Memory Kit
Free for noncommercial use under PolyForm Noncommercial 1.0.0; commercial use needs a licence from Rico. View the source
It installs a small memory folder the AI reads at the start of every chat and writes to at the end. Three plain, dated, append-only sources:

The problem
Chat assistants forget everything between conversations. An owner who settled a price, a policy or a territory rule last week has to re-explain it, and nothing warns them when a new idea quietly contradicts a call they already made.
How it works
The memory behavior is Markdown: a skill (skills/memory/SKILL.md) with a startup ritual, a wrap ritual and four workflows. A self-adapting installer.md detects what the host AI can do and picks an install path, from a real filesystem install to a Project/Gem source list to a paste-into-chat fallback. On filesystem-capable agents, a plain-Node helper (scripts/install-files.mjs) owns install and upgrade file operations so collision refusal, backups and preservation of existing memory are deterministic rather than improvised.

Features
- Contradiction check
- Named refusals
- Day-dump wrap
- Monthly brief
- Portable MEMORY BRIEF block
- Honest degradation
- Pasted text is data
How it was built
Version 1.0 shipped on 2026-07-15 as a Markdown skill that gives a chat assistant memory: a startup ritual that reads a memory folder, a wrap ritual that writes dated notes and a self-adapting installer.
A GPT-family review and a multi-agent swarm reviewed it before release, and the coding-agent install path was verified in a clean-room sandbox for Codex and Claude Code.
A quality panel on 2026-08-11 called it a filing convention, not a product. Version 1.0.1 added a contradiction check that quotes the owner's own dated reasoning when a new decision argues with an old one, a monthly brief and named refusals.
Three rounds of simulated buyers found twelve defects. One was that the wrap assumed the work happened in chat, so a day-dump wrap was added for a contractor who sends the whole day in one evening message.
AI agents built and tested the kit under Rico's direction. The assistant never overrules the owner and never invents a prior decision, and Rico placed the kit as a free starting point the other kits build on.