Context
Your company has expectations for its software: how it should look, how information should be protected, and how it should behave. Those expectations may come from different teams and vary between products.
The problem
An AI assistant needs those requirements to do the job correctly. Without them, developers and reviewers spend time correcting work that could have been guided from the start. Copying instructions into every project creates another maintenance task: when a standard changes, each copy needs updating.
Approach
The tool connects Claude Code, an AI coding assistant, to standards maintained by the teams responsible for them.
When a developer starts a task, the assistant receives the standards relevant to that project and the work being done. Teams can update their standards centrally, and connected projects receive those changes on their next request.
A separate AI reviewer then checks the completed work against those standards and identifies issues for correction. It can also assess an existing project to help a team understand where improvements are needed.
People stay in control
Owners keep their standards
The teams responsible for the standards continue to own them and update them in one place.
The project decides what applies
Each project’s requirements determine which standards apply to it.
Exceptions are recorded
An exception requires a person’s decision and a recorded reason.
Gaps go to the team
Concerns the standards don’t address are reported separately, so the team can decide what to do.
Human review remains
The AI reviewer checks the work first; a person still reviews it.
Why it matters
The goal is fewer repeated corrections, giving reviewers more time to assess whether the software meets the business need.
It gives teams a consistent way to bring their own requirements into AI-assisted development while keeping ownership and decisions with people.
We also use the tool in our own development workflow.
This is a working proof of concept built for Claude Code. The source and technical documentation are available on GitHub; production adoption would require further hardening.