agentchecker is a command-line tool for resolving rule conflicts between instruction files for AI coding assistants. Its core problem is "rule drift": when you use multiple AI coding tools at the same time, each tool reads a different instruction file, for example Cursor reads .cursorrules or .cursor/rules/*.mdc, Claude Code reads CLAUDE.md or .claude/CLAUDE.md, GitHub Copilot reads .github/copilot-instructions.md, while humans and shared agents read AGENTS.md. Over time, these rules gradually diverge, for example Cursor starts using pnpm and eslint, Claude defaults to npm and oxlint, Copilot reformats everything with biome, ultimately leading to conflicting generated content, compilation errors, and wasted API tokens.
Purpose and main capabilities: agentchecker is a zero-dependency, ultra-fast command-line tool that can scan project-local agent files as well as global tool configurations (such as ~/.claude/, ~/.cursor/rules/, ~/.codex/, etc.), extract and parse technology stack decisions (including package managers, linters, formatters, test runners, shell environments, etc.), and then highlight conflicts across all sources with a clear terminal-based diagnostic report. It supports interactive fixes, and only modifies local project files; global configurations are always read-only. If no configuration exists locally, it automatically creates AGENTS.md to lock in the selected configuration for this project.
Usage: no installation is required; simply run it with npx in the project root directory, with no configuration file, global package, or environment setup needed. It can be integrated into daily workflows: add the check to .husky/pre-commit or a Git hook runner to scan project files and terminate the commit with exit code 1 when conflicts are found (omitting the --local-only option also includes global tool configurations under the user's home directory); you can also use the official GitHub Action or run it in CI directly via npx, using --local-only to validate only repository files rather than each developer's global configuration.
Supported tools and instruction files: this tool unifies the configurations of the industry's most popular AI coding assistants, editors, and agents. The types of decisions covered include package managers (pnpm, npm, yarn, bun), linters (oxlint, eslint, biome), formatters (prettier, biome, dprint), test runners (vitest, jest, playwright, cypress), package runners (npx, pnpm dlx, bunx, yarn dlx), response languages (spanish / español, english / inglés), and shell environments (wsl, wsl2, windows, macos, linux).
Applicable scenarios and limitations: suitable for developers and teams that use multiple AI coding assistance tools at the same time and want to keep instructions consistent and avoid rule drift, especially for consistency checks before commits and in CI workflows. Currently it can only add tools that store instructions in on-disk files; settings that exist only in the IDE (such as the Copilot global UI, Cursor configuration profiles without files, etc.) cannot be reliably scanned. To add support for a new tool, you can submit a support request, which must include the tool name, project rule path (exact file or glob), global path (if any), format (Markdown, JSON, YAML, TOML, etc.), official documentation link, and example snippet; PRs are also welcome to add patterns in packages/cli/src/scan/known-files.ts and add fixtures under packages/cli/tests/fixtures/. Details such as command-line options, local development instructions, and complete licensing information (MIT License) are not specifically provided in the current materials.
