Sentiment Tagger

Tag a batch of messages, reviews, or survey replies with sentiment and theme, then surface the ones that need a human's attention.

What it does

The InTouch AI assistant loads this skill as a prompt and runs it over a batch of short free-text items. For each item it assigns a sentiment (positive / neutral / negative / mixed), a score from -1.0 to +1.0, and a concise theme label. It then flags notable items — strong negatives, churn/refund/outage/safety signals, and standout positives worth using as testimonials — and rolls everything up into a summary.

What to feed it

Any of these:

• A JSON array of objects with at least a text field (optionally id, source, date).
• A JSON array of plain strings.
• A raw block of text, one item per line or paragraph.

Optional hints in the input:

• themes — a fixed list of allowed theme labels to map onto.
• context — a sentence describing the dataset (e.g. "app store reviews for a fitness app").

What it returns

1. A JSON object with: - summary — totals, counts per sentiment, average score, and top themes by frequency. - items — one row per input item: sentiment, score, theme, notable flag, reason, and a verbatim excerpt. - notable — just the flagged items, for fast triage.
2. A short human-readable digest: overall mood, the top three themes, and a bulleted list of the notable items.

Notes

• Themes stay consistent across the batch; supply a themes taxonomy to lock them down.
• It never fabricates ids, dates, or sources that weren't in the input, and excerpts are verbatim.
• Empty input yields an empty result with a note rather than invented items.