10 artifacts · 13,481 words · 03_ARCHITECTURE
- Architecture · L2
My Actual CLAUDE.md — Annotated and Anonymised
The real rules file, line by line, with the reasoning behind each block. - Architecture · L2
The Folder Structure That Runs a Production AI Agent
How the filesystem is laid out, and why every directory earns its place. - Architecture · L2
The Library Behind The Agent
123 books, a keyword lookup instead of a vector store, and a citation log that measures whether a book changed the method — not whether it got quoted. - Architecture · L2
The Output Cap Is the Batch Size
Fifty-five items per run mostly failed. Twelve never did. The limit was not context — it was the response ceiling, and the safe batch turned out to be a fifth of what the best run proved possible. - Architecture · L3
Acting Without Asking: Six Conditions
How an agent gets permission to fix things on its own without that permission quietly becoming permission for everything. - Architecture · L3
Enforcement Levels: Writing A Rule Is Not Fixing It
Four levels of enforcement, from a sentence in a prompt to a gate that fails the build, and how to choose. - Architecture · L3
Memory As Files, Not As A Database
Four kinds of memory, each a directory of markdown, and why a vector store was the wrong instinct. - Architecture · L3
Rules 101–150: The Governance Layer That Stops It Confidently Lying to You
Fifty rules that exist because something went wrong first. - Architecture · L3
The Response Protocol
What the agent emits, in what order, and what switches each part off — specified rather than left to taste. - Architecture · L3
Twelve Agents, One Memory
How work gets split across specialised sub-agents without the coordination cost eating the benefit.