You are generating evaluation queries to test whether a Claude Code skill's description triggers correctly.

Skill name: {skill_name}
Skill description: {description}
Skill content:
{skill_content}

Create 20 eval queries — a mix of should-trigger and should-not-trigger.

The queries must be realistic and something a Claude Code or Claude.ai user would actually type. Not abstract requests, but requests that are concrete and specific and have a good amount of detail. For instance, file paths, personal context about the user's job or situation, column names and values, company names, URLs. A little bit of backstory. Some might be in lowercase or contain abbreviations or typos or casual speech. Use a mix of different lengths, and focus on edge cases rather than making them clear-cut.

should-trigger (8-10): cover different phrasings — formal/casual; cases where the user does NOT name the skill or file type but clearly needs it; cold/unusual use cases; cases that compete with other skills but this one should win.

should-not-trigger (8-10): the most valuable are near-misses — queries that share keywords or concepts with the skill but actually need something else. Adjacent domains, vague phrasings that naive keyword-matching would falsely trigger, scenarios that touch the skill's function but another tool fits better. Negatives must be genuinely tricky — NOT obviously unrelated.

Also assign each query a short "topic" tag (a few words grouping similar queries).

Respond with ONLY a JSON array, no other text:
[{"query": "...", "should_trigger": true, "topic": "..."}, ...]
