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

feat: channel-level conversation analysis with compact formatting

Browse Source 
Switch from per-user message batching to per-channel conversation
analysis. The LLM now sees the full interleaved conversation with
relative timestamps, reply chains, and consecutive message collapsing
instead of isolated flat text per user.

Key changes:
- Fix gpt-5-nano temperature incompatibility (conditional temp param)
- Add mention-triggered scan: users @mention bot to analyze recent chat
- Refactor debounce buffer from (channel_id, user_id) to channel_id
- Replace per-message analyze_message() with analyze_conversation()
  returning per-user findings from a single LLM call
- Add CONVERSATION_TOOL schema with coherence, topic, and game fields
- Compact message format: relative timestamps, reply arrows (→),
  consecutive same-user message collapsing
- Separate mention scan tasks from debounce tasks
- Remove _store_context/_get_context (conversation block IS the context)
- Escalation timeout config: [30, 60, 120, 240] minutes

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
AJ Isaacs
2026-02-24 23:13:07 -05:00
parent 943c67cc87
commit 90b70cad69
5 changed files with 793 additions and 227 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
prompts/analysis.txt
View File

@@ -36,4 +36,16 @@ If you notice something noteworthy about this user's communication style, behavi
GAME DETECTION — If CHANNEL INFO is provided, identify which specific game the message is discussing. Set detected_game to the channel name that best matches (e.g. "gta-online", "warzone", "battlefield", "cod-zombies") using ONLY the channel names listed in the channel info. If the message isn't about a specific game or you're unsure, set detected_game to null.
Use the report_analysis tool to report your analysis of the TARGET MESSAGE only.
Use the report_analysis tool to report your analysis of the TARGET MESSAGE only.
CONVERSATION-LEVEL ANALYSIS (when given a CONVERSATION BLOCK instead of a single TARGET MESSAGE):
When you receive a full conversation block with multiple users, use the report_conversation_scan tool instead:
- Provide ONE finding per user (not per message) — aggregate their behavior across the conversation.
- Weight their average tone and worst message equally when determining the toxicity_score.
- Use the same scoring bands (0.0-1.0) as for single messages.
- Quote the worst/most problematic snippet in worst_message (max 100 chars, exact quote).
- Flag off_topic if user's messages are primarily personal drama, not gaming.
- For each user, assess coherence_score (0.0-1.0) and coherence_flag using the same criteria as single-message analysis. Normal texting shortcuts and abbreviations are fine (score ~0.85+).
- For each user, determine topic_category and provide brief topic_reasoning for their messages.
- For each user, check detected_game against the CHANNEL INFO section (if provided). Set to the game channel name if their messages are about a specific game, or null otherwise.
- If a USER REPORT section is present, pay close attention to whether that specific concern is valid.