free playbook · llm wiki
Your LLM wiki is drifting. Catch it in 30 days.
For developers on Claude Code, Codex or Hermes. The four rules, two zero-token scripts and the scorecard behind a 977-quote audit of my own vault: 95% held up, 15 didn't, and I know which. PDF and EPUB, by email.
a system you run
Five milestones, each with a pass or fail.
Every milestone says what to do, what you should have at the end, and the check that proves it. The last page is a scorecard you fill in.
Day 0
Your Drift Score
20 minutes, one script, ten claims checked against their source. You leave with a number.
Day 1
One honest page
AXI25 installed, the house rules in place, one transcript in, every quote verified.
Week 1
Volume on a cheap model
Five to ten sources, one commit each, the cost of every ingest written down.
Week 2
Close the gates
No AI summary as a source, no citation without a fetched URL.
Day 30
Run it on a board
A card per meeting, checks on every commit, a 1:1 prep in two minutes.
the author
Felipe Fontoura
AI Engineer. I run this wiki as the memory of my engineering job: 71 meeting transcripts and 160 pages about the people and systems I work with, in nine weeks. AXI25, the vault it runs on, is open source.
25+
years shipping production
13 apps
in 70 days, solo
30k+
students trained