Keep AI Agent Config Files Lean by Indexing Conventions, Not Pasting Them

Developers using AI coding agents often bloat their AGENTS.md files with hundreds of lines of rules, consuming valuable LLM context tokens on every single task. A more efficient approach is to keep AGENTS.md as a slim index that references separate documentation files, so the agent only loads the specific conventions relevant to each task. For teams using multiple agents like Claude Code alongside others, a symlink between CLAUDE.md and AGENTS.md ensures both tools always read from a single source of truth, preventing rule drift. In monorepo setups, placing an AGENTS.md in each subfolder allows agents to inherit shared rules while also picking up stack-specific guidelines. The core principle is treating AGENTS.md as a context budget rather than full documentation, keeping rules detailed without inflating per-task token costs.