Architecture Decision Records
How this system is built and why. Every decision that changes the architecture gets written down. Generated with the adr-skill for Claude Code and Pi.
npx skills add skillrecordings/adr-skill --skill adr-skill --yes --global- 0005
AIAA OS — TurboClaw integration for per-student OpenClaw deployment
AIAA teaches agency owners to use AI agents, but students struggle with deployment infrastructure. Three existing repos contain pieces of the solution: TurboClaw (per-user deployment), Agentcy-OS (…
proposed - 0004
ACP runtime for coding agents
Coding agents (Codex, Claude Code) were managed via tmux sessions and PTY hacks — fragile, hard to monitor, and no clean lifecycle management. OpenClaw introduced ACP (Agent Client Protocol) as a f…
accepted - 0003
Five-layer memory architecture
Previous memory systems (Kybernesis, Supermemory, QMD, a 7-layer stack) added complexity without payoff. Memory retrieval was unreliable, indexing broke frequently, and multiple truth sources cause…
accepted - 0002
Five-agent team structure
The previous setup had 15+ scattered agents with overlapping roles, unclear ownership, and expensive model usage. Agents were being spawned as disposable subagents, losing context between runs. We …
accepted - 0001
Adopt architecture decision records
Architecture decisions in this project are made implicitly — through code, conversations, and tribal knowledge. When a new contributor (human or AI agent) joins the codebase, there is no record of …
accepted