Skill: fraimz

fraimz is the canonical proof artifact: a single fraimz.html (evals/results/<run-id>/fraimz.html) with one frame per step. Each frame binds a claim, the action the end user took, the assertion that witnesses the side effect, a voiceover that narrates the step, and a validated screenshot. It is the thing a human looks at (and listens to) to understand, at a glance, that an experience works.

This skill owns the prove-it loop. In the demo-driven journey (voiceover skill: script → worktree → build → fraimz → PR) this is how the build is verified; when orchestrating, the orchestrator drives this loop and delegates code repairs to the executor.

Every fraimz is a demo

A fraimz is never a bare test log. The flow declares which demo it is via kind in evals/flows/<id>.flow.mjs:

• kind: "user-facing", the end user is the protagonist; the frames walk a real journey through the UI. Default for any feature, fix, or UX change.
• kind: "internal", for changes with no visible surface (perf, storage swaps, invariants, refactors): the frames demonstrate the internal claim in a way a reviewer can still follow. Terminal/tooling demos may set requiresApp: false to run without CDP; their frames carry claims, assertions, and ctx.output command output instead of screenshots (see evals/flows/voiceover-first-dx.flow.mjs).

If you cannot say which of the two your fraimz is, you have not framed the experience yet.

Voiceover first