Claude Code Setup¶

Get CloneGuard running with Claude Code in under 5 minutes.

Prerequisites¶

Python 3.11+
Claude Code installed and working

Install¶

pip install "cloneguard[mini]"

This installs CloneGuard with the semantic classifier (recommended). Without [mini], only pattern matching is active.

Configure Hooks¶

cloneguard init --global

This writes hook entries into ~/.claude/settings.json for three events:

Hook Event	What It Does
`PreToolUse`	Gates writes, builds, and config changes before execution
`PostToolUse`	Scans tool output for injected instructions
`InstructionsLoaded`	Scans CLAUDE.md and rules files when loaded

Verify¶

Open Claude Code in any repository:

claude

CloneGuard scans transparently. When a detection fires, you'll see output in the hook's stderr. To confirm hooks are registered, check your ~/.claude/settings.json for CloneGuard hook entries.

Project-Level Hooks¶

For per-project configuration instead of global:

cd /path/to/project
cloneguard init --project

This writes hooks to .claude/settings.json in the project directory.

Trust Cache¶

For repositories you scan repeatedly, enable the trust cache to skip unchanged files:

cloneguard init --global --trust-cache

The cache stores SHA-256 content hashes in ~/.cloneguard/trust-cache.json. Any file modification -- even a single byte -- invalidates the entry.

False Positives¶

Security documentation and research files may trigger detections. Allowlist reviewed files by content hash:

cloneguard allow README.md --reason "Documents attack patterns for reference"

The allowlist lives in ~/.cloneguard/allowlist.json -- outside the repository, so repo content cannot tamper with it. If the file changes, the hash no longer matches and it's scanned normally.

What Happens When Something Is Detected¶

By default, CloneGuard operates in detection-only mode:

HIGH severity findings exit with code 2, which tells Claude Code to block the tool call
MEDIUM severity findings log a warning but allow the tool call
LOW severity findings are logged only

To configure enforcement (sandboxing, blocking), see the policy engine guide.

Uninstall¶

Remove CloneGuard hook entries from ~/.claude/settings.json (or the project-level .claude/settings.json), then remove the package:

pip uninstall cloneguard