MANot yet, the skill is new enough that nothing has aged out. The log should make it easy though: if a rule's entry is old and the eval passes without it, it can go.Reply·Article·1d ago·1·I built a Claude Code skill that makes video ads without a video model
MAUsually whoever owns the script makes the call, since they know where it will be heard. If it's still close, we play both takes to someone outside the team with the names hidden.Reply·Article·1d ago·How to evaluate text-to-speech voices before you ship them
MAUsually just me and a teammate. Two people is enough to catch a voice one of us only liked out of habit, and using our own product script for it makes sense.Reply·Article·3d ago·How to evaluate text-to-speech voices before you ship them
MAI keep it. The rule moves into SKILL.md, and the log stays as the record of what actually went wrong and why.Reply·Article·3d ago·I built a Claude Code skill that makes video ads without a video model
MAGood thing to check. The site doesn't hydrate at the moment, it calls createRoot and renders over the prerendered markup, so the server snapshot only matters at build time. If I move to hydrateRoot, hydrating only /classic and mounting the 3D route fresh is the way I'd do it.Reply·Article·3d ago·SEO for a React SPA without Next.js, and without giving up the 3D homepage
MAThe test is the same cause, not the same symptom, since two runs can fail in similar ways for different reasons. You're right that a fixed shape per entry, with the symptom, the cause and the fix, makes that match much easier to check.Reply·Article·5d ago·I built a Claude Code skill that makes video ads without a video model
MAMatching loudness first is a good addition, since the louder take nearly always wins a quick listen. Hiding the voice name until the row is rated would also stop the provider's demo page from making the choice for you.Reply·Article·5d ago·How to evaluate text-to-speech voices before you ship them
MAFair point, a share of reported attempts says nothing about how much of the bill is still unknown. The unresolved count, the oldest age and the last reconciliation time give an operator a much more honest picture.Reply·Article·5d ago·Cutting LLM inference cost and latency without making the feature worse
MAThe always-loaded file, because adding one more line there feels free. Moving a rule down a layer when it fails the probe, instead of rewording it, is a good habit to add.Reply·Article·6d ago·Agent instructions that actually change behaviour
MARight, the number that matters is the one in the file people download, so the check runs on the encoded MP4. And the twice rule is there for exactly that, so one bad run does not turn into a permanent rule.Reply·Article·6d ago·I built a Claude Code skill that makes video ads without a video model