Agent Loops

Ship improvements while you sleep. Agent loops are GPTDiff's killer feature—a simple while true pattern that turns one-off fixes into continuous, autonomous code improvement.

Without Agent Loops	With Agent Loops
Manual code reviews	Automated 24/7 scanning
Reactive bug fixes	Proactive issue detection
Weekend tech debt sprints	Continuous improvement
One change per prompt	Hundreds of changes overnight

Real Results: Test Coverage Overnight

The Setup: A Python project with sparse test coverage. Run this loop before bed:

while true; do
  gptdiff "Add missing test cases for uncovered functions" --apply
  sleep 10
done

8 Hours Later:

Metric	Before	After
Test files	3	14
Test cases	18	127
Functions with tests	12%	71%
Edge cases covered	~5	40+

Each cycle, GPTDiff identified the next untested function, wrote appropriate test cases, and moved on. No manual intervention required.

Real Results: Security Hardening Sprint

The Setup: A Node.js API with authentication and database queries. Run this loop during a security sprint:

while true; do
  gptdiff "Find and fix OWASP Top 10 vulnerabilities" \
    --model deepseek-reasoner \
    --apply
  sleep 30
done

48 Hours Later:

Vulnerability Class	Before	After
SQL injection risks	8	0
XSS vulnerabilities	12	0
Missing input validation	23	2
Hardcoded secrets	3	0

Each cycle scanned for a different vulnerability pattern, applied fixes, and moved to the next issue. What would have taken a security consultant a week completed autonomously over a weekend.

---

AI-Powered Feedback Loops

Autonomous Code Improvement

Agent loops let GPTDiff work continuously on your codebase—finding issues, generating fixes, and applying them automatically. Each cycle makes your code better.

graph LR
    A[Codebase] --> B(GPTDiff Analysis)
    B --> C{Identify Improvement}
    C --> D[Generate Patch]
    D --> E[Apply Changes]
    E --> F[Verify]
    F --> A

Copy-Paste Starter

Run this tonight and check the results tomorrow:

while true; do
  gptdiff "Find and fix the most obvious code quality issue" --apply
  git add -A && git commit -m "Auto-improvement $(date +%H:%M)" 2>/dev/null
  sleep 30
done

This loop automatically commits each improvement, giving you a clean history of what changed and when. Start with 30-second intervals; adjust based on your API limits.

Test Enhancement Recipes

Missing Test Case Detection

# Continuously analyze test coverage gaps
while :
do
  gptdiff "Identify and add missing test cases for edge conditions" \
    --apply \
    --temperature 0.3
done

Improvements Per Cycle:

1. Null input handling tests
2. Boundary value validations
3. Error state simulations
4. Concurrent execution checks

Flaky Test Remediation

while true; do
  gptdiff "Find and fix intermittent test failures by adding retries/cleanup" \
    --model deepseek-reasoner \
    --apply
done

Code Quality Automation

Tech Debt Reduction Loop

while true
do
  # Prioritize worst code first
  gptdiff "Refactor functions with high complexity scores" \
    --apply \
    --temperature 0.1
done

Security Hardening Daemon

while :; do
  gptdiff "Find and fix OWASP Top 10 vulnerabilities" \
    --apply \
    --model deepseek-reasoner
done

Documentation Syncing

Code-Docs Alignment

while true;
do
  gptdiff "Update documentation to match current implementation" \
    --apply \
    --temperature 0.5
done

Performance Optimization

while true; do
  gptdiff "Identify and optimize slow database queries" --apply
done