Feedback

When a skill produces unexpected output, the feedback command captures a structured report so the issue can be traced and fixed.

Using the Feedback Command

In a Claude Code session with the marketplace plugin installed:

/webplatform4sync:feedback

Claude will ask you to describe:

1. Which skill produced the unexpected behavior
2. What you expected to happen
3. What actually happened — the output, error, or wrong result
4. Your project context — stack, active theme, tenant tier

Where Reports Go

Feedback reports are written to .platform/feedback/ in your project directory. Each report is a timestamped JSON file:

.platform/feedback/
  2026-02-22T14-30-00-marketplace-plugin.json
  2026-02-22T15-45-00-theme-inspired-tokens.json

These files are local to your project. They are not transmitted anywhere automatically. You decide whether to include them in a GitHub issue or keep them for your own reference.

Report Format

Each report contains:

{
  "timestamp": "2026-02-22T14:30:00Z",
  "skill": "theme-inspired-tokens",
  "expected": "Generated tokens.css with all 12 color scales",
  "actual": "Missing neutral scale in output",
  "context": {
    "theme": "nihon-traditional",
    "stack": "vite-7, tailwind-4",
    "tier": "dormant"
  }
}

Why This Exists

Skills are AI-native instructions, not deterministic templates. The same skill can produce different output depending on project context, model behavior, and how the request is phrased. Structured feedback gives maintainers the context they need to reproduce and fix issues.

If you prefer, you can also open a GitHub issue directly.