vibeguard: Command for Claude Code

Core Concept

• Review is not to find faults, but to systematically verify code quality
• Stratified by priority: Security issues > Logic bugs > Code quality > Performance
• Each discovery comes with specific fix recommendations

Steps

1. Get guard baseline

   • Run bash guards/<language>/check_*.sh <target_dir> to get the current guard status
   • Record existing problems (no repeated reports)

2. Determine the scope of review

   • If a file path is specified: review the file
   • If a directory is specified: review recently modified files (git diff --name-only)
   • If no parameters: review the files in the current git staging area

3. P0 — Security Review

   • Refer to rules/claude-rules/common/security.md
   • Check OWASP Top 10 related questions
   • Check for key/credential leaks
   • Check input validation and sanitization

4. P1 — Logical Correctness

   • Boundary condition processing
   • Error handling integrity
   • Concurrency safety
   • Data consistency (multiple entry paths are consistent U-11~U-14)

5. P2 — Code Quality

   • Duplicate code detection (whether there is an existing implementation that can be reused)
   • Naming convention (refer to L2 naming constraints)
   • Exception handling (disable silent swallowing of exceptions L3)
   • File size (>800 line mark)
   • Refer to the corresponding language rule file

6. P3 — Performance

   • Performance issues on the hot path
   • N+1 query
   • Unnecessary memory allocation

7. Goal-Backward, borrowed from GSD)

   • From the user's perspective: What difference can users observe after these changes are completed?
   • Backward verification of third-level products:
     • L1 Existence: Do all declared/committed files and functions exist?
     • L2 Substantiveness: Is it a real implementation? Scan todo!(), unimplemented!(), empty function body, pass # and other stubs
     • L3 Wiring: Is the new code wired correctly? (Called, imported, covered by tests)
   • If L1/L2/L3 is found missing, report it as P1 logic problem

8. Output review report

   Markdown format (default):

   ## Review Report
   
   ### Guard the baseline
   <Summary of guard script results>
   
   ### Discover
   | Priority | File:line number | Question | Suggestion |
   |--------|-----------|------|------|
   | P0     | ...       | ...  | ...  |
   
   ### Passed items
   - <Confirm that there are no problems>
   
   ### suggestion
   - <Improvement suggestions (optional)>
   

   JSON format (optional, convenient for trend comparison of check command consumption):

   {
     "command": "review",
     "scope": "<review scope>",
     "findings": [
       {"priority": "P0", "file": "file:line", "issue": "...", "suggestion": "...", "ruleId": "RS-03"}
     ],
     "passedItems": ["No security holes", "..."],
     "verdict": "pass | warn | fail"
   }
   

   See docs/command-schemas.md for Schema details.

Guardrails

• It is not recommended to add unnecessary abstractions (L5)
• Adding a backward compatibility layer (L7) is not recommended
• When duplicate code is found, it is recommended to extend the existing implementation rather than create a new one (L1)
• AI generated tags are not included in review reports

Reference

• Security rules: rules/claude-rules/common/security.md
• Common rules: rules/claude-rules/common/
• Language rules: rules/claude-rules/<lang>/