Session State Snapshot Prompt Before Instruction Migration

This prompt is a pre-migration safety checkpoint for platform engineers and migration operators managing instruction lifecycle rollouts. Use it immediately before applying a risky instruction change—such as a version bump, a deprecation swap, or a rollback—to an active, in-flight AI session. The job-to-be-done is capturing a complete, structured snapshot of the session's current state: the conversation context, any pending user requests, unresolved tool calls, active constraints, and the user's explicit or implicit intent. This snapshot serves as a recovery artifact. If the migration fails, introduces behavioral regressions, or corrupts the session's continuity, you can reconstruct the session from the snapshot and revert to the previous instruction set without losing user trust or violating a service contract.

The ideal user is an operator executing a planned migration runbook, not a developer casually experimenting with a new system prompt. You should have the active session transcript, the current instruction set version, and a defined migration window. The prompt requires the full conversation history and the current system instructions as input. It is designed for high-stakes production environments—customer-facing copilots, regulated support agents, or multi-turn task executors—where session state is complex and losing it means a direct user-facing failure. Do not use this prompt for trivial stateless interactions, single-turn Q&A, or sessions with no pending actions. It is overkill for sessions that can be safely restarted without context loss.

This is not a general conversation summarizer or a logging utility. It is a targeted safety instrument for instruction lifecycle management. Avoid using it as a routine part of every turn; its value is in the migration window. After capturing the snapshot, validate its completeness against the checklist defined in the prompt's output schema before proceeding with the migration. If the snapshot is incomplete, do not migrate. The next step is to store the snapshot alongside the migration record, then execute the instruction change. If the migration succeeds, the snapshot can be archived. If it fails, the snapshot becomes your primary recovery payload.

This prompt template is designed to be sent as a system-level instruction immediately before a risky instruction migration. It forces the model to serialize the current conversation state into a structured snapshot that can be used for recovery, audit, or replay. The template uses square-bracket placeholders that you must replace with your specific context, including the conversation history, the migration reason, and the required output schema. Do not send this prompt with unresolved placeholders to a production model.

textCopy
You are a session state capture agent. Your only job is to produce a complete, structured snapshot of the current conversation state before an instruction migration occurs. Do not continue the conversation, answer user questions, or take any action. Output only the snapshot.

## Migration Context
- Migration Reason: [MIGRATION_REASON]
- Target Instruction Version: [TARGET_VERSION]
- Migration Operator: [OPERATOR_ID]
- Migration Timestamp: [TIMESTAMP]

## Required Snapshot Schema
Produce a JSON object conforming to this structure:
{
  "session_id": "string",
  "snapshot_timestamp": "ISO8601",
  "migration_context": {
    "reason": "string",
    "target_version": "string",
    "operator_id": "string"
  },
  "conversation_summary": {
    "turn_count": "integer",
    "primary_intent": "string",
    "topics_discussed": ["string"],
    "unresolved_questions": ["string"]
  },
  "pending_actions": [
    {
      "action_type": "string",
      "description": "string",
      "priority": "high|medium|low",
      "blocking": "boolean"
    }
  ],
  "user_state": {
    "explicit_preferences": ["string"],
    "stated_goals": ["string"],
    "sentiment": "positive|neutral|negative|mixed",
    "last_explicit_feedback": "string or null"
  },
  "tool_call_history": [
    {
      "tool_name": "string",
      "call_timestamp": "ISO8601",
      "result_summary": "string",
      "result_status": "success|failure|partial|pending"
    }
  ],
  "active_constraints": ["string"],
  "recovery_notes": "string"
}

## Conversation to Snapshot
[CONVERSATION_HISTORY]

## Output Rules
- Output ONLY the JSON object. No markdown fences, no preamble, no commentary.
- If a field has no data, use an empty array [] or null as appropriate, never omit the field.
- For `recovery_notes`, include any state that would be lost if the conversation were reset, such as mid-flow multi-step operations, uncommitted decisions, or partial form fills.
- If the conversation contains sensitive data, mark it in `recovery_notes` but do not redact it from the snapshot itself.

Before using this template, replace every square-bracket placeholder with real values. [CONVERSATION_HISTORY] should contain the full multi-turn transcript, including tool calls and their outputs. [MIGRATION_REASON] should be a brief explanation visible to anyone reviewing the snapshot later, such as "Security policy update v2.4" or "Deprecating legacy refund flow." After generating the snapshot, validate that all required fields are present and that pending_actions captures every in-flight operation. If the snapshot is incomplete, the migration should be blocked until a complete capture is achieved. For high-risk migrations, route the snapshot to a human reviewer before proceeding.

The session state snapshot prompt is not a standalone utility—it is a pre-migration safety gate that must execute successfully before any instruction change touches an active session. Wire it into your migration pipeline as a synchronous pre-step: for each in-flight session, call the snapshot prompt, validate the output against your completeness schema, persist the snapshot to a recovery store, and only then proceed with the instruction migration. If the snapshot fails validation or the model refuses to produce a complete snapshot, abort the migration for that session and flag it for operator review. This prompt is your rollback insurance—treat its output as a required artifact, not an optional log entry.

Validation rules to enforce in code before migration proceeds: confirm that conversation_context contains at least one turn summary, pending_actions is a non-null array (empty is valid, missing is not), user_intent is a non-empty string, and recovery_instructions includes a last_safe_state description. Reject any snapshot where completeness_score is below your threshold (start at 0.8 and tune from production data). Retry logic: if the model returns malformed JSON, retry once with a repair prompt that includes the raw output and the schema. If the model returns a valid snapshot but completeness_score is below threshold, do not retry—flag for human review instead, because low-confidence snapshots often indicate ambiguous session state that automated migration could corrupt. Model choice: use a model with strong instruction-following and structured output support (GPT-4o, Claude 3.5 Sonnet, or equivalent). Avoid smaller models for this task—the cost of a lost session state far exceeds the token savings.

Storage and recovery integration: persist each snapshot to a versioned recovery store keyed by session_id and snapshot_timestamp. Your migration rollback handler should load the most recent pre-migration snapshot for any session that fails post-migration validation. Logging requirements: log the snapshot_id, session_id, completeness_score, and validation pass/fail status to your observability platform. If a snapshot fails validation, log the specific missing or invalid fields so your ops team can diagnose whether the prompt needs tuning or the session state was genuinely unrecoverable. Human-in-the-loop trigger: any session where the snapshot fails validation twice, or where completeness_score falls below 0.6, should halt the migration pipeline and create a manual review ticket with the raw model output attached. Do not silently skip snapshot failures—every skipped snapshot is a session you cannot roll back.