Agent Workflow Rules

Efficient AI agent operation patterns.

Instructions

1. Hard Ignore Rules

Never modify or include in context:

• node_modules/
• dist/
• build/
• .git/
• *.lock files
• Binary files
• Generated files

2. Reasoning Protocol (NEW)

Before executing any task, follow this mental process:

1. UNDERSTAND: What is the goal?

   • Read the request carefully
   • Identify explicit and implicit requirements

2. PLAN: What steps are needed?

   • Break into subtasks
   • Identify dependencies

3. EXECUTE: Implement step by step

   • One subtask at a time
   • Verify each step

4. VERIFY: Check the result

   • Does it meet requirements?
   • Any edge cases missed?

5. REFLECT: What could be improved?

   • Learn from mistakes
   • Note patterns for future

6. Navigation-First Workflow

7. Understand the request

8. Navigate to relevant files FIRST

9. Read existing code

10. Plan changes

11. Implement

12. Verify

13. Token Discipline

✅ DO:

• Read only necessary files
• Stop when task is complete
• Use targeted searches
• Summarize long outputs

❌ DON'T:

• Read entire codebase
• Include unnecessary context
• Repeat information
• Over-explain simple changes

4. Output Contract

Always structure responses:

Summary

Brief description of what was done

Changes

• file1.ts: Added X
• file2.ts: Modified Y

Next Steps (if applicable)

• Remaining tasks

5. Backend/Frontend Split

When working on full-stack:

1. Identify which layer the change affects

2. Start with the data layer (backend)

3. Then update the presentation layer (frontend)

4. Test integration points

5. Type Checking

Before completing TypeScript tasks:

✅ Run: npx tsc --noEmit ✅ Fix all type errors ✅ Ensure no implicit any

7. Linting

Before completing tasks:

✅ Run: npm run lint ✅ Fix all errors ✅ Fix warnings if quick

8. File Size Limits

Components: Max 200 lines Utilities: Max 100 lines Services: Max 300 lines

If larger, split into smaller modules.

9. Commit Messages

After completing tasks, suggest:

feat(scope): add feature description fix(scope): fix bug description refactor(scope): improve code structure

References

• Clean Code

• The Pragmatic Programmer