Team Brain – Shared Context for Humans and Agents

The problem

An agent can only use the context it can reach. Research, product decisions, and operating knowledge held in one engineer’s prompt collection leave everyone else starting from scratch.

I designed and stood up the team’s shared knowledge repository, then helped turn it into an operating system for maintaining that context: an agent-maintained wiki schema, hybrid search, a structural knowledge graph, a memory lifecycle, reusable skills, lifecycle hooks, and lint checks in CI.

What it makes possible

The repository preserves source material, synthesized knowledge, decisions, contradictions, and the confidence behind emerging conclusions. Retrieval helps an engineer or agent reach the relevant slice rather than load the entire corpus.

Product work can open the product repository and the brain in the same session while preserving ownership. The product repository remains authoritative for its code, tests, architecture, and local constraints. Broader team knowledge stays in the brain. A reusable lesson returns through a separate, reviewable contribution.

I also wrote the team’s harness thesis and Agent Delegation Stack, connecting context work to the controls, evaluations, and operating practices needed for safe delegation.

Delivered capability

Engineers and agents can retrieve source-backed research and decisions from a shared repository, while product-specific code and constraints remain in the product repository. A reviewed lesson can be contributed once and reached from another agent tool rather than remaining in an individual prompt collection.

My responsibility was the repository architecture and its schema, retrieval, graph, memory, and quality tooling. This establishes a maintained context capability; broader adoption and productivity effects remain questions to measure.