AI chat

## SDK Gotchas
- [2026-02-25] MemoryClient.list_events max_results default is 100, not unlimited

## Recurring Review Findings
- [2026-02-25] DRY violation in LambdaChatService Lambda invoke pattern — deferred, track for refactor

## Workflow Learnings
- [2026-02-25] Always verify deployed Lambda package contents, not just config timestamps

Phase 4: Confirm

Print: "Updated .kiro/steering/memory.md with N new entries."

Rules

• One line per learning. Keep it scannable.
• Include the date for temporal context.
• Don't duplicate — check before appending.
• Don't editorialize — state facts, not opinions.
• This file is loaded every session via steering — keep it under 200 lines. Archive old entries to .kiro/context/memory-archive.md if it grows too large.
• Refer to the user as "The Brougham 22".