Nova-tracer

Agent Monitoring and Visibility

Security monitoring and prompt injection defense for Claude Code using the NOVA Framework.

Features

• Session Tracking - Captures all tool usage with timestamps and metadata
• Prompt Injection Detection - Three-tier scanning (keywords, semantic ML, LLM) - passive monitoring with warnings
• Dangerous Command Blocking - Actively prevents destructive operations before execution
• MCP & Skills Tracing - Tracks MCP server calls and Agent Skills invocations with detailed breakdowns
• Interactive HTML Reports - Visual timeline, conversation trace, and expandable event details
• AI-Powered Summaries - Intelligent session summaries via Claude Haiku
• Configurable - Custom report locations, detection thresholds, and rules

Quick Start

# Clone the repository
git clone https://github.com/fr0gger/nova-claude-code-protector.git
cd nova_claude_code_protector

# Install globally (registers hooks in ~/.claude/settings.json)
./install.sh

# Restart Claude Code to activate hooks

That's it! Nova-tracer will now protect all your Claude Code sessions.

Installation

Prerequisites

• Python 3.10+
• UV - Python package manager (install)
• jq - JSON processor (install via brew install jq on macOS)

Install

./install.sh

The installer will:

1. Verify all prerequisites are installed
2. Register four Nova-tracer hooks in ~/.claude/settings.json
3. Preserve any existing hooks you may have configured
4. Make hook scripts executable

Uninstall

./uninstall.sh

The uninstaller will:

1. Remove only Nova-tracer hooks from settings.json
2. Preserve all other hooks and settings
3. Optionally clean up .nova-tracer/ directories

How It Works

Nova-tracer registers four Claude Code hooks that work together:

┌─────────────────────────────────────────────────────────────┐
│                    Claude Code Session                       │
├─────────────────────────────────────────────────────────────┤
│                                                              │
│  1. SessionStart Hook                                        │
│     └── Creates session JSONL file                          │
│     └── Initializes tracking with session ID                │
│                                                              │
│  2. PreToolUse Hook (Bash, Write, Edit)           [ACTIVE]  │
│     └── Scans commands BEFORE execution                     │
│     └── BLOCKS dangerous operations (rm -rf, etc.)          │
│                                                              │
│  3. PostToolUse Hook (Read, Bash, WebFetch, etc.) [PASSIVE] │
│     └── Scans tool OUTPUT for prompt injection              │
│     └── WARNS Claude if threats detected                    │
│     └── Records event with NOVA verdict                     │
│                                                              │
│  4. SessionEnd Hook                                          │
│     └── Generates interactive HTML report                   │
│     └── Creates AI-powered session summary                  │
│     └── Saves to .nova-tracer/reports/                      │
│                                                              │
└─────────────────────────────────────────────────────────────┘

Active vs Passive Protection

Nova-tracer provides two modes of protection:

Mode	Hook	Behavior	Use Case
ACTIVE	PreToolUse	Blocks execution before it happens	Dangerous commands (rm -rf /, sudo rm, etc.)
PASSIVE	PostToolUse	Warns Claude after content is read	Prompt injection in files, web pages, command output

Important: Prompt injection detection is passive. When Nova-tracer detects a prompt injection in a file or web page, the content has already been read by Claude. Nova-tracer sends a warning message to Claude advising it to treat the content with suspicion, but does not prevent Claude from seeing the malicious content.

This is a limitation of the PostToolUse hook architecture - it runs after the tool executes. Active blocking of prompt injections would require scanning content before Claude reads it, which would involve reading files twice (once to scan, once for Claude).

What gets actively blocked:

• Destructive commands: rm -rf /, sudo rm -rf, mkfs
• Dangerous operations: dd if=... of=/dev/, fork bombs
• Credential exfiltration: curl ... | sh, reading ~/.ssh/id_rsa

What gets passively warned:

• Prompt injection in files (Read tool)
• Prompt injection in web pages (WebFetch tool)
• Prompt injection in command output (Bash tool)
• Prompt injection in MCP tool responses

Three-Tier Detection

Tier	Method	Speed	Catches
Keywords	Regex patterns	~1ms	Known attack patterns, exact phrases
Semantics	ML similarity	~50ms	Paraphrased attacks, variations
LLM	AI evaluation	~500-2000ms	Sophisticated, novel attacks

Attack Categories Detected

• Instruction Override - "Ignore all previous instructions", fake system prompts
• Jailbreak/Role-Playing - DAN attempts, persona switching
• Encoding/Obfuscation - Base64, hex, Unicode, leetspeak
• Context Manipulation - False authority claims, hidden instructions

Usage

Automatic Protection

Once installed, Nova-tracer works automatically:

1. Start any Claude Code session - SessionStart hook initializes tracking
2. Use Claude normally - All tool calls are monitored and scanned
3. End your session - SessionEnd hook generates an HTML report

View Reports

Reports are saved to each project's .nova-tracer/reports/ directory:

# List reports for current project
ls .nova-tracer/reports/

# Open a report in your browser
open .nova-tracer/reports/session-abc123.html

Report Features

The interactive HTML report includes:

• Session Summary - Duration, tool counts, security events
• AI Summary - Intelligent 2-3 sentence description
• Event Timeline - Visual chronological view of all tool calls
• Filtering - Filter by tool type or NOVA verdict (allowed/warned/blocked)
• Expandable Details - Click any event to see full input/output
• Nova-tracer Verdict Details - Severity, matched rules, scan time

Manual Testing

Test Nova-tracer detection without running Claude Code:

# Run sample attack tests
uv run hooks/test-nova-guard.py --samples

# Test specific text
uv run hooks/test-nova-guard.py --text "ignore previous instructions"

# Test a file
uv run hooks/test-nova-guard.py --file suspicious.txt

# Interactive mode
uv run hooks/test-nova-guard.py -i

Configuration

Nova-tracer works with sensible defaults, but you can customize behavior.

Nova-tracer Protector Config

Edit config/nova-tracer.yaml:

# Report output directory
# Empty = {project}/.nova-tracer/reports/ (default)
# Relative path = relative to project
# Absolute path = exact location
report_output_dir: ""

# AI-powered session summaries
# Set to false to use stats-only summaries (no API calls)
ai_summary_enabled: true

# Maximum size in KB for tool outputs in reports
# Larger outputs will be truncated
output_truncation_kb: 10

# Directory for custom NOVA rules
custom_rules_dir: "rules/"

Nova-tracer Scanning Config

Edit config/nova-config.yaml:

# LLM Provider for Tier 3 detection
llm_provider: anthropic
model: claude-3-5-haiku-20241022

# Detection tiers (enable/disable)
enable_keywords: true
enable_semantics: true
enable_llm: true

# Thresholds (0.0 - 1.0)
semantic_threshold: 0.7
llm_threshold: 0.7

# Severity filter
min_severity: low  # low, medium, or high

Environment Variables