Please do not open public issues for security vulnerabilities.

If you discover a security vulnerability in Argus, please email security@argus.local with:

1. Description of the vulnerability
2. Steps to reproduce
3. Potential impact
4. Suggested fix (if you have one)

We will:

• Acknowledge receipt within 48 hours
• Work on a fix and release a patch
• Credit you in the release notes (unless you prefer anonymity)

Argus is designed with a transparent, local-first security model:

• Monitors running processes on your machine
• Tracks which files are accessed by AI applications
• Logs network connections made by AI applications
• Sends native OS notifications when alerts are triggered
• Stores all data in a local SQLite database

• No network requests — All monitoring happens locally
• No telemetry — No data is sent to Anthropic, OpenAI, or any third party
• No cloud sync — All data stays on your machine
• No shell commands executed — Argus only reads process state; it doesn't modify anything
• No privilege escalation — Runs as a regular user (LaunchAgent/systemd)

All monitoring data is stored in:

~/.argus/data.db          # SQLite database (main event log)
~/.argus/logs/daemon.log  # Daemon logs (rotating, max 10 MB)

These files are:

• Local only — Never synced to cloud or sent anywhere
• User-readable — Belong to your user account, not root
• Inspectable — You can query the database with sqlite3
• Deletable — You can safely delete ~/.argus/ at any time

All code is in the src/ directory. To verify Argus doesn't make network calls:

# Search for network libraries
grep -r "http\|https\|fetch\|request" src/

# Expected: 0 results (Argus doesn't initiate outbound connections)

Inspect all npm dependencies:

npm ls

All dependencies are production-ready, open-source packages:

• better-sqlite3 — SQLite driver
• chokidar — File system watcher
• node-notifier — Native OS notifications
• pino — Logging
• ps-list — Process listing
• Others are utilities with no network access

cat src/notifications/notifier.js

Notifications are sent via:

• macOS: node-notifier (uses native macOS notification API)
• Linux: node-notifier (uses D-Bus)

No network calls are made.

# Check for hardcoded API keys
grep -r "sk-" src/
grep -r "api-key" src/
grep -r "token" src/ | grep -v "// " | grep -v "comment"

# Should return nothing

# Check that all database queries use parameterized statements
grep -r "prepare\|all\|run" src/db/

# All queries should use `?` placeholders, never string concatenation

Example of safe query:

db.prepare('SELECT * FROM events WHERE app_name = ?').all(appName);

Example of unsafe query (DO NOT USE):

// ❌ WRONG — vulnerable to SQL injection
db.prepare(`SELECT * FROM events WHERE app_name = '${appName}'`).all();

• Argus runs as a LaunchAgent in ~/Library/LaunchAgents/com.argus.daemon.plist
• Does NOT require sudo or root access
• Can read from standard user directories (Documents, Downloads, Desktop)
• Uses launchctl to manage the service
• Respects macOS System Integrity Protection (SIP) and code-signing restrictions

TCC (Transparent User Consent):

• Argus may request Full Disk Access if you explicitly grant it via System Preferences
• This is optional — Argus will still work without it, but with reduced file monitoring

• Argus runs as a systemd user service
• Does NOT require sudo or root access
• Can read from user home directory (~)
• Uses systemctl --user to manage the service

argus install
# Prompts for:
# - macOS user password (to write LaunchAgent plist)
# - Optional: Full Disk Access permission (for enhanced file monitoring)

argus install
# No password required (systemd user service)
# Runs under your own user account

Argus monitors what these applications do — it does NOT modify their behavior or intercept their network traffic. You should:

1. Review each AI app's own privacy policy:

   • Claude: https://www.anthropic.com/privacy
   • ChatGPT: https://openai.com/policies/privacy-policy
   • Cursor: https://www.cursor.com/privacy

2. Understand what data each app sends to its vendor:

   • Claude sends prompts to Anthropic's servers
   • ChatGPT sends prompts to OpenAI's servers
   • Cursor may send code snippets to Codeium
   • Argus cannot stop this — it only monitors and alerts you

3. Use Argus to verify expected behavior:

   • "Did Claude really only connect to api.anthropic.com?"
   • "Why is Cursor connecting to an unfamiliar domain?"
   • "Which files did the AI app read?"

1. File Access Monitoring — Argus uses file system events to detect access. Some accesses may not trigger events (e.g., memory-mapped files, in-process reads from already-open files).

2. Network Monitoring — Argus uses lsof to enumerate connections. This has a small race condition window where very fast, short-lived connections might not be captured.

3. Process Classification — The 6-signal AI detection engine has false positive/negative rates. High-confidence signals (network endpoints, code signing) are reliable; lower-confidence signals (keywords, pipes) can vary.

4. TCC Database — macOS TCC database queries require reading a privileged SQLite file. This may require Full Disk Access, which is optional.

If you discover a vulnerability:

1. Do not open a public GitHub issue
2. Email security@argus.local with details
3. Allow 90 days for a patch before public disclosure
4. We will credit you in release notes (unless you prefer not to)

Examples of security issues:

• SQL injection in database queries
• Hardcoded API keys or credentials
• Unvalidated file paths leading to path traversal
• Insecure deserialization
• Privilege escalation vulnerabilities

Argus respects:

• macOS: System Integrity Protection (SIP), Gatekeeper, code signing
• Linux: SELinux, AppArmor (if installed), standard user permissions
• GDPR: No personal data collection, transmission, or storage on external systems
• Data Privacy: All monitoring data is yours — you can delete it anytime

Subscribe to releases to be notified of security patches:

# Watch the repository for releases only
# Settings → Notifications → Only releases

Or check the releases page regularly:

https://github.com/yourusername/argus/releases

Email security@argus.local or open a discussion on GitHub (for non-sensitive topics).

Transparency is security. All Argus code is visible, reviewable, and local-only.