The problem
Agents and teams lose quality when decisions, specifications, and the real state of the code live in separate places. The goal was to build an engineering layer that kept those elements connected and detected when implementation moved away from the agreed context.
The solution
Flow Engineering organizes work as an explicit flow. A state machine governs progress, deterministic snapshots make runs comparable, and drift detection flags divergence. The CLI is the primary interface; MCP extends integration without becoming a required dependency.
Architecture decisions
- Core before interface: state and verification logic do not depend on MCP.
- Reproducible results: deterministic snapshots reduce ambiguity between runs.
- Bounded memory: the Engram bridge preserves context while the prompt registry exposes the instructions governing each operation.
- Quality as a gate: typing, linting, security, tests, and coverage run in CI across two Python versions.
What it demonstrates
This project demonstrates systems design, developer tooling, and operational discipline. It is not presented as a commercial success story: the available evidence covers the repository’s structure and verification.