Create a compact summary of the visible work episodes in this conversation that relate to the main task developed in this chat.

Identify that task only from the visible context. If several clearly distinct tasks appear and you cannot determine which one I want summarised, ask me in one sentence and wait for my answer.

The purpose is to explain how the person and the AI interacted. Do not assess authorship, effort, quality, misconduct or percentages of contribution.

SCOPE
- Use only earlier content that remains visible in this conversation. Exclude this prompt and earlier HREVN exports.
- Do not use memories, profiles, other chats, external information, or new searches or actions.
- Do not claim to see the complete history. Add an item to context_gaps only when a loss of context is explicitly visible.

PRIVACY
- Use short paraphrases and generic roles. Do not include quotations, names, email addresses, phone numbers, addresses, identifiers, credentials or URLs.
- Do not identify your provider or model; the person will indicate it in HREVN.

EPISODES
- Group consecutive messages with the same purpose into one episode. Do not create one episode per message.
- Start another episode when the purpose changes, a distinct human decision appears, a revision or transfer occurs, or the outcome changes.
- Preserve substantive actions by the person and AI as separate steps.
- Omit greetings, repetitions and technical explanations that do not affect the task.
- Group repetitive material and keep every description below 280 characters. There is no fixed maximum number of episodes.

INTERPRETATION
- Do not infer approval from continuation, final use from production, authorship from providing material, or communication between chats from similarity.
- Do not use “author”, “co-author” or equivalent expressions. State only that a person prepared, reviewed, coordinated or was assigned a part.
- Record connections only when they are visible. Their absence from the chat does not prove that they did not occur.

HUMAN ACTION TYPES
request, define_goal, set_criterion, provide_material, clarify, correct, reject, select, approve, request_verification, coordinate, document, continue, other.

AI ACTION TYPES
ask, propose, plan, research, draft, revise, analyze, explain, summarize, verify, format_or_create_file, technical_failure, other.

CONNECTIONS
connections is a list that may contain only: input_from_other_ai, output_to_other_ai, input_from_other_person, output_to_other_person, peer_review, integration, coordination.
Each item in connections must be only one of those quoted names. Never use objects, explanations, type or summary inside connections. Valid example: ["peer_review", "integration"].

PROCESS DOCUMENTATION
documentation is a list that may contain only: external_work_memory_created, external_work_memory_updated, dated_entry_added, objective_logged, plan_logged, roles_logged, decision_logged, correction_or_rejection_logged, source_logged, source_check_status_logged, file_or_version_logged, handoff_logged, pending_tasks_logged, ai_use_logged, external_save_confirmed.

Use documentation only when a work record, continuity note, log or file intended to preserve the process outside the chat was deliberately created or updated. Do not mark a conversation summary, an ordinary answer, the final deliverable or merely discussing a topic as process documentation. external_save_confirmed requires the person to have confirmed an external download or save. If no such evidence exists, use [].

EPISODE RESULT
produced, revised, accepted, rejected, continued, failed or not_established.

OUTPUT
After preparing the JSON and before delivering the file, explain in your own words, in two to four short sentences that a non-technical person can understand:
- which parts or moments of the work you were able to summarise;
- which decisions, actions or collaborations actually appear;
- which measures for preserving the work record appear, if any;
- which information you could not recover, if any;
- that the file does not contain the complete conversation, quotations, personal data, links, or assessments of authorship, effort or quality.

Also explain, without jargon, that the file will be used to produce an educational workbook about the process. Do not mention field names, codes, schemas or validation rules. Do not repeat a fixed formula, do not describe information that is absent, and do not claim that HREVN has verified the content. This explanation remains outside the file.

Then create a downloadable file named exactly JSON-AI-workbook.json containing only the JSON object, with no Markdown, comments or additional text. If you cannot create files, show the explanation and then the JSON object in one clearly separated code block, and say that it can be saved under that name.

Use null or empty lists where appropriate. episode.order must start at 1 and be consecutive. In steps, use only actions compatible with the actor.

{
  "schema_version": "hrevn-ai-chat-summary/2.0.1",
  "context_gaps": [],
  "work_focus": "Brief description of the work visible in this chat",
  "responsibility": null,
  "episodes": [
    {
      "order": 1,
      "purpose": "Specific purpose of the episode",
      "steps": [
        {
          "actor": "human",
          "action": "request",
          "summary": "Short paraphrase of what the person did"
        },
        {
          "actor": "ai",
          "action": "draft",
          "summary": "Short paraphrase of what the AI did"
        }
      ],
      "result": "produced",
      "result_summary": "Visible outcome, without claiming final use",
      "connections": [],
      "documentation": []
    }
  ]
}

Before replying, check only that the JSON file is valid, that episode order is consecutive, that connections contains names only and never objects, that you did not invent actions, connections, documentation or gaps, and that you did not include personal data, quotations, URLs or information that appeared after this prompt. The explanation remains outside the file.