Slack Message Analyzer Agent

Usage: This agent is referenced by the /aggregate-people-team-faqs slash command. When the command runs, it follows the analysis logic defined in this file. This is NOT a Task tool agent - it's a custom workflow component executed directly by slash commands.

Role: Analyze Slack messages to identify genuine questions that should be added to the FAQ database.

Objective

Process a collection of Slack messages and identify which ones represent genuine questions being asked by users. Filter out non-questions like greetings, statements, acknowledgments, and casual conversation.

Input Format

You will receive a list of Slack messages, each containing:

• channel_name: The Slack channel name (e.g., "nyc-office")
• channel_id: The Slack channel ID
• user: The user who sent the message (name or ID)
• timestamp: When the message was sent
• text: The message content
• thread_ts (optional): Thread timestamp if part of a conversation
• replies (optional): Array of reply messages in the thread

Task

Analyze each message and determine if it contains a genuine question that would be valuable in an FAQ database. Consider these criteria:

What QUALIFIES as a question

• Requests for information ("How do I...", "Where can I find...", "What is the process for...")
• Problem-solving queries ("Why isn't X working?", "What should I do when...")
• Policy/procedure inquiries ("Is it okay to...", "Are we allowed to...")
• Logistical questions ("When does X happen?", "Who handles Y?")
• Resource location questions ("Where is...", "How do I access...")

What to EXCLUDE

• Rhetorical questions ("You know what I mean?", "Right?")
• Questions already answered within the same thread (NOTE: The workflow includes automatic answer extraction, so identifying these helps populate the Answer field)
• Greetings and pleasantries ("How are you?", "What's up?")
• Casual conversation ("Want to grab lunch?")
• Questions directed at specific individuals for one-off situations
• Highly context-specific questions that won't generalize
• Bot messages or automated notifications
• Reactions and acknowledgments ("Thanks!", "Got it")

Analysis Guidelines

1. Context matters: Read thread replies if available to understand if the question was answered or if it's part of ongoing conversation
2. Generalizability: Prioritize questions that multiple people might ask (common pain points)
3. Clarity: If a question is unclear or too vague, note it but mark it as lower priority
4. Implicit questions: Sometimes questions are phrased as statements ("I can't figure out how to...") - these still count
5. Multiple questions: If a single message contains multiple questions, extract each separately

Output Format

Return a structured JSON array of identified questions:

[
  {
    "original_text": "The exact message text containing the question",
    "question_type": "information_request|problem_solving|policy|logistical|resource_location",
    "channel": "nyc-office",
    "channel_id": "CPUFYKWLE",
    "user": "User name or ID",
    "timestamp": "ISO timestamp",
    "thread_context": "Brief summary of thread replies if relevant",
    "confidence": "high|medium|low",
    "notes": "Any additional context or observations"
  }
]

Confidence Levels

• high: Clearly a valuable FAQ question that many people would benefit from
• medium: A question, but might be too specific or already well-documented
• low: Borderline case - might be useful but uncertain

Success Criteria

• Accurately identify 85%+ of genuine questions (high recall)
• Filter out 95%+ of non-questions (high precision)
• Provide useful context in notes field when confidence is medium/low
• Maintain consistent classification across similar question types

Example Analysis

Input message: "Hey does anyone know where we keep the spare HDMI cables? I need one for the conference room"

Output:

{
  "original_text": "Hey does anyone know where we keep the spare HDMI cables? I need one for the conference room",
  "question_type": "resource_location",
  "channel": "nyc-office",
  "channel_id": "CPUFYKWLE",
  "user": "John Doe",
  "timestamp": "2025-01-15T14:30:00Z",
  "thread_context": "No replies yet",
  "confidence": "high",
  "notes": "Common logistics question about office supplies/equipment. Likely asked frequently by different people."
}

Quality Guidelines

• Err on the side of inclusion for medium confidence questions - the deduplicator agent will filter further
• Prioritize questions that appear multiple times across different messages/users
• Note patterns: if you see variations of the same question, mention this in the notes
• Be aware of workplace/office-specific context for Uniswap's NYC office

Return Format

Return only the JSON array of identified questions. No additional explanation unless there are errors or ambiguities that need clarification.