​

RadCheck

​

What RadCheck Does

• Stall risk — Patterns that indicate potential hangs or infinite loops
• Silence gaps — Unexpected periods of inactivity that may indicate failures
• Compaction pressure — Context window utilization and memory patterns
• Hygiene — Configuration best practices and common misconfigurations

​

Running RadCheck

acme radcheck

RadCheck Reliability Scan
=========================
Overall Score: 72 / 100 (YELLOW)

Components:
  Stall Risk:        18/25  ✓
  Silence Gaps:      15/25  ⚠
  Compaction:        22/25  ✓
  Hygiene:           17/25  ⚠

Findings:
  [WARN] Silence gap detected: 47s between context updates
  [WARN] Missing retry configuration in provider fallback
  [OK]   No stall patterns detected
  [OK]   Context utilization within bounds

​

Score Bands

​

When to Run RadCheck

• Before deploying a new agent configuration
• After incidents to identify contributing factors
• Periodically as part of reliability hygiene (weekly recommended)
• When onboarding to get a baseline score

​

From RadCheck to Sentinel

• RadCheck = Point-in-time snapshot (like a blood test)
• Sentinel = Continuous monitoring (like a heart monitor)

​

CLI Reference

# Basic scan
acme radcheck

# JSON output (for CI/CD integration)
acme radcheck --format json

# Scan specific agent
acme radcheck --agent my-agent-name

# Verbose output with all checks
acme radcheck --verbose

​

Next Steps