What happened?

_chunk_by_exchange() in mempalace/convo_miner.py (line 73) caps AI responses
at 8 lines before storing them in the palace:

ai_response = " ".join(ai_lines[:8])

Any content after line 8 is silently discarded and never stored anywhere. A
50-line code example, a step-by-step guide, or any detailed AI response gets
stored as 8 lines with no warning.

What did you expect?

The full AI response to be stored verbatim, consistent with the project's stated
core principle: "Verbatim first: Never summarize user content. Store exact words."

I also want to confirm whether the 8-line limit was intentional before submitting
a fix — I can imagine it was meant to keep the Q+A chunk within CHUNK_SIZE, but
an arbitrary line count silently drops content rather than splitting it cleanly.
Happy to hear if there's a reason for it I've missed.

How to reproduce:

1. Install mempalace and run the following script:

from mempalace.convo_miner import _chunk_by_exchange

conversation = (
    "> How do I implement JWT authentication in Flutter?\n"
    "Step 1: Add the jwt_decoder package\n"
    "Step 2: Create an AuthService class\n"
    "Step 3: Implement login with token storage\n"
    "Step 4: Add token to HTTP headers\n"
    "Step 5: Handle token expiry\n"
    "Step 6: Implement logout\n"
    "Step 7: Add StreamController for auth state\n"
    "Step 8: Wire up AuthBloc\n"
    "Step 9:  SHOULD be stored\n"
    "Step 10: SHOULD be stored\n"
).split("\n")

chunks = _chunk_by_exchange(conversation)
print(chunks[0]["content"])

2. Observe that Steps 9 and 10 are missing from the output.
3. Mine any real conversation export where the AI gave responses longer than
   8 lines and search for content from those responses — it will not be found.

Environment:

• OS: macOS 15
• Python version: 3.11
• MemPal version: 3.1.0 (git SHA: 068dbd9)