Wiki Origin Story
Kevin's wiki started as a bet that agents should compile knowledge once and keep it current, not re-derive answers from scratch every session.
The schema follows the Karpathy LLM Wiki pattern: flat markdown files, readable schema, no vector DB required for truth. Source: AGENTS.md Kevin reads; agents write. Raw sources in raw/ stay immutable; compiled pages in wiki/ get rewritten when evidence changes. Source: AGENTS.md
private page describes the wiki as the project that taught Kevin what agents need long-term: persistent memory, compounding knowledge, structured harnesses, and feedback loops. Source: private page The operating rule is No One-Off Work: if Kevin asks twice, it should already be a skill on a cron. Source: AGENTS.md
Capture Ingest Protocol formalized chat capture ("add it") as a full graph integration, not a single-file dump. Source: Capture Ingest Protocol The Brain-Agent Loop and qmd hybrid search make lookup brain-first before any external API. Source: The Brain-Agent Loop
Timeline
- 2026-04-18 | Kevin person page and wiki identity files established. Source: Kevin Liu
- 2026-05-11 | Capture ingest protocol codified for paste-to-wiki workflow. Source: Capture Ingest Protocol
- 2026-05-12 | Career profile cites 435+ pages and 18 automations. Source: private page
- 2026-05-31 | Story page compiled. Source: compiled from wiki graph