BridgeCalle Active Listener

Use this skill when an elderly person, family member, or caregiver (with explicit recipient consent) wants an empathetic, active-listening phone check-in call to reduce loneliness and foster emotional connection between visits.

BridgeCalle flips the traditional voice-agent paradigm: instead of an AI that talks at the user, it configures CALL-E task instructions for prompt-guided active listening, encouraging the AI agent to speak sparingly with gentle verbal nods ("Mmhmm", "I hear you", "How lovely"). Note: Active-listening speaking ratios and turn behavior are prompt-guided system instruction directives rather than hard platform enforcement, subject to provider model adherence.

When to use

• Outbound phone check-ins for elderly loved ones or seniors living alone.
• Capturing nostalgic memories, daily reflections, and emotional state in a structured post-call summary.
• Exporting warm call summaries to family members via SMS or WhatsApp.

When not to use

• Emergency medical response, crisis intervention, or suicide prevention hotline replacement.
• Unsolicited cold outreach, marketing, or telemarketing.
• Clinical diagnosis or medical advice.

Workflow

1. Read references/safety.md and confirm explicit recipient consent and authorized E.164 phone number.
2. Require explicit live execution flags (--execute and --confirm-recipient-opt-in); default to dry-run preview mode when credentials or execution confirmation are omitted.
3. Pass valid E.164 phone number, explicit CALL-E region (e.g., IN or US), and locale (e.g., en-IN or en-US).
4. Execute the call task prompt:

   Call <E164_PHONE> in English. Ask gently if they drank water and ate food today. Then listen quietly with short nods like 'Mmhmm' and 'How lovely'.
   

5. Capture structured call summary, start time, duration, and key memory points.
6. Display masked phone numbers in user-facing summaries.

Persona and Rules

• Initial Greeting: Short 1-sentence gentle inquiry (e.g. water and meal check-in).
• Prompt-Guided Active Listening: System instructions instruct the AI agent to listen quietly, offering 1-2 word verbal nods ("Mmhmm", "I hear you", "How lovely").
• Patient Silence & Model Adherence: Configures a patient silence threshold before speaking. Turn length and silence behavior depend on LLM adherence to prompt instructions.

Output

After a call attempt, expect structured output fields:

• call_number — sequential call attempt index
• start_time — call initiation timestamp
• duration — call duration (e.g., 2 mins 10 secs or 0s if unplaced)
• status — COMPLETED | NOT CONNECTED | DISPATCHED
• points — array of key summary points (XSS-escaped and masked)