Ask JSON via claude-wire

Use me when you need structured results. If you're about to call Agent / Task and then parse JSON out of free-form text -- stop. Use this skill instead. You get schema-validated data back, not prose. Uses sonnet by default for reliable schema adherence, one round-trip, no regex.

When to Use Me

Reach for this skill whenever downstream logic needs a structured payload:

• Classify into an enum ("bug" | "feature" | "chore")
• Extract an array of records (entities, TODOs, imports, etc.)
• Triage / rank a list with scores
• Routing decisions ({ route: "refund" | "support", confidence: 0.9 })
• Any subagent call whose output you were about to regex out of prose

If your next thought after an Agent call is "now parse the JSON from the response text" -- that's the signal. Use me.

ALWAYS pass --model sonnet when invoking the CLI from this skill. The CLI's own default is haiku for CI/script users who know their workload, but sonnet adheres to JSON schemas far more reliably on ad-hoc prompts. Haiku's --json-schema compliance is best-effort; it frequently returns prose-wrapped JSON or pure prose for short "classify X" style prompts, producing confusing exit-1 validation errors. Skip the thrash -- start at sonnet.

When NOT to Use