Claude Code skill: a few prompt-quality tweaks to skills/daisyui/SKILL.md

[longrunningprocess]

I use the daisyUI Claude Code plugin and ran a prompt audit of its skill, with Claude Opus 5.5 as the reading model. Line numbers are against master at 9adbeaa (skills/daisyui/SKILL.md), which is identical to the installed 5.7.0. The component guides came out clean, and they read as tight contracts. The ideas below are all about the top-level SKILL.md.

1. Line 28 vs line 107 disagree. Line 28: "Read more than one candidate guide before you select a component." Line 107: "If the choice is not clear, read the guides for the candidate components…" One says always, the other says only when unclear, so the model has to guess which one applies. Could line 28 drop its second sentence and leave the rule to line 107?

2. The trigger text is repeated in the body (lines 12-18). "When to use this skill … Even if the user does not request this skill, use it." repeats the frontmatter description. The description is what makes the skill trigger, and calibrated urgency is fine there. Once the body has loaded, the skill has already triggered, so this section adds nothing.

3. The "Component discovery protocol" (lines 101-112) is a six-step script for a judgment call, followed by "You must match the meaning… Always examine the intended function." Current models follow strong wording very literally, and a fixed procedure for choosing a component tends to make them over-read guides. The idea fits in one or two sentences: "Choose components by intended function and behavior, not by name. When more than one could fit, read their guides and compare before you choose."

4. "You must read" appears three times in the references table (lines 20-28). Plain "Read X before using Y" would carry the same weight without the escalation.

Happy to open a PR if any of these sound right.