AI agents are now a primary reader of your documentation: a coding agent opens AGENTS.md before it touches your repo, and a browsing agent reads llms.txt before it uses your site. Both are young conventions, and most are missing, vague, or padded with words that waste the very context window they are meant to fill. This skill writes them properly and reviews existing ones against what an agent actually needs. For AGENTS.md it covers runnable build, test, and run commands, code conventions, a map of where logic lives, and the gotchas that trip agents up. For llms.txt it produces the correct structure — H1 name, summary, and curated, linked sections an agent can navigate. It has a write mode that drafts the file ready to commit, and a review mode that returns a PASS or REVISE verdict with a prioritized fix list, including a context-efficiency pass that cuts the filler tokens agentic engineers warn about. Works with Claude Code, Cursor, Codex CLI, Gemini CLI, and any SKILL.md agent.