Summary

After a recent update to claude-code-action@v1 (SHA fad22eb3fa582b7357fc0ea48af6645851b884fd), invoking a plugin skill via the prompt parameter no longer works. The skill installs successfully but fails to execute, and the model falls back to manual tool use for the rest of the session.

Workflow configuration

- uses: anthropics/claude-code-action@v1
  with:
    plugins: 'code-review@claude-code-plugins'
    plugin_marketplaces: 'https://github.com/anthropics/claude-code.git'
    prompt: '/code-review:code-review owner/repo/pull/123 --comment'

What happens

The plugin installs successfully and the skill appears in slash_commands, but execution fails:

✔ Successfully installed plugin: code-review@claude-code-plugins (scope: user)

{
  "name": "Skill",
  "input": { "skill": "code-review:code-review" },
  "tool_use_result": "Error: Execute skill: code-review:code-review"
}

The model then logs: "The skill invocation is failing, but I have the full review procedure from the command instructions. I'll execute it manually" — and proceeds without the skill.

Last known working SHA

a92e7c70a4da9793dc164451d829089dc057a464