Code Review Cycle

Iteratively review local changes, fix issues, and re-review until clean.

Prerequisites

Verify shavakan-agents:code-reviewer is installed:

if ! claude plugins list 2>/dev/null | grep -q "shavakan-agents"; then
  echo "MISSING: shavakan-agents plugin required"
  echo "Install: claude plugins add shavakan-agents@shavakan"
  exit 1
fi

---

Additional Instructions

$ARGUMENTS

---

Execution

Initialize

Track iterations:

• iteration_count = 0
• issues_by_iteration = []
• max_iterations = 5

Review Loop

Repeat until no actionable issues or max_iterations reached:

1. Run Review

   • Increment iteration_count
   • Launch shavakan-agents:code-reviewer agent via Task tool
   • Parse findings: 🔴 Critical, 🟠 High, 🟡 Medium

2. Check Exit Condition

   • If no findings at any priority → exit loop, report success
   • If iteration_count >= max_iterations → exit loop, report remaining issues

3. Record Issues

   • Store findings in issues_by_iteration[iteration_count]
   • Include: file:line, type, description, severity

4. Fix Issues

   • Address all actionable issues (🔴🟠🟡) in priority order
   • Use Edit tool for code fixes
   • Stage changes: git add -A

5. Continue Loop

Report Summary

Output structured iteration report:

## Code Review Summary

**Iterations**: {iteration_count}
**Final Status**: {Clean | {N} issues remaining}

### Iteration Details

#### Iteration 1
- [file:line] type - description → Fixed
- [file:line] type - description → Fixed

#### Iteration 2
- [file:line] type - description → Fixed

### Statistics
- Total issues found: N
- Issues fixed: M
- Time per iteration: ~Xs average

---

Constraints

• Max 5 iterations to prevent infinite loops
• Stop immediately if agent reports "No critical or high-priority issues found"
• Don't fix style-only issues unless explicitly flagged as actionable
• Each iteration must have at least one fix to continue (no empty iterations)

---

Error Handling

• Agent unavailable: Exit with install instructions
• Max iterations reached: Report remaining issues, recommend manual review
• Git dirty state: Warn but continue (agent handles this)
• Fix breaks tests: Rollback fix, mark issue as "requires manual intervention"