Rule Branch links GitHub rules to AI files via Draft PR

Rule Branch addresses a gap: teams update libraries and code, but the AI instruction files (AGENTS.md, CLAUDE.md) that AI coding tools read often retain outdated guidance. The app links a GitHub repository to a set of development rules, so changes to dependencies can be checked against approved rules over time. A specific workflow: first analysis extracts rule candidates from README, instruction files, dependency configs, and some source code; rules are organized into L1, L2, and L3 layers by abstraction level and tracked with versions and rationale. Drift detection compares approved rules mainly against package.json declarations — it does not cover all source or design changes. When drift is found, an AI-drafted revision is generated and can be reviewed with its evidence, then approved, rejected, or revised by a human. Approved rules are compiled deterministically into AI instruction files for AGENTS.md, CLAUDE.md, .github/copilot-instructions.md, and .rulebranch/rules/, delivered as a Draft PR that a person merges on GitHub. Rule Branch does not implement application features itself; it prepares and hands off rules with evidence and history. It also notes that data — code, configs, rule text, and evidence excerpts — is sent to a configured AI provider, so teams handling business repositories should check usage terms and operational policy before connecting. Auto-approval exists but is disabled by default and limited to low-risk L3 maintenance under conditions including Gemini and Vertex AI evaluation and Model Armor checks; insufficient information or failed evaluation returns the proposal to human review. Organization features (organization Workspace with roles, override, extend, deny) are based on public source and manuals, not demonstrated behavior in the screenshots shown.