Overview

agent-skills-toolkit is two things at once:

• The Standard - a normative spec (the Advanced Skill Library Standard) for what a conformant skill plugin looks like: its components, the conformance tiers that make it portable across agents, and the CI and release discipline that keeps it healthy at scale.
• The toolkit that builds and grades against it - a catalog of askit-* builder skills plus a deterministic validation spine that any plugin (including this one) runs to learn its tier.

The problem it solves

A folder of skills is not a library. As a collection grows, the questions that matter are no longer “is this one skill good?” but “does the whole thing hang together?” - do the components agree with the manifest, do the chains resolve, do deprecated components carry a migration path, does it work the same on a second agent? The toolkit answers those questions mechanically and tells you the tier your plugin earns.

The wedge

Cross-agent emission alone is crowded. The defensible position is the combination of three things:

1. Library-scale grading. Most tools assess a skill in isolation. This grades the plugin as a system and reports the tier it actually satisfies.
2. A deterministic, CI-safe gate. Objective conformance is checked by portable scripts with no model in the loop, so the grade can be trusted and enforced.
3. Honest cross-agent support. It emits for Claude Code and Codex and encodes each agent’s real limits, so a plugin cannot claim portability it does not have.

How you use it

• Point askit-migrate at an existing repo to get a staged Bronze-to-Silver plan.
• Run askit-evaluate to see your tier and exactly what blocks the next one.
• Use the askit-build-* skills to author each missing component, conformant by construction.
• Let the gate (node scripts/check.mjs) travel with the repo and enforce it in CI.

See Getting started, the Standard, and the catalog.