Slack Digest

Produce a weekly digest of key Slack threads from configured channels using Slack semantic search MCP tools. The digest surfaces high-engagement threads, groups by topic, and highlights unanswered questions.

Configuration

See _shared/report-config.md for the standard config resolution pattern.

Config schema (~/.claude/skills/slack-digest/config.json):

• channel_ids: list of Slack channel ID strings
• lookback_days: number (default: 7)

Required scope: at least one channel_ids entry.

Arguments

• channel_ids: list of Slack channel IDs to scan
• lookback_days: how far back to search (default: 7)

Workflow

1. Search Slack channels. See _shared/slack-search.md for the standard Slack search workflow. Use query terms: "discussion update question issue decision" with size=20 per channel. Count replies per thread from the full thread fetch — this is the engagement signal.
2. Classify threads. For each thread, determine:
   • Topic category: incident, decision, question, announcement, discussion, or request
   • Engagement level: high (10+ replies), medium (4-9 replies), low (1-3 replies)
   • Resolution status: resolved (answer given, decision made), unresolved (question still open, no conclusion), or informational (no resolution needed)
   • Classification is based on thread content: questions end with ? or start with "does anyone", "how do we"; decisions contain "let's go with", "we decided", "agreed"; incidents contain "incident", "outage", "SEV", "pages".
3. Identify unanswered questions. A thread is "unanswered" if: it was classified as a question AND has fewer than 3 replies AND no reply contains a clear answer (code block, link, or statement without a trailing ?).
4. Group threads by topic category. Within each category, sort by engagement (highest first).
5. Generate report. Output markdown:

## Slack Digest — {date}
Channels: {channel names or IDs} | Period: last {lookback_days} days | Threads analyzed: {N}

### Summary
- Total threads: {N} | High engagement: {N} | Unanswered questions: {N}

### Unanswered Questions (action needed)
(For each: channel, thread summary, who asked, when, permalink)

### Key Decisions
(For each: thread summary, decision outcome, participants, permalink)

### Incidents & Outages
(For each: thread summary, status, permalink)

### Active Discussions
(For each: thread summary, reply count, key participants, permalink)

### Announcements & FYI
(Compact list: summary, permalink)

6. Deliver as HTML. Follow the shared HTML delivery pattern in _shared/html-delivery.md. Report name: slack-digest. TLDR includes thread count, unanswered questions count, and top discussion topic.
7. Terminate. Report is complete when all channels are searched and threads are grouped with permalinks.

Design Principles

1. Team-level only. Aggregate to team level — it is the right granularity for a periodic digest. Individual-level detail (messages per person, reply count per author) is too noisy for a team report.
2. Deterministic math first, LLM narrates only. Thread counts, reply counts, and engagement levels must be computed from actual thread data. The LLM writes summaries around the numbers but never invents engagement metrics.
3. Pair metrics with counter-metrics. Never report activity volume without resolution rate. If showing total threads, also show how many were resolved vs left open.

Golden Rules

1. Fetch full threads, not just snippets. The RAG result is a snippet. Use fetch-slack-thread to get the complete conversation before classifying.
2. Unanswered questions go first. They represent work stuck waiting for input. Surface them before everything else.
3. Include permalinks for every thread. The reader must be able to click through to the original conversation.
4. Count replies from the actual thread, not from RAG metadata. RAG returns relevance-matched snippets, not engagement data. The full thread fetch gives the real reply count.
5. Check ALL configured channels, not a subset.

Anti-Rationalization Counter-Table

Termination Labels

Self-Review Checklist

• Every configured channel was searched
• Full threads were fetched (not just RAG snippets) for classification
• Reply counts come from actual thread data
• Unanswered questions section appears first after summary
• Every thread entry includes a permalink
• Threads are grouped by topic category, sorted by engagement within each group
• HTML version uploaded to S3 with commuter link (unless --no-html or upload failed with noted fallback)
• Slack/chat delivery uses TLDR + link, not the full report