Prelint

Prelint sits next to your existing code review and looks at a layer most tools ignore: whether the code matches what the team actually agreed to build. It reads markdown specs, architecture decision records, and product requirement docs stored in your repo, then turns them into review criteria. When a PR contradicts one of those decisions, Prelint says so.

The problem it targets is specific. AI coding agents write clean, passing, well-structured code that still breaks your product rules. A pricing page that quotes 60-day terms when billing policy says 30, or a signup flow that adds a GDPR consent step for US-only customers. Your linter won't catch that. Your human reviewer might, if they have time and remember the policy. So what fills the gap when neither does?

The biggest limit is scope by design. Prelint doesn't replace your linter, type checker, or security scanner, and it won't fix the code for you. It reads context within a character budget and depends on the quality of the docs you feed it. Garbage specs in, garbage findings out.

Getting Started

1. Install the GitHub App and pick which repositories Prelint should review. No CI changes and no YAML are needed.
2. Add context to the repo by dropping product specs, decision logs, or architecture docs into it. Markdown files get indexed automatically.
3. Open a pull request. Prelint reviews it within seconds and posts findings as inline comments pinned to the lines that triggered them.
4. Read the decision ledger to see who decided what, when, and whether a person or an AI made the call. Approved decisions become context your agents reuse next time.