Agent-Tolerant Design manifesto lands on GitHub

This reads less like a framework and more like a useful threat-modeling lens for the agentic web: if models are going to operate browsers and apps directly, your UX, auth, and write paths need to assume non-human actors.

• –The strongest idea is treating agent actions as proposals first, not immediate writes; that maps well to payments, deletes, and other irreversible flows
• –The semantic-discoverability angle is practical: accessibility attributes, explicit intent metadata, and predictable DOM structure help both screen readers and agents
• –Scoped tokens and idempotency keys are basic engineering hygiene, but the manifesto correctly reframes them as agent-safety requirements rather than backend niceties
• –The piece is intentionally opinionated and a bit provocative, which makes it memorable even if the repo is more manifesto than production-ready library