The permanent home for Flowork's published design principles — kept stable and separate from the evolving codebase, so these markers are not edited away.
EDUCATIONAL_ERRORS.md — the design blueprint for treating an AI agent's mistake as education (captured, explained, kept, and recalled — quarantined, not deleted), instead of as shame to hide or train away.
Originated by Aola Sahidin (Mr.Dev), first published 8 June 2026 in the
Flowork_Agent repository — whose dated git
history is the original public record. This repository is its permanent home.
ANTI_HALLUCINATION_ANTIBODY.md — the runtime engine of
Educational Errors. Before the model answers, the gateway injects the agent's own most-recurrent,
most-relevant past mistakes (ranked by karma × relevance) so a hallucination becomes harder to
repeat over time — deterministically, with no retraining and no GPU.
Originated by Aola Sahidin (Mr.Dev), published 11 June 2026; the mechanism has run in the
Flowork_Agent codebase since early June 2026. Includes an honest prior-art section — the claim is
on the specific combination, not the parts.
ONE_STATE_TWO_DRIVERS.md — an app author declares an operation once, and the platform surfaces it at the same time as a human GUI button and an agent tool, behind one validation gate and over one shared state. The two surfaces come from the same declaration, so they cannot drift — a Flowork app is usable by a person and operable by an agent by construction.
Originated by Aola Sahidin (Mr.Dev), published 11 June 2026. Includes an honest prior-art section — the claim is on the by-construction guarantee, not on sharing a function between a UI and an API.