Ezra Monitor

Ezra Monitor

Monitor and examine ezra-prime's activities to propose memory block updates.

Purpose

Self-reflection tool that analyzes ezra-prime's Discord support conversations to:

• Identify recurring patterns and issues
• Detect knowledge gaps from repeated questions
• Propose updates to memory blocks based on learnings
• Track team corrections and user feedback

Scripts

scripts/fetch-recent.sh

Fetches recent conversations and messages from ezra-prime.

./scripts/fetch-recent.sh [hours] [limit]
# Default: last 24 hours, 50 messages

scripts/analyze-patterns.sh

Analyzes fetched messages for patterns.

./scripts/analyze-patterns.sh <messages.json>
# Outputs: recurring topics, unanswered questions, corrections

scripts/propose-updates.sh

Generates proposed memory block updates based on analysis.

./scripts/propose-updates.sh <analysis.json>
# Outputs: suggested edits to specific memory blocks

scripts/scan-corrections.sh

Scans conversations for user/team corrections that need action.

./scripts/scan-corrections.sh [agent_id] [num_conversations]
# Default: ezra-prime, 10 conversations
# Outputs: corrections.json with HIGH/MEDIUM/NORMAL priority items

Detects:

• Team corrections (cameron, swooders, pacjam, 4shub) - HIGH priority
• User corrections ("that's wrong", "actually", etc.)
• Link shares after responses (user providing sources)

Environment

Requires:

• LETTA_API_KEY - API access
• EZRA_PRIME_AGENT_ID - Target agent (default: agent-57ce3ea1-72ad-43e5-a444-7e3724f706e8)

Workflow

1. Run fetch-recent.sh to download conversation data
2. Run analyze-patterns.sh to identify learnings
3. Run propose-updates.sh to generate memory edit suggestions
4. Review and apply updates manually or via API