Stet is built for coding agents and developers. Use project skills and validated plans with your agent, or attach annotations directly through the typed API.
- Add agent authoring tools alongside the typed runtime API with typed annotation plans, generated
schemas/capabilities, actionable standalone validation, local CLI, 30 framework
snippets, five persistent-control lifecycle patterns and four project-installable
Agent Skills. Existing runtime APIs are unchanged. - Recover interrupted skill installation through ownership journals and locks,
preserving local edits and reporting precise conflicts. - Add bounded read-only project discovery and a generated illustrative settings
plan. Missing arrow destinations no longer gate Vue/Svelte sample controls. - Add contract drift/malformed-input tests, rendered lifecycle and SSR/hydration
checks, supported Ubuntu WebKit verification, and packed CLI consumers on Node
20.0.0/24 for Linux and Windows. Browser size budgets and zero required runtime
dependencies are preserved. - Complete actual fresh-context routing and application trials, with retained
failures, focused review/arrow guidance corrections, fresh reruns and a separate
ordinary-docs baseline. Add pristine-layout comparisons and model-free artifact
replay. See docs/agent-evals.md for outcomes and limitations.
The README and stetkit.com now provide starting points for both workflows.
Plans and compiler checks do not establish live UI correctness. Source, interaction and screenshot review remain necessary; arrow paths do not automatically avoid intervening UI. See evaluation results for retained failures and recoveries.