Avocetta — Testing skill for Claude Code

avocetta [](https://github.com/korya/avocetta/actions/workflows/ci.yml) [](https://github.com/marketplace/actions/avocetta-lint)

avocetta is a deterministic linter for [agent skills](https://agentskills.io/specification) and [agent plugins](https://agent-plugins.org/). It verifies compliance with the open specs and with what specific runtimes (Claude Code, Codex) actually enforce, so you catch incompatibilities in CI before your users do.

It is plain static analysis: it never calls a model or the network at lint time, and it works without a config file.

• Four compliance dialects, each versioned and frozen: the agentskills.io and agent-plugins.org specs, Claude Code, and Codex
• Auto-detection of what you point it at (a skill, a plugin, a marketplace repo) and of which dialects your layout targets
• Union runs: lint against several runtimes at once, with guidance when their layouts seem to conflict
• Vendor rules backed by verified runtime behavior, with sources cited in every dialect file
• Text, JSON, SARIF, and GitHub-annotation output
• A zero-config GitHub Action and an installation-free CLI (npx avocetta)

GitHub Action

- uses: korya/avocetta@v2

That's it. The action detects what your repo is (a skill, a plugin, a marketplace of plugins), picks the right compliance targets (the spec dialects always; Claude Code and Codex when your layout shows you target them), annotates PR diffs inline, and fails the job on errors.

Optional inputs mirror the CLI flags:

- uses: korya/avocetta@v2
  with:
    path: plug