AgentLint: The Single Source of Truth for AI

AI coding agents such as IBM Bob, Claude, and Copilot rely on repository instruction files like AGENTS.md, CLAUDE.md, and .github/copilot-instructions.md. As projects evolve, these files can become outdated or contradict each other. For example, a repository may migrate from npm to pnpm, while an old instruction still tells AI agents to use npm. The result is incorrect code, broken workflows, and wasted developer time. AgentLint solves this by acting as a linter for AI coding instructions. It automatically discovers agent instruction files and compares their rules with evidence from the actual repository, including lockfiles, package configurations, scripts, paths, and project settings. AgentLint then identifies stale instructions, contradictions, missing references, and conflicts between different agent files. Developers can run scans through the AgentLint CLI and review results in a visual dashboard showing findings, severity, instruction sources, and the repository evidence behind every detection. AgentLint also uses IBM Bob to understand natural-language instructions, analyze conflicts, suggest safe repairs, and help create a unified policy.yaml that acts as a single source of truth for AI coding agents. The target users are software developers and engineering teams using one or more AI coding assistants. What makes AgentLint different is that it does not try to replace coding agents. Instead, it improves the context they depend on. By keeping AI instructions synchronized with the actual repository, AgentLint helps every coding agent produce more reliable and consistent results.