Agent Scripts is a collection of shared instructions, skills, and helper scripts for AI agents used across local workspaces.
The project solves the problem of managing reusable agent capabilities and guardrails across multiple repositories and agent platforms. It provides a centralized canonical source for agent rules, workflow skills, and portable helper scripts. The core mechanism is a skill routing layer where each skill is defined in a SKILL.md file with YAML front matter describing its purpose and implementation. A sync script manages discovery and symlink topology across different agent platforms, handling name collision resolution and validating skill integrity before exposing them to agents like Codex and Claude Code.
The tool is designed for developers maintaining multiple projects that use AI agents and need consistent, shared skill definitions across those projects. It suits workflows where you want to define reusable agent capabilities once and expose them to multiple agent platforms without duplication. The project emphasizes keeping skill descriptions short and generic to optimize for routing rather than documentation, and provides validation tooling to catch errors after edits. It handles the specific discovery patterns of different agents—Codex scans nested directories while Claude Code requires a flat structure—by maintaining appropriate symlink mirrors and resolving collisions with a clear precedence order.
The project maintains strict validation rules for skill naming and structure, rejecting duplicates and unknown arguments during sync operations. It includes audit tooling to detect problematic symlink topologies like nested self-references. The sync script performs rechecks before each operation but does not provide atomic concurrency protection, stopping on errors and reporting any removals already completed. Test fixtures are available for isolated coverage of the sync behavior, and the project distinguishes between broad repair operations and scoped legacy topology handling.
