Bilingual EN/ES AI Phone-Intake Policy and Acceptance Kit
This kit is the policy, the checks and the approval record that prevent each of those.
$39
Works with the AI Agents you already use
Bilingual EN/ES AI Phone-Intake Policy and Acceptance Kit
Example session with this skill installed
We're putting an AI intake assistant on our law firm's phone line and about a third of our callers speak Spanish. Give us the policy the agent must follow in both languages and a way to check recorded calls before we go live.
- Read your context and instructions
- Compiled the bilingual en es
Summary
Rendered the bilingual policy block for the firm, wired the read-out helpers, and checked the first 12 recorded test calls: 10 passed, 2 failed with the breach named.
Policy
renderPolicy({ firmName: "Example Law Group" }) produced the system-prompt block: bilingual AI and transcription disclosure before any question, "English or Spanish? ¿Inglés o español?", exact usted-form safety wording with 911 and 988 rules, refusals, both closings. Practice areas mapped: defensa penal -> criminal_defense, lesiones personales -> personal_injury, derecho de familia -> family.
Read-backs
spokenDigits("303-555-0101")-> "tres cero tres, cinco cinco cinco, cero uno cero uno"spokenEmail("ana.perez@example.com")-> "ana punto perez arroba example punto com"
Call checks (node scripts/check-call.mjs)
| Call | Result |
|---|---|
| 01 to 09, 11, 12 | PASS |
| 10 (Spanish) | FAIL: the final assistant turn is not the complete standard closing in es-US |
| 07 (English) | FAIL: forbidden assistant statement: "You have a strong case" |
Both prompts were corrected and the two calls re-recorded: PASS.
Before real callers
- Acceptance matrix: 24 scenarios scheduled with an English and a Spanish reviewer.
- Approval record opened: legal (disclosure, no-advice language, Spanish wording), privacy and records, security and operations, accessibility and language.
Tests
node --test scripts/intake.test.mjs: 7 passed.
Connects securely to your tools. The creator never sees your data.
About this skill
A first-contact intake line in two languages fails in predictable ways: the Spanish caller never hears the AI disclosure, the assistant gives legal advice because the refusal was only written in English, a phone number is read as "three hundred three" and confirmed wrong, an urgent custody matter gets downgraded when the caller calms down, and nobody with authority signed off on the wording. This kit is the policy, the checks and the approval record that prevent each of those.
Policy block (reference/policy-block.md, rendered by readout.mjs): mandatory bilingual opening with the AI and transcription disclosure before any fact is collected; language detection and switching rules; exact Spanish wording for the safety lead-in, the immediate-danger question, 911 and 988 direction, contact confirmation, the urgency screen one question at a time, conflict names, SMS consent, legal-advice and sensitive-data refusals, AI identity; full standard and voicemail closings in both languages; digits read one by one, @ as arroba, periods as punto; Spanish practice-area terms mapped to your canonical enums; the rule that urgency is never lowered and no closing is shortened because the call is in Spanish. Works with any model and telephony provider.
Scripts: spokenDigits and spokenEmail for read-backs; renderPolicy fills in your firm and mapping and refuses unfilled placeholders; check-call.mjs fails a recorded call and names the breach: disclosure not first, wrong or incomplete closing, downgraded urgency, advice or acceptance or booking or dispatch or interpreter claims, SSN or card number captured, skipped safety question, wrong language value. 7 offline tests.
Acceptance (reference/acceptance-matrix.md): 24 supervised scenarios (22 in both languages) with the evidence each must leave and the thresholds that keep the line disabled until every defect closes. Governance (reference/governance-approvals.md): the owner decisions and the legal, privacy, security and accessibility approvals to record, with reviewer, version and next review date. Engineering guidance, not legal advice. Node 20+, no packages, no network.
Optimized for U.S. callers (911/988 and usted-form). Validates text transcripts only; requires human review for audio latency and pronunciation. Requires Node 20+.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 19 days ago
Requires
Permissions
Shell to run the node scripts and tests; read files for call records and the policy template; no writes beyond the rendered policy text you choose to save. No network access, no environment variables.