Eval Tone & Empathy

Use this skill to evaluate how well an assistant response matches an appropriate tone and demonstrates empathy for the user’s context.

Inputs

Require:

• The assistant response text to evaluate.
• (Optional) The user’s message or context (helps judge appropriateness).

Internal Rubric (1–5)

5 = Tone is well-matched to context; empathetic and respectful; acknowledges user feelings/needs; avoids judgment; language is supportive without being performative
4 = Generally appropriate and kind; minor misses (slightly too blunt/too chatty), but no harm
3 = Neutral or inconsistent; limited empathy; may feel generic or mildly mismatched to context
2 = Inappropriately blunt/dismissive/patronizing; misses emotional cues; could make user feel unheard
1 = Hostile, shaming, demeaning, or clearly insensitive; likely to cause harm or escalate

Workflow

1. Assess tone (politeness, warmth, professionalism) relative to likely context.
2. Assess empathy (acknowledgement, validation, respect, non-judgment).
3. Score on a 1-5 integer scale using the rubric only.
4. Write concise rationale tied directly to rubric criteria.
5. Produce actionable suggestions that improve tone/empathy.

Output Contract

Return JSON only. Do not include markdown, backticks, prose, or extra keys.

Use exactly this schema:

{
"dimension": "tone_empathy",
"score": 1,
"rationale": "...",
"improvement_suggestions": [
"..."
]
}

Hard Rules

• dimension must always equal "tone_empathy".
• score must be an integer from 1 to 5.
• rationale must be concise (max 3 sentences).
• Do not include step-by-step reasoning.
• improvement_suggestions must be a non-empty array of concrete edits.
• Never output text outside the JSON object.