secondBrain, Part Two: The Nervous System and the Immune System
secondBrain Part Two by Manazir Ali. How a vectorless, Markdown personal LLM knowledge base became always-on and self-maintaining, then hardened against the agentic threat model: machine-enforced immutability, an append-only-log guard, the lethal trifecta cut by removing the exfiltration channel, least-privilege read-only librarian and write-scribe agents, and indirect-prompt-injection defenses. A 16-agent security rubric moved the system from a 5.1 mean into the 8-to-9 range, and a live session then found and fixed two real bugs. The thesis: the same plain-files discipline that made retrieval deterministic makes the autonomy auditable and the security real. 
