init

To optimize AGENTS.md files, apply a discoverability filter that includes only non-discoverable, operationally significant, and actionable guidance for AI agents. This prevents stale or generic instructions from confusing agents during software development.

Context engineering for AI agent instructions involves filtering out discoverable repository information to focus on crucial, non-obvious project conventions, tooling quirks, and workflow preferences that agents need to perform tasks effectively.

Maintain AGENTS.md files by pruning stale or generic instructions and verifying each remaining instruction against current repository context and project workflows, ensuring the file remains concise, accurate, and relevant for agile maintenance.

You need to set up an AGENTS.md file when initializing a new project if agents require specific coding standards and tooling choices that are not readily evident from the codebase itself to avoid preventable mistakes.

Yes, the init skill works with existing repository setups by updating the current AGENTS.md file, applying a discoverability filter to remove generic instructions, and verifying operational significance against existing project workflows.

Your AI agent keeps making mistakes because the AGENTS.md file may contain overly long, generic, or stale instructions. Optimizing the file to include only non-discoverable, actionable guidance verifies accuracy against repository context and reduces agent confusion.