Agent SwarmPlaybooks
Eight production playbooks for running an agent swarm — feature development, prospecting, content, UX audits, customer support, code health, cross-source reports, and self-documenting releases. Real flows, real schedules, real agents.
These playbooks document how we actually run our internal swarm at Desplega Labs. Each one corresponds to a "Known Use Case" from the agent-swarm README and gives you the same three things:
- Agents — which roles do the work, with links to the official agent templates.
- Tools & Skills — what each agent calls, grouped by source (see the legend below).
- Workflows / Schedules — the actual scheduled jobs and on-demand flows that drive the use case.
These are starting points, not the only way. They're deliberately high-level — copy the structure, adapt the specifics, and lean on the hot patterns (litmus tests, drain loops, HITL gates, per-customer working directories, no-op workflows) that recur across every flow.
Skill-source legend
Every skill referenced in a playbook is tagged with one of three sources:
| Tag | Meaning | Where it lives |
|---|---|---|
[Built-in] | Ships with the agent-swarm package | desplega-ai/agent-swarm → plugin/ |
[ai-toolbox] | Open-source Claude Code plugin | desplega-ai/ai-toolbox → cc-plugin/ |
[Custom] | Authored by us, currently swarm-managed via skill-create / skill-install (publishing model under discussion) | Not in a public repo yet |
Agent links always point at the matching official template so new swarm users can clone the template as a starting point.
The eight playbooks
- Feature Development — Slack/Linear request → research → plan → implementation PR → review → merge.
- Lead Prospecting — Daily discovery + drafting + scheduling, with a human approval gate before sending.
- Content Generation — Topic mining, blog/social/meme production, all gated by a different-model-family LLM-as-judge reviewer and finalized by humans.
- UX Command Center — Weekly UX audits, agentic session recording, design-system enforcement, telemetry-driven proposals.
- Proactive Customer Support — Per-customer working directories, scheduled value-showcase reports, post-meeting follow-ups.
- Code Health & Alert Management — Datadog/New Relic/Sentry/SigNoz alerts kick off fixes or proposals; daily/weekly health audits.
- Reports from Multiple Sources — Join your data warehouse, analytics, billing, search, and observability into one swarm.
- Self-Documenting & Release Reports — Auto-update docs, generate release notes, render release videos with Remotion + browser captures.
Hot patterns
Patterns that recur across multiple playbooks — read these too if you want to skip the same mistakes we made:
- Litmus Tests — LLM-as-judge quality gates that hard-reject sub-bar output.
- Drain Loops — turn one big ticket into stacked, individually-reviewable PRs, then a merge loop.
- HITL Gates — pause a workflow until a human approves in Slack.
- Per-Customer Working Directories — persistent per-account context in agent-fs.
- No-op When Nothing Changed — detect "nothing changed" and skip silently.
Workflows
Run multi-step AI agent pipelines automatically — trigger workflows from webhooks, schedule them with cron, route tasks conditionally, and recover from crashes. Orchestrate your agent swarm with DAG-based automation.
Feature Development
Take feature requests from Slack or a Linear ticket and turn them into pull requests, end to end — with research, planning, implementation, review, and merge handled by specialized agents that hand context off via agent-fs.