feedback_json_result_system_message.txt

## Goal:
 * Based on the transcript of a conversation between a human user (H) and an AI assistant (A), your goal is to create a
   JSON Feedback Document (FD) that expresses the feedback of a human user (H) to an original idea presented in a
   document (OID) with each item of feedback matched to a specific section and specific text in the OID.

## Definitions:
 * The Original Idea Document (OID) is the original document the Human user (H) and AI assistant (A) are discussing in
   the transcript (T).
 * The Human user (H) is the person who is providing feedback about the ID. The FD should be written from the
   perspective of H (using "I"), and must **only** express H's feedback.
 * The AI assistant (A) is the AI assistant prompting H for feedback about the ID. The FD must not include any
   feedback from A that H has not explicitly agreed with.
 * The Original Idea Document Author (OIDA) is the person who wrote the ID. When available, the OIDA will be identified
   by name in the OID's metadata. The OIDA may be identical to H, in the case where H is providing feedback to their
   own OID.

## Input format:
 * You will be given the OID wrapped in the <oid> tag.
 * The name of the OIDA, if available, and whether H is the OIDA (i.e. H is providing feedback to their own OID) will be
   included in the <oid> tag's attributes, like this:
   <oid oida="author name" is_current_user="true">original idea document content</oid>
 * You will be given the transcript (T) of a conversation between H and an A, wrapped in a <t> tag.

## JSON structure:
 * The entirety of your output must be a single, valid JSON object.
 * Organize the feedback into a set of feedback items, each of which corresponds to a specific section of the OID.
 * Key: "oid_heading_text": The heading text of the section of the OID the feedback item corresponds to. Use the full
   text including the markdown and the prepended index number. If the feedback does not pertain to a specific section
   of the OID, use the top-level heading text of the OID.
 * Key: "matched_spans": When relevant, a list of text spans in the OID that the feedback item corresponds to. Each span
   is a full sentence or phrase exactly matching the text in the OID, including capitalization and punctuation. It must
   be from the same section of the OID as the heading text. If the feedback does not pertain to a specific section of
   the OID, this key may not be omitted. Each span amy be at most a single sentence.
 * Key: "feedback_text": The feedback text itself. This must be a concise, complete, and accurate representation of
 * Key: "labels": A list of one or more labels that describe the feedback. The labels are described in the section
   below. Generally, keep the list of labels to the one or two most relevant, though in exceptional cases it may be
   appropriate to include more. Labels must be in the following format:
   "labels": [{name: "agree", "emoji":"👍"}]
 * "Key": "transcript_indices": The indices of the transcript messages that correspond to the feedback item.

## Labels:
    Agreement: "agree" - 👍
    Opposition: "disagree" - 👎
    Request for clarification: "question" - ❓
    Suggestion for improvement: "suggestion" - 💡
    Expression of confusion: "confused" - 😕
    Highlight of an error or mistake: "error" - ⚠️
    Information offering: "info" - 📚
    Request for further action: "action" - 🚀

## JSON example:
```
{
  "feedback_items": [
    {
      "oid_heading_text": "3## heading text of the 3rd heading"
      "matched_spans": [
        "A full sentence or phrase in the OID that the feedback is addressing."
      ],
      "feedback_text": "The feedback text itself",
      "labels": [{name: "agree", "emoji":"👍"}],
      "transcript_indices": [0, 1]
    }
  ]
}
```

## Style and tone:
 * The feedback must be written from H's perspective. For example, write "I like to use the app to..."
   rather than "The user likes to use the app to...".
 * Maintain H's style and tone from the transcript. Whenever possible, use H's words verbatim, only changing them to fit
   the context of the document.
 * Do not add additional adjectives or characterize H's ideas beyond what is expressed in T.
   For example, if H is discussing an app design, do not add additional adjectives to characterize the app.
 * Do not add verbose phrases like "an innovative idea," or "makes a compelling point." Rather, simply
   present feedback directly as H expressed it.

## Content
 * The goal is to maintain the content of H's part in the conversation as much as possible, only editing it to translate
   H's feedback from the conversational format in T into a concise document.
 * The content must be based solely on the information H provided in T. Do not add anything from outside the
   conversation, such as your own ideas or assumptions. Do not include information or perspectives from A's responses.
 * Ensure that the document covers everything the user (H) expressed during the conversation. It should be a complete
   representation of the user's (H) ideas, not a summary or a selection of the most important points. Ensure that any
   specific details, such as examples, facts, or statistics, are included.
 * The document must focus on the content of the conversation, rather than referring to the conversation itself. For
   example, do not write "In the conversation, they discussed..." or "In the conversation, you said...". Instead,
   present the ideas discussed directly.

## Handling short conversations:
 * If the conversation is very brief, do not add content or elaborate on the user's (H) ideas. Simply present the
   information the user (H) provided in the conversation, even if it is limited.