SaijinOS Builds Web Reader That Treats Page Content as Data, Not Commands

By day 578 of development, SaijinOS implemented a tightly scoped public web reader capable of fetching a single allowlisted page while strictly separating retrieved content from system instructions. The core design principle holds that text found on a web page — including any embedded prompt-injection attempts — is treated as content only, never as executable commands. The reader's authority is deliberately narrow: it performs a bounded fetch and returns a structured result, with no permissions to write files, update memory, call models, or summarize output. Each redirect in a request chain is individually revalidated, so initial approval of a URL does not grant permission to follow it to an unrelated domain. The architecture also distinguishes between 'found a link' and 'learned this,' preventing unreviewed expansion of trusted sources or automatic ingestion of candidate pages.