testdino-mcp: Skill for Claude Code

Review $ARGUMENTS for bugs, correctness, and convention violations.

If $ARGUMENTS is "staged" or empty, review all staged/unstaged git changes.

Process

1. Read CLAUDE.md — load project rules and conventions
2. Read the target code — understand before judging
3. Check against each category below — report only real issues with file:line references

Categories

MCP Protocol Compliance (highest priority)

• Does the tool return the correct response shape? { content: [{ type: "text", text: string }] }
• Is console.log used anywhere? (corrupts stdio — must be console.error)
• Does the tool name match the routing in index.ts?
• Is the tool registered in both the tools array AND the routing if-blocks?
• Does inputSchema accurately describe what the handler expects?

Tool Pattern Consistency

• Does the tool follow the same pattern as existing tools?
• Auth: getApiKey(args) called first with proper error?
• URL: Built via endpoints.ts, not inline?
• Response: Formatted with JSON.stringify(response, null, 2)?
• Error: Wrapped in try/catch with contextual message?
• Types: Consistent use of String()/Number() for conversions?

Data Integrity

• Can any code path lose the API response silently?
• Are required parameters validated before the API call?
• Could a missing optional param cause the URL to be malformed?
• Are boolean params correctly converted to strings for query params?

Tool Descriptions (critical for AI agent usability)

• Is the tool description specific enough for an AI to choose correctly?
• Does every inputSchema property have a description?
• Are enum values provided for fields with known options?
• Are required fields accurately marked?
• Would an AI know when NOT to use this tool based on the description?

Security

• Is the PAT exposed in any error message or response?
• Are user inputs passed to URLs without sanitization?
• Are file paths validated before reading (for attachment tools)?