| name | memory-insights |
| description | Track recurring session friction and success patterns across memory (sessions/, patterns). Use periodically or when improving agent instructions—not every turn. |
Memory Insights
Aggregate signals from existing memory artifacts to suggest improvements to AGENTS.md, app AGENTS.md, or .ai/memory/patterns.md.
When to use
- User asks to improve agent reliability or onboarding
- Several
sessions/ files exist and you want themes
- After multiple related incidents (same confusion repeated)
Process
- Skim recent
.ai/memory/sessions/*.md (newest first; cap at ~5 files to limit context).
- Skim patterns.md “Contents” and any
@category: learning / bugfix lines (use memory-search / rg if large).
- Identify:
- Repeated friction (same class of mistake or missing doc)
- Repeated wins (what agents should keep doing)
- Output concrete recommendations:
- One bullet = one proposed change (which file, what sentence or pattern to add)
- Apply only with user approval—or run memory-sync to append approved bullets to
patterns.md.
Anti-patterns
- Do not duplicate full session logs into patterns (keep patterns short).
- Do not invent project facts; only generalize from existing session/pattern text.
See also