aj/Breehavior-Monitor
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Keep only the last segment after bracketed metadata in LLM responses

Browse Source 
The model dumps paraphrased context and style labels in [brackets]
before its actual roast. Instead of just removing bracket lines
(which leaves the preamble text), split on them and keep only the
last non-empty segment — the real answer is always last.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
AJ Isaacs
2026-02-23 15:31:09 -05:00
parent 02b2870f2b
commit e4239b25c3
1 changed files with 10 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
cogs/chat.py
View File

@@ -205,12 +205,13 @@ class ChatCog(commands.Cog):
await typing_ctx.__aexit__(None, None, None)
# Strip leaked metadata the LLM may echo back.
# The LLM paraphrases/reformats injected context in unpredictable ways,
# so nuke any [bracketed block] that sits on its own line — real roasts
# don't use standalone bracketed text.
# The LLM often dumps paraphrased context and style labels in [brackets]
# before/between its actual answer. Split on those bracket lines and
# keep only the last non-empty segment — the real roast is always last.
if response:
response = re.sub(r"^\s*\[[^\]]*\]\s*$", "", response, flags=re.MULTILINE)
response = re.sub(r"\n{2,}", "\n", response).strip()
segments = re.split(r"^\s*\[[^\]]*\]\s*$", response, flags=re.MULTILINE)
segments = [s.strip() for s in segments if s.strip()]
response = segments[-1] if segments else ""
if not response:
log_channel = discord.utils.get(message.guild.text_channels, name="bcs-log")
@@ -317,10 +318,11 @@ class ChatCog(commands.Cog):
recent_bot_replies=recent_bot_replies,
)
# Strip leaked metadata (same catch-all as main chat path)
# Strip leaked metadata (same approach as main chat path)
if response:
response = re.sub(r"^\s*\[[^\]]*\]\s*$", "", response, flags=re.MULTILINE)
response = re.sub(r"\n{2,}", "\n", response).strip()
segments = re.split(r"^\s*\[[^\]]*\]\s*$", response, flags=re.MULTILINE)
segments = [s.strip() for s in segments if s.strip()]
response = segments[-1] if segments else ""
if not response:
return