The GenPark ContextDev Web Scraper Skill is a minimal implementation with severe quality deficiencies. Only one tool is defined (parse_html), and it lacks essential production-grade attributes. The tool description is 27 characters ('Mock regex tags stripping'), which barely meets the non-empty threshold but provides no actionable guidance on WHAT the tool does, WHEN to use it, or WHAT the agent should expect. The input schema is present but minimal, the 'raw_html' parameter has a description ('Raw HTML content to parse and clean'), but the tool offers no output schema documentation in code comments or docstrings. The implementation is a trivial regex-based HTML stripper with no error handling, validation, or recovery guidance. The tool name 'parse_html' uses the correct verb_noun pattern but the description fails to clarify what 'parsing' entails (is it extracting text? Links? Metadata?) or when the agent should prefer this tool over alternatives. No pagination, no output structure docs, no parameter constraints, no error recovery patterns are evident. This server reads as a proof-of-concept stub rather than a production-ready tool.

• Expand tool description to 50-150 characters. Explain: 'Parses raw HTML and strips tags to extract plain text. Use this when you need to clean article text, body content, or unstructured HTML before indexing or analysis. Returns extracted_text and optional metadata.',
• Document the output schema explicitly in a docstring or JSON comment. Specify: {'extracted_text': string, 'meta_description': string, 'extracted_tags': [string], 'total_chars': integer}. Let LLMs know what fields to expect.
• Add input validation and error handling. Check raw_html is not None and under 10MB. Catch regex exceptions and return a structured error: {'error': 'HTML parsing failed', 'reason': '<specific error>', 'recovery': 'Try splitting the HTML into smaller chunks.'}.
• Add parameter constraints to the input schema: set maxLength for raw_html (e.g., 10485760 bytes), add minLength of 1, and document the encoding (UTF-8). Include in the description: 'Must be valid UTF-8 HTML, max 10 MB.'
• Implement proper HTML parsing beyond regex. Consider using a lightweight library like html.parser (stdlib) or lxml to extract structure, links, headers, and metadata accurately. Return per-element data so agents can filter meaningfully.
• Remove hardcoded static responses. Compute meta_description from actual content (first non-empty paragraph, or first 160 chars of text) so the response reflects what the tool actually extracted.
• Add tool annotations (readOnlyHint, idempotentHint) once you upgrade to a proper MCP server framework that supports them.