What problem does it solve? Repositories often accumulate scattered, duplicated, or conflicting AI agent instruction files (AGENTS.md, CLAUDE.md, ad-hoc docs) that confuse coding agents and waste context. This Skill establishes a single canonical instruction hierarchy so Claude Code and Codex read consistent, well-scoped guidance. ## Core Features & Use Cases - Canonical AGENTS.md setup: Creates or refactors AGENTS.md as the sole repository-level instruction entry point, keeping only universal behavioural rules. - Scoped instructions and references: Moves path-scoped rules into .agents/instructions/ with non-global applyTo patterns and task-triggered knowledge into .agents/references/. - Legacy handling and validation: Detects obsolete CLAUDE.md files, hands cleanup to agent-instructions-audit, and validates that no global scopes, duplicates, or lost instructions remain. - Use Case: You onboard a new repository for AI-assisted development and want Claude Code and Codex to follow one consistent, minimal set of instructions without a CLAUDE.md compatibility shim. ## Quick Start Ask the agent to run agent alignment for this repository and confirm when prompted before any files are changed.
