Buzz is a community management agent. It watches every mention of your brand across Twitter/X, Reddit, Bluesky, LinkedIn, and Hacker News in real time — and for each one, decides whether to draft a reply, suggest a like, or skip, then surfaces the result as an inline card in Slack before any human has to look at a feed.

The core triage logic is nuanced by design. Buzz reads full thread context before deciding, distinguishes genuine product signal from AI-generated engagement, and follows a detailed voice guide so every suggested reply sounds like someone who actually builds the product, not a community manager running from a script. It also monitors community Slack channels on a rolling schedule, catching unanswered user questions and routing them to the right owner with a pre-drafted reply attached.

At the end of every day and every week, Buzz produces sentiment reports — total mention volume, a scored sentiment index, pattern summaries, and testimonial quotes — and optionally posts them to Slack. The weekly report saves directly to the repository so the team has a running historical record.

Open the repo in Warp — or point any agent that loads skills from .agents/skills/ at its URL — and paste one prompt:

Run the setup skill

The setup skill clones the repo if needed, checks prerequisites, creates .env from the template, walks you through credentials for only the skill groups you want, and verifies readiness with check_setup.py. It is safe to re-run anytime.

Or set it up manually — .env is git-ignored, scripts auto-load it, and you only need values for the skills you actually run:

cp .env.example .env

# Optional: chart-generating skills (e.g. reaction-analytics)
python3 -m venv .venv
.venv/bin/pip install -r requirements.txt

# Check which skill groups are ready
python3 .agents/skills/setup/scripts/check_setup.py

Connect Buzz to Octolens for brand monitoring and configure your Slack workspace tokens: one for reading community channels (COMMUNITY_SLACK_TOKEN) and one for posting to internal channels (BUZZ_SLACK_TOKEN). Set up the mentions-proxy webhook to trigger triage on each incoming mention, and schedule the community-channel scan to run every 30 minutes.

The triage skill ships with engagement rules tuned to developer community norms: platform-specific guidelines for Twitter, Reddit, and LinkedIn; reach thresholds that adjust reply priority by audience size; filters for AI-generated posts, organic peer conversations where a brand reply would intrude, and stale or removed threads. These defaults cover most cases out of the box. The reply-learning skill extracts patterns from feedback reactions in Slack and opens PRs to refine the drafting skill over time — the agent improves itself as your team rates its suggestions.

The top-contributor finder and changelog-social-swarm skills run on a separate schedule, surfacing open-source contribution activity and turning merged changelog entries into tweet drafts for your content calendar.

Once live, every suggested reply sits in #feed-mentions for a human to review before anything goes out. You can configure fully autonomous mode for high-confidence actions — likes and skip confirmations — while keeping reply drafts in human-review mode indefinitely.

Out of the box, Buzz connects to Octolens for brand mention monitoring across social platforms, Slack for internal triage cards and daily/weekly reports, and the X API for reading full tweet threads and reply chains. Reddit threads are fetched via the public JSON API with no credentials required.

Additional engagement tools — Typefully for scheduled publishing, community Slack workspaces, and GitHub for issue linking and contributor analysis — can be connected via the Oz skill system. The reply-learning loop uses Slack reactions as its training signal, so no separate feedback infrastructure is needed.