dif.sh puts feature flags in Markdown

dif.sh has a sharp premise: rollout logic and experimentation knowledge should live beside the code that uses them. The model is compelling for agent-driven development, though teams still need a reliable analytics pipeline to turn local experiment definitions into trustworthy decisions.

• –Each file captures flags, audiences, variants, hypotheses, metrics, rationale, and conclusions in a reviewable format.
• –`dif build` generates a typed client and `context.json`, while conflict detection and exclusion groups can fail CI before incompatible tests reach production.
• –Local evaluation avoids a runtime dependency on a hosted control plane, keeping the production client small and predictable.
• –The no-signup CLI makes the workflow easy for coding agents to install and scaffold autonomously.
• –Dif Cloud adds dashboards and confidence intervals, but the repository remains the canonical source of truth.