The problem
Most organizations now monitor models, versions, and tokens, but the prompt layer is still unmanaged text. That means no clear answer to basic governance questions: who changed this instruction, what risk did it introduce, and can the behavior be reproduced later?
What I built
Prompt Control Plane is a deterministic governance layer for prompts. It adds structure before execution: risk scoring, policy enforcement, cross-provider routing, and a tamper-evident audit trail. It ships in three useful surfaces at once:
- an MCP server for AI-native workflows
- a CLI for local and CI usage
- a GitHub Action for pull-request quality gates
Why it matters
I kept LLM calls out of the engine itself. Rules and policy checks run deterministically, so the team can inspect how a prompt decision was made and reproduce it later.
Distribution
- Product site with full documentation
- npm package for local and CI integration
- GitHub Action for pull-request quality gates