How do I make AI follow my team's coding style?

Why style drift is a policy problem

Most teams enforce coding style through review comments — which is expensive, inconsistent, and training-data-hostile for AI. An agent that gets a style note after writing the code has no way to internalise it for next time; the correction does not persist. Style drift is not a quality-of-writing problem; it is a policy-not-codified problem. The fix is to move style from reviewer headspace into machine-readable artefacts the agent consults before writing.

Codifying style into the graph

bRRAIn ingests .editorconfig, linter configs (eslint, ruff, golangci-lint, whatever), and a curated set of exemplar PRs into the POPE graph. Each rule becomes a node with example code; each exemplar PR becomes a pattern the agent can emulate. The Document Portal is where style guides and exemplar bundles live. The Consolidator keeps the style layer in sync as the team evolves conventions — no more "the wiki says one thing and the repo does another".

Inference-time enforcement via the Handler

The Handler loads the style layer into every generation prompt. When the agent writes code, it does so under the team's explicit rules: naming conventions, import order, error-handling patterns, comment density. The Security Policy Engine double-checks the output as a merge gate — linter failures become policy failures, not reviewer nitpicks. Cursor and Claude Code, via the Embedded SDK, pull the same style layer so behaviour is consistent across tools and engineers.

What this changes for reviewers

Review time shifts from style policing to logic review. The agent's output already matches house style, so the reviewer reads for correctness, design, and test coverage — the work that actually benefits from human judgment. New engineers stop getting style nitpicks in their first PRs; the agent applies the rules on their behalf. The team's style becomes a versioned artefact rather than tribal knowledge, and onboarding new tools or engineers to the same rules takes minutes instead of months of review feedback.

Relevant bRRAIn products and services

• Handler — applies style rules as constraints during every code generation.
• POPE Graph RAG / Style Layer — stores linter rules, exemplar PRs, and conventions as queryable nodes.
• Document Portal — landing zone for style guides and exemplar bundles.
• Security Policy Engine — gates merges on lint and style policy failures.
• Consolidator — keeps the style layer fresh as conventions evolve.
• Embedded SDK — propagates the style layer to Cursor, Claude Code, and any agent runner.