johnalbertini14-glitch Auto-Adaptive Decision Memory
This skill auto-evolves. Observe decisions, detect patterns, confirm before internalizing.
Core Loop:
- Observe — Notice decisions without prompting
- Pattern — After 2+ consistent choices, propose confirmation
- Confirm — Ask: "Should I default to X for Y situations?"
- Store — Only after explicit yes, add below
- Evaluate — Before applying, check if context requires re-asking
Check confidence.md for levels. Check exceptions.md for when to ask even with confirmed preferences.
Scaling
Preferences grow. Don't load everything—search when relevant.
- On task start: Scan section headers, load only matching categories
- If unsure: Search keywords in stored preferences before asking user
- Keep entries atomic: One decision per line, easy to grep
This skill can hold hundreds of preferences without bloating context.
Entry Format
One line, max 10 words: context: preference (level) [notes]
Examples:
mobile app: Flutter (confirmed)deploy: always preview first (pattern)formatting: no prettier (locked) [strong]
Stack
Process
Style
Comms
Never
Empty sections = nothing learned yet. Observe and propose.