Family Conflict Aftercare Analysis

Problem to solve

After a household conflict, emotions often stay elevated: a slammed door or an angry hand gesture can turn into a cold silence or a second argument. Typical security systems identify what happened, but they do not suggest how to settle the moment without escalating it. This skill targets shared family areas such as the living room, kitchen, and dining room. It detects conflict events and then proposes non-intrusive aftercare prompts, without replacing counseling or judging family members.

How the skill works

The input should be camera audio/video with an audio track, commonly mp4, avi, or mov, up to 10MB; network URLs can be downloaded by the service. It first analyzes audio signals such as peak decibel level, sustained 75 dB duration, shouting voice characteristics, door slams, object impacts, crying, and silence duration. It then uses video signals such as aggressive arm swings, pacing, a person leaving the frame, minimum physical distance, and face-away duration. Results are classified into none, mild_dispute, conflict, intense_conflict, and critical_redline, and a calm window is checked, by default around 10 minutes of silence, no new conflict, distance recovery, and at least one person returning to view. When satisfied, the skill may suggest smart_speaker_soft_music, smart_speaker_gentle_voice, or family_app_push, with a soft lead-in bell. If suspected physical violence, minors present, dangerous objects, or injury signs appear, it must escalate to safety resources such as 12338, 110, and 400-161-9995 instead of offering aftercare.

Boundaries

It is intended for shared family areas only, not bedrooms, bathrooms, or private children's rooms. It does not perform marriage scoring, personality analysis, or long-term storage of raw family conversations; normally only metrics and short event clips are retained. Prompts during active conflict may worsen escalation, so they must wait for the calm window. Historical reports require a valid open-id and should be queried from the cloud API, not local memory. The output is event detection and aftercare advice, not psychological treatment.