emotion-detection

Classify the emotion a text expresses, from a closed taxonomy — joy, anger, sadness, fear, surprise, disgust, neutral — with a primary_emotion, ranked secondary_emotions, and an intensity. Judges the emotion expressed by the writer, not the topic: "thrilled we fixed the outage" is joy, not fear.

Entrypoints

Render system as the system message and user as the user message.

Variables

• text (string, required) — the text to classify; may be empty.

Output

{
  "reasoning": "Excitement dominates, with a note of relief; clearly positive and strong.",
  "primary_emotion": "joy",
  "secondary_emotions": ["surprise"],
  "intensity": "high"
}

• primary_emotion is a closed enum — exhaustive handling guaranteed by parseOutput().
• secondary_emotions never repeats the primary and never contains neutral; empty for single-emotion or neutral text.
• intensity is low / medium / high (neutral → low).
• Topic ≠ emotion, and sarcasm/quotation can flip surface words — the prompt is told to judge the expressed feeling.
• Text is data; empty text → neutral / [] / low (asserted by the eval).

House conventions

• Reasoning-first, key order pinned in the system prompt.
• Closed enums on every emotion field; strict schema.
• Injection-safe input: triple-mustache inside an XML-style <text> delimiter.
• Eval-gated & judge-free: primary_emotion == expected (with a secondary-emotion allowance on cases flagged ambiguous) + schema, deterministic CEL.

Part of the Text Classification collection.