Legal Prompt Library: 30 Tested Prompts for Brazilian Court Publications

A legal prompt library gives lawyers and legal operations teams a repeatable way to turn unstructured court publications into useful information. Instead of asking an AI tool to simply “summarize this,” a well-designed prompt defines the fields, legal distinctions, evidence requirements, and review rules that matter for the workflow.

The first problem is volume. A single publication may contain a case reference, procedural history, a decision, a deadline, a party obligation, and administrative information in the same block of text. A generic summary can be readable while still omitting the one sentence that starts a response period.

The second problem is ambiguity. Expressions such as “intimate-se,” “manifeste-se,” “junte-se,” and “dê-se ciência” do not all create the same operational consequence. The prompt must ask the model to classify the event, identify its recipient, preserve the source wording, and state when human confirmation is required.

Brazilian deadlines also require legal context. A prompt should not calculate a final date from a publication alone unless the workflow has reliable information about the applicable rule, the date of official publication, local holidays, suspension periods, and the type of procedural period. The Brazilian Code of Civil Procedure on Planalto is a useful primary reference for teams designing their validation rules.

The goal is not to replace legal judgment. The goal is to reduce repetitive reading, make relevant facts visible sooner, and give a lawyer a structured starting point that can be checked against the original publication.

The prompts below are designed as reusable templates. Replace the fields in brackets with your own variables, and preserve the instruction to quote the supporting passage. The examples assume Brazilian court publications and should be reviewed against the applicable procedural context.

1. Publication summary Prompt: “Summarize the publication in five bullets. Identify the court, case number, parties, procedural event, decision, and next action. Do not add facts that are not present. Quote the sentence supporting the next action.”

2. Executive summary for a corporate legal team Prompt: “Create a plain-English summary for a legal manager. Include case status, business impact, required action, responsible team, deadline information, and open questions. Separate confirmed facts from interpretation.”

3. Lawyer-facing technical summary Prompt: “Summarize this publication for counsel using procedural terminology. Preserve the legal basis cited by the court, identify the ruling’s practical effect, and distinguish findings from orders. End with a review checklist.”

4. Deadline extraction Prompt: “Extract every explicit deadline. For each one, return the exact expression, recipient, required action, trigger described in the text, and source excerpt. If no deadline is explicit, return an empty list.”

5. Deadline candidate detection Prompt: “Find language that may create a procedural deadline, including ‘manifestar-se,’ ‘contestar,’ ‘recorrer,’ ‘juntar,’ ‘pagar,’ and equivalent expressions. Classify each item as explicit deadline, possible deadline, or no deadline, and explain why.”

6. Deadline calculation guardrail Prompt: “Calculate a deadline only when the publication supplies a valid trigger date, period, unit, and responsible party. If any element is missing, set calculated_date to null, list the missing element, and mark human_review as true.”

7. Business-day versus calendar-day classification Prompt: “Determine whether the stated period is expressed in business days, calendar days, months, or another unit. Do not infer the unit from custom or memory. If the text is ambiguous, return ‘ambiguous’ and quote the relevant passage.”

8. Multiple deadlines in one publication Prompt: “Create one record for each deadline in this publication. Do not merge deadlines with different recipients or actions. Sort the records by operational urgency and retain the original order as a separate field.”

9. Appeal-related deadline check Prompt: “Identify whether the publication communicates a decision that may require analysis of an appeal or response. Do not state that an appeal is available unless the text supports that conclusion. Return decision_type, possible_action, stated_period, and review_questions.”

10. Payment or compliance deadline Prompt: “Check whether the publication orders payment, compliance, document production, or another concrete act. Identify the obligated party, amount or object, deadline wording, consequences mentioned, and evidence needed to confirm completion.”

11. Urgent action detector Prompt: “Flag urgent actions involving a deadline, injunction, seizure, hearing, asset restriction, release order, suspension, service failure, or immediate compliance. Assign high urgency only when the text supports it, and quote the evidence.”

12. Hearing identification Prompt: “Determine whether a hearing, session, mediation, inspection, or expert appointment is mentioned. Extract date, time, location or virtual access information, participants, preparation tasks, and missing logistics.”

13. Injunction or provisional relief detector Prompt: “Identify any order granting, denying, maintaining, modifying, or requesting analysis of provisional relief. State the practical obligation, affected party, effective period if stated, penalty if stated, and review trigger.”

14. Administrative notice filter Prompt: “Classify the publication as substantive decision, procedural order, administrative notice, service confirmation, docket update, or unclear. Explain the classification using the publication’s wording and identify whether a lawyer action is required.”

15. Interlocutory decision versus administrative order Prompt: “Distinguish a decision that resolves a procedural or substantive issue from an administrative act that only records, forwards, certifies, or organizes information. Return classification, legal effect, action required, and supporting excerpt. Use ‘unclear’ when the text does not permit a safe distinction.”

16. Judgment detector Prompt: “Check whether the publication contains a judgment or final ruling. Extract the outcome, relief granted or denied, monetary amounts if stated, party obligations, appeal information stated in the text, and next operational step.”

17. Order to amend a filing Prompt: “Identify whether the court ordered an amendment, correction, supplement, or regularization of a filing. Extract what must be corrected, who must act, the stated period, required attachments, and consequences of noncompliance.”

18. Document request extractor Prompt: “List every document or piece of evidence requested by the court. For each item, identify the responsible party, deadline, format or authentication requirement, and whether the request is explicit or inferred.”

19. Expert evidence workflow Prompt: “Identify orders involving expert evidence, expert fees, questions for the expert, appointment, challenge, or report deadlines. Create a task list for counsel and flag any date or responsibility that requires confirmation.”

20. Party and counsel identification Prompt: “Extract the parties, counsel, law firm, and any recipient named in the publication. Preserve spelling exactly as written, distinguish parties from representatives, and flag inconsistent names against the case metadata.”

21. Case-number and instance validation Prompt: “Check the case number, court, chamber, jurisdiction, and instance against the supplied case metadata. Return matching fields, conflicting fields, and fields absent from the publication. Do not correct a value without evidence.”

22. Duplicate publication detection Prompt: “Compare this publication with the supplied recent publications. Determine whether it is a duplicate, a correction, a continuation, or a new event. Explain the comparison using date, case number, event wording, and document identifiers.”

23. Contradictory information detector Prompt: “Find contradictions involving dates, parties, deadlines, amounts, decisions, or procedural status. Quote both conflicting passages, describe the operational risk, and mark human_review as true.”

24. Plain-language client summary Prompt: “Rewrite the confirmed content for a client with no legal training. Explain what happened, what it means, what must happen next, who owns the task, and what remains uncertain. Do not provide legal advice beyond the publication.”

25. Internal task creation Prompt: “Convert the publication into actionable tasks. Each task must include title, description, owner_role, due_date if confirmed, priority, dependencies, evidence_required, and source_excerpt. Never create a due date from an unsupported assumption.”

26. Email draft for responsible counsel Prompt: “Draft a concise internal email about this publication. Include case identifier, event, required action, confirmed deadline, uncertainty, and requested confirmation. Use a neutral tone and do not imply that a date is final when review is required.”

27. Calendar event draft Prompt: “Create a calendar event only for a confirmed hearing, meeting, inspection, or explicit deadline. Return title, start date, end date if known, timezone, location, attendees, preparation notes, and source excerpt. If the date is uncertain, return no event and explain why.”

28. Risk scoring Prompt: “Score operational risk from low, medium, high, or critical using only deadline proximity, irreversible consequences, financial exposure, service or hearing information, and uncertainty. List the factors used and do not treat uncertainty as proof of high legal merit.”

29. Human review decision Prompt: “Set human_review to true when the text is ambiguous, deadlines conflict, a calculation depends on external calendars, a decision may require legal interpretation, OCR quality is poor, or a high-impact order is detected. Explain the exact review reason.”

30. Structured final response Prompt: “Return valid JSON with these fields: publication_summary, classification, case_number, court, parties, actions, deadlines, urgent_flags, source_excerpts, uncertainties, confidence, and human_review. Use null for unavailable scalar values and [] for unavailable lists. Do not include commentary outside the JSON.”

A compact example shows the intended behavior. Input: “Intime-se a parte autora para, no prazo de 15 dias, manifestar-se sobre a contestação.” Output: “action: respond to the defense; responsible_party: plaintiff; stated_period: 15 days; trigger: service or publication event, as confirmed by the workflow; calculated_date: null; urgency: medium; human_review: true.” The model should not invent the final date from that sentence alone.

The best prompt is not necessarily the longest one. It is the prompt that produces a consistent result for a defined decision, exposes uncertainty, and fits the way the legal team already assigns work.

A small office might use three flows: a daily summary, a deadline detector, and a human-review queue. A corporate legal department may add business impact, outside counsel ownership, contract references, and escalation rules for injunctions or hearings.

The review screen should make comparison easy. Show the original publication, extracted fields, source excerpts, confidence, and reviewer controls in one place. Comments attached to the publication are more useful than scattered emails because they preserve why a deadline was accepted, changed, or rejected.

Cost control comes from routing. Run a short classification prompt on every new item, use a richer prompt only when a deadline or urgent signal appears, and avoid sending duplicate publications through the full flow. Estimate cost using the number of monitored names and terms, publication volume, average text size, retry rate, and model pricing rather than relying on a single average.

A useful governance report includes total publications processed, percentage summarized automatically, deadline precision, urgent-action recall, human-review rate, invalid-output rate, average processing time, and cost per monitored term. These measures tell leadership whether the workflow is reducing manual checks without hiding legal risk.