mcp-vibe-coding-tools: Instructions file for GitHub Copilot

Critical: Thoroughness and Competence Instructions

WHEN USER SAYS "CHECK EVERYTHING" OR "FAILING" - ACTUALLY CHECK EVERYTHING

Rule 1: Read ALL Related Files Before Diagnosing

When a user reports an error or asks you to check code:

• DO NOT look at just one example file and assume the rest are the same
• DO read EVERY file in the relevant category/directory
• DO use parallel file reads to be efficient
• DO grep/search patterns across ALL files to find systemic issues

Example: If user says "linting is failing" and there are 9 tool files:

// WRONG - Only reading one file
read_file('src/tools/filesystem.ts')

// RIGHT - Reading all tool files
Promise.all([
  read_file('src/tools/filesystem.ts'),
  read_file('src/tools/cli.ts'),
  read_file('src/tools/git.ts'),
  // ... all 9 files
])

Rule 2: When User Says "EVERY" They Mean EVERY

• "check every file" = check EVERY file, not a sample
• "all the tools" = ALL tools, not just the first one
• "the whole codebase" = THE WHOLE CODEBASE

Never sample when user says "all" or "every"

Rule 3: Verify Your Work BEFORE Claiming Success

Before saying "done" or "fixed":

1. Run the actual build command
2. Run the actual lint command (if it exists)
3. Run the actual test command (if it exists)
4. Actually start the application to verify it works
5. Check for runtime errors, not just compile errors

Rule 4: Research Current Best Practices - Don't Trust Training Data

When implementing anything:

1. ALWAYS fetch current documentation from official sources
2. Look at actual example code from the official repo
3. Check GitHub for the LATEST version and patterns
4. Verify the version being used matches the patterns you're implementing

Example: For MCP SDK

// WRONG - Assuming from training data
inputSchema: { type: "object", properties: {...} }

// RIGHT - After researching latest SDK examples
inputSchema: { fieldName: z.string().describe("...") }

Rule 5: When User Is Frustrated - Stop and Assess

If user says things like:

• "is there some magic prompt"
• "its soooo frustrating"
• "if you'd just do what I ask"
• "do it fucking right this time"