Conversation JSON To MD

Convert a user-provided chat-export JSON into multiple Markdown files with consistent Q/A formatting.

When to Use

Use this skill when the user asks for:

• Splitting one JSON chat export into many .md files
• One conversation per markdown file
• Keeping only question/answer content from user and assistant
• Renaming response section to 回答
• Normalizing exported files with a second formatting pass

Do not use

Do not use this skill for:

• Plain text transformation that does not involve JSON chat exports
• Non-conversation JSON processing tasks
• Requests requiring semantic summarization instead of structural conversion

Instructions

1. Read the input file path provided by the user. Do not assume default file names.
2. Detect conversation/message structure automatically.
3. Export one markdown file per conversation.
4. Keep only user/assistant Q&A content.
5. Format each Q/A block as:
   • ## <question text>
   • ### 回答
6. Preserve answer markdown and demote answer-internal heading levels by one level.
7. Run an independent second-pass formatting check and fix naming/title structure before final delivery.

Supported Input Structures

The bundled script supports common export formats including:

• DeepSeek/ChatGPT-like mapping tree (mapping/root/children/fragments)
• Qwen-like exports (data[].chat.messages[], content_list with phase=answer)
• Claude web export style (list[{ name, chat_messages: [...] }])
• Generic message arrays (messages, history, conversations, dialog, turns)
• Pair fields (question-answer, prompt-response, input-output)

If format detection fails, stop and ask the user for a sample snippet, then extend parsing rules.

Run Script

python3 scripts/convert_conversations.py \
  --input /path/to/<user-provided>.json \
  --output-dir /path/to/output_md \
  --clean

Output Format

Each output file uses this structure:

# <conversation title>

## <user question 1>
### 回答
<assistant answer markdown>

## <user question 2>
### 回答
<assistant answer markdown>

Second-Pass Formatting (Required)

After export, run a second-pass check/fix on output files:

1. Filename normalization:

• Keep title-only naming
• Remove illegal filename characters
• Resolve duplicates with (2), (3)...

2. Heading normalization:

• Keep only one H1: # <conversation title>
• Ensure questions are H2
• Ensure responses are exactly ### 回答

3. Body normalization:

• Keep answer body markdown
• Keep answer-internal heading demotion

4. Final verification:

• Confirm no files still violate naming or heading rules

Validation Checklist

• File count equals detected conversation count
• No random suffixes in filenames
• No ## REQUEST or ## RESPONSE headers in output
• Response blocks are present as ### 回答
• Output preserves markdown rendering correctly