Slack Gather Context

Slack: Gather Context & Attention Triage

When to Activate

User asks what's happening in Slack, what needs their attention, searches for specific information, or wants a summary of activity.

Step 1: Identify What the User Needs

Attention triage (what needs me?): → Go to Step 2: Attention Scan

Specific search (what did X say about Y?): → Go to Step 3: Smart Search

Channel pulse (what's happening in #channel?): → Go to Step 4: Channel Pulse

Step 2: Attention Scan

Gather everything that needs the user's attention:

DMs and mentions:

SLACK_SEARCH_MESSAGES(query="to:me", sort="timestamp", sort_dir="desc", count=20)

Urgent/important signals — search for keywords:

SLACK_SEARCH_MESSAGES(query="urgent OR ASAP OR blocked OR critical OR help after:today", sort="timestamp", sort_dir="desc", count=20)

Channels with activity — check key channels:

SLACK_FIND_CHANNELS(query="general") → channel_id
SLACK_FETCH_CONVERSATION_HISTORY(channel=channel_id, limit=10)

Synthesize into priority buckets:

URGENT (needs response now):
  - @mention from Sarah in #engineering about prod issue (2m ago)
  - DM from Mike: "Can you review my PR ASAP?"

IMPORTANT (respond today):
  - Thread in #product about Q2 planning (you were mentioned)
  - DM from Alex about meeting reschedule

FYI (can wait):
  - #announcements: New onboarding doc shared
  - #random: Team lunch poll

Step 3: Smart Search

Construct precise queries using Slack search modifiers:

Query modifiers:

• in:#channel — search within specific channel
• from:@user — search by sender
• to:me or to:@user — messages directed to someone
• before:YYYY-MM-DD / after:YYYY-MM-DD — time range
• on:YYYY-MM-DD — specific date
• has:link / has:file / has:reaction — content filters
• is:thread — only threaded messages

Example queries:

# Find deployment discussions this week
SLACK_SEARCH_MESSAGES(query="deployment after:2025-01-20 in:#engineering", sort="timestamp", sort_dir="desc")

# Find what a specific person said
SLACK_FIND_USERS(search_query="Sarah") → get user display name
SLACK_SEARCH_MESSAGES(query="from:@sarah project update", sort="timestamp", sort_dir="desc")

Progressive search:

1. Start specific: "deployment error in:#production from:@ops after:today"
2. If no results, broaden: "deployment error after:today"
3. If still empty, broaden more: "deployment error"

Step 4: Channel Pulse

Read what's happening in a channel:

SLACK_FIND_CHANNELS(query="product") → channel_id
SLACK_FETCH_CONVERSATION_HISTORY(channel=channel_id, limit=30)

For threads, always expand:

SLACK_FETCH_MESSAGE_THREAD_FROM_A_CONVERSATION(channel=channel_id, thread_ts=message_ts)

Using spawn_subagent for Multiple Channels

When gathering context from multiple channels in parallel:

spawn_subagent(task="Get recent messages from #engineering channel", context="Limit: 20 messages, focus on technical discussions")
spawn_subagent(task="Get recent messages from #product channel", context="Limit: 20 messages, focus on roadmap updates")
spawn_subagent(task="Get recent messages from #general channel", context="Limit: 20 messages, focus on announcements")

This allows parallel context gathering from multiple channels.

Summarize by topic, not chronology:

#product channel — last 24h:
  Topics discussed:
  1. Q2 Roadmap (3 threads, 15 messages) — Decision: Focus on mobile
  2. Bug triage (1 thread) — 4 Critical bugs assigned to eng
  3. Customer feedback (2 threads) — NPS survey results shared
  
  Pinned: Product roadmap doc (updated yesterday)

Step 5: Synthesize & Present

Always provide:

• Source: channel name + timestamp for each finding
• Context: thread summary, not just individual messages
• Action items: What the user should respond to
• Links: Message permalinks when relevant

Anti-Patterns

• Searching without using date filters (gets stale results)
• Not expanding threads (miss critical context)
• Dumping raw messages without synthesis
• Ignoring DMs when user asks "what needs my attention"
• Using sort="score" when user wants recent (use sort="timestamp")