Conversation Context Persistence Contract

Problem/Feature Description

An AI customer service system stores user conversation context in a JSON-based key-value store. When a user returns for a follow-up session, the agent is supposed to "remember" prior preferences, issue history, and open action items. Engineering has observed that some sessions start cold — the agent answers as if no prior conversation occurred — even though no errors appear in the application logs. The suspected cause is a race condition where stale or deserialisation-failed cache entries are silently ignored.

The team needs a detectability contract for the memory-resume boundary so that every time the agent attempts to restore session context, the restoration is validated before the session begins. If the context is unrestorable for any reason, the system should know exactly why and what to do about it, rather than silently defaulting to an empty session.

Output Specification

Produce the following files:

1. contract.md — A boundary contract document covering the memory-resume stage of this workflow, with all required columns filled in. Include the specific verification probes that are appropriate for checking memory persistence.
2. resume_check.py — A Python script that accepts a JSON file path as a command-line argument. The script should load the file, check that the required session keys are present (session_id, user_id, context, timestamp), verify the timestamp is fresh (within 5 minutes of current time), and print a structured report to stdout indicating which checks passed and which failed.

The script should be runnable with: python resume_check.py <path_to_session_file>