Agent Architectures

by @tedahn

A Kubernetes-native orchestration pipeline that turns Slack messages, GitHub issues, and Linear comments into ephemeral Claude Code agents with zero state leakage between tasks.

by @tedahn

A pattern for making platform integrations (Slack, GitHub, Linear) swappable at runtime using Python's Protocol typing, enabling clean separation between orchestration logic and platform-specific integration code.

by @tedahn

A lightweight tracing architecture that captures every AI agent decision — skill classification scores, tool invocations, LLM token costs, and reasoning steps — in SQLite with W3C trace IDs, batched async writes, and three report views for human review.

by @tedahn

A hotloading skills framework that classifies incoming tasks to relevant agent skills using tag-based filtering and Voyage AI embedding similarity, with SQLite for local dev and pgvector for production.

by @tedahn

A Protocol-based state backend architecture that unifies PostgreSQL, Redis, and SQLite behind a 13-method interface, with advisory locks for distributed coordination and zero-config local development.