Autonomous agent systems that do real engineering and operations work — in public.
We build and run a live multi-agent fleet: agents that claim work from a priority queue, ship merged pull requests, and recover from their own failures — kept honest by evals, OpenTelemetry, and infrastructure-as-code. The repositories here are the open-source pieces of that system.
- laneq — a tiny lease-based priority queue for coordinating parallel agents without collisions.
- agent-skills — reusable
SKILL.mdworkflows that give agents judgment, not just tools. - MCP servers (capabilities for agents): agent-mcp · telemetry-mcp · memory-mcp · dispatch-mcp · reddit-mcp · dns-mcp
- terraform-modules — reusable OpenTofu modules for cloud/provider resources.
- python-repo-template — the scaffold our public repos start from.
- Done means evidence — a merged PR, a loaded page, a green run — not "the model said so."
- Declarative everything — IaC/GitOps; no hand-mutated live state.
- Evals + observability first — agent work is measured (OpenTelemetry traces) and gated, not hoped for.
- Public-first — share the framework openly; keep only what must be private, private.
Built by Patrick Selamy · selamy.dev