structured-summarization

Published

Public

Condense text into a structured JSON summary — title, key points, action items, and entities mentioned — instead of prose. Built for meeting notes, email threads, articles, and agent handoffs. Optional cap on the number of key points.

structured-summarization

Summaries your code can use. Instead of a paragraph of prose, this returns a structured object — title, key_points, action_items, and entities_mentioned — so you can render, filter, and route the pieces independently.

Built for meeting notes, email threads, support conversations, articles, and agent handoffs.

Entrypoints

Entrypoint	Renders	Inputs
`systemPrompt`	Instructions + embedded schema	optional `maxKeyPoints`
`userPrompt`	The delimited text to summarize	`text`

Variables

text (string, required) — the text to summarize.
maxKeyPoints (integer, optional) — cap the number of key points. Omit to let the model choose.

Output

{
  "title": "Q3 roadmap review and launch decision",
  "key_points": [
    "Beta feedback is positive; main gap is onboarding friction.",
    "Launch slips one week to finish the new setup flow."
  ],
  "action_items": [
    "Priya to finalize the onboarding flow by Friday.",
    "Marcus to draft the launch announcement."
  ],
  "entities_mentioned": ["Priya", "Marcus", "Q3 roadmap"]
}

key_points are ordered most-important-first.
action_items capture concrete tasks/commitments only — the prompt is told not to invent them. Empty array when the text has none.
Empty/low-content input → empty title and empty arrays (specified, not left to chance).

House conventions

Triple-mustache for text.
Grounded only in the input — no outside facts, opinions, or fabricated action items.
Strict schema — additionalProperties: false, fixed key order.

Notes

Verified by local render with and without maxKeyPoints. Suggested default claude-sonnet-4-6 at temperature: 0.

@sufleur/