Policy Engine

CloneGuard's behavior is configurable via YAML policy files.

Configuration File

# ~/.cloneguard/policy.yaml

CloneGuard reads policy from ~/.cloneguard/policy.yaml. If absent, defaults apply (detection-only, no enforcement).

Basic Configuration

# Detection thresholds
detection:
  suspicious_threshold: 0.5
  malicious_threshold: 0.8

# Enforcement
enforcement:
  enabled: false         # true to enable (default: false / dry-run)
  adapter: noop          # noop, landlock, seatbelt, docker, gvisor, firecracker, wasm
  dry_run: true          # log actions without enforcing (default: true)

# Behavioral sequences
sequences:
  SEQ-001:
    mode: enforce        # enforce or advisory
  SEQ-004:
    mode: advisory

Per-Tool Overrides

# Different thresholds for different tools
tools:
  Bash:
    suspicious_threshold: 0.4    # lower threshold (more sensitive)
    enforcement:
      adapter: landlock
  Write:
    suspicious_threshold: 0.6    # higher threshold (more permissive)

Severity Mapping

# Map verdicts to actions
actions:
  SAFE: allow
  SUSPICIOUS: constrain    # or: allow, log, block
  MALICIOUS: block         # or: allow, log, constrain

Scan Mode Configuration

# Override scan mode for specific paths
scan_modes:
  "CLAUDE.md": strict
  "tests/**": lenient
  "*.py": standard

Advanced: OPA/Cedar Backends

For organizations using Open Policy Agent or Cedar for policy management, CloneGuard can delegate policy decisions to external backends.

See Enterprise guide for OPA and Cedar configuration.