agents-md

The `agents-md` skill is designed to streamline the creation and maintenance of concise AGENTS.md and CLAUDE.md files within software projects. It addresses the challenge of managing agent instructions and coding-agent workflows in repositories, ensuring that instructions remain actionable, minimal, and easy to reference. By standardizing the process of documenting agent commands and conventions, this skill prevents redundancy, reduces the risk of outdated instructions, and helps maintain clarity in collaborative environments where multiple agents or AI contributors are involved.

Key features of this skill include inspecting repository structures such as package managers, command definitions, CI workflows, and existing documentation before generating instructions. It enables users to create root-level or nested AGENTS.md files with repository-specific defaults while maintaining brevity (targeting under 60 lines). The skill supports precise path verification, structured tables for commands and external references, and integration with CLAUDE.md through symlinks when necessary. It emphasizes the inclusion of only essential sections, adherence to file-relative paths, and avoidance of duplicated content from other documentation sources. Furthermore, it enforces writing rules that prefer bullets, tables, and headings over paragraphs, maintaining a clear, reference-driven format.

Typical use cases for `agents-md` include creating initial AGENTS.md files, updating existing agent documentation, setting up CLAUDE.md entrypoints, and maintaining coding-agent instructions that are minimal yet fully reference-backed. Target users are repository maintainers, AI workflow integrators, and development teams who need to enforce consistent agent conventions across projects. This skill is particularly valuable in environments where multiple agents interact with the codebase, and precise, actionable guidance is critical for automated or semi-automated workflows.