CAGENTS – Context-aware AGENTS.md from templates (github.com)

cAGENTS is a new npm tool (alpha) that composes context-aware agent instruction files from small templates instead of maintaining multiple static AGENTS.md/CLAUDE.md/.cursorrules files. You initialize with npx cagents init, author Markdown templates with YAML frontmatter in .cAGENTS/templates/, and build outputs with npx cagents build. Templates can include metadata like name, order, when (conditions), variables (shell checks), globs, and output targets, letting a single source produce different formats (agents-md, claude-md, cursorrules) and package-scoped outputs in monorepos. It also offers migrate, lint, preview, render, and context commands, guidance to auto-rebuild on install (postinstall/prepare scripts), and a helper to add generated files to .gitignore. For AI/ML engineering teams that maintain multiple agent instruction sets (human-facing rules, model-specific prompts, CI/cloud vs local agent behavior), cAGENTS reduces duplication, drift, and merge conflicts by making rules conditional and reproducible. Technical implications include environment-aware inclusion via shell-evaluated variables (e.g., check for bd), ordered composition, multi-format rendering, and support for custom template engines (Handlebars/Jinja/Liquid). Because templates are version-controlled and outputs are rebuilt per environment, teams can enforce consistent agent behavior across cloud agents, local dev, and model-specific runtimes — but expect rough edges and breaking changes while the project is in alpha.