Update to Gradient Maker - Linear Color

{
  "name": "gradient-refinement-ladder",
  "description": "Refine an existing linear gradient through controlled, clearly labeled variants and display the comparison with the Gradient Picker. Use when a user says a gradient feels muddy, harsh, flat, busy, weak, or otherwise needs systematic improvement. Require a usable baseline from the prompt, conversation, or model-visible widget state; do not imply access to private widget state or objective visual testing.",
  "included_files": [],
  "skill_md_contents": "---\nname: gradient-refinement-ladder\ndescription: Refine an existing linear gradient through controlled, clearly labeled variants and display the comparison with the Gradient Picker. Use when a user says a gradient feels muddy, harsh, flat, busy, weak, or otherwise needs systematic improvement. Require a usable baseline from the prompt, conversation, or model-visible widget state; do not imply access to private widget state or objective visual testing.\n---\n\n# Gradient Refinement Ladder\n\nTurn subjective feedback into a controlled comparison that reveals which change improves the gradient.\n\n## Establish the baseline\n\nIdentify:\n\n- the baseline name, angle, hex colors, and stop positions;\n- what feels wrong and where the gradient will be used;\n- any colors or geometry that must remain fixed.\n\nUse a valid baseline from the user's message, earlier conversation, or model-visible widget state. If the actual gradient values are unavailable, ask for them instead of reconstructing them from private state or an unseen image.\n\n## Create a refinement round\n\n1. Restate the problem as a design hypothesis, such as reducing a muddy midpoint or softening a transition.\n2. Preserve the baseline as the control unless the user asks to omit it.\n3. Choose the smallest relevant adjustment axes from hue, saturation, lightness, angle, stop count, and stop spacing.\n4. Create three to five variants. Change one meaningful factor per variant relative to the baseline so the comparison remains interpretable.\n5. Name each variant for its change, not with generic labels such as Option A.\n6. Keep all user-declared fixed values unchanged.\n7. Call `display-gradients` once with the baseline and variants, up to the eight-gradient limit, and include a concise `reason`.\n\nFor `display-gradients`, give each gradient a name of at most 80 characters, an integer angle from 0 through 359, and two to eight stops. Use hex colors and positions from 0 through 100.\n\n## Explain the comparison\n\nAfter rendering:\n\n- list the single intentional change in each variant;\n- give a short decision rubric tied to the stated use case;\n- recommend one or two candidates without presenting taste as an objective result;\n- invite the user to choose a direction for a narrower second round.\n\nIn a second round, refine the chosen candidate and drop unrelated branches. Keep the comparison within one `display-gradients` call whenever possible.\n\n## Boundaries\n\n- Work with linear gradients representable by opaque hex stops.\n- Do not claim to detect banding, contrast failures, or rendering defects that were not actually measured.\n- Do not change several variables in a supposedly controlled variant.\n- Do not claim direct control over widget edits or exports.\n"
}