1. Plaintext API keys in config files

   • Risk: Anyone with file access sees your keys
   • Mitigation: Keys stored encrypted in system keyring

2. Accidental git commits

   • Risk: Keys pushed to GitHub/GitLab
   • Mitigation: Keys never in files that get committed

3. Config file sharing

   • Risk: Sharing ~/.claude.json exposes keys
   • Mitigation: Keys injected dynamically, not stored permanently

4. Session-scoped credential injection (v1.1.0+)

   • Risk: Keys permanently in ~/.claude.json between sessions
   • Mitigation: Automatically removed when Claude Code exits
   • Window: Keys only accessible while you're actively using Claude

1. Malicious code with your permissions

   • Any process running as you can access your keyring
   • If malware has your login, it can extract keys

2. Memory dumps

   • Keys are in memory while Claude Code runs
   • Memory forensics can extract them

3. Physical access to unlocked machine

   • Keyring unlocks when you log in
   • Physical access = full key access

4. Claude Code itself

   • You're trusting Claude Code not to exfiltrate keys
   • Keys are available to MCP servers Claude spawns

• Use unique keys per machine - Don't reuse same key across systems
• Rotate keys regularly - Change every 90 days minimum
• Use least-privilege keys - Read-only when possible
• Revoke immediately if machine compromised

• Lock your screen when away from keyboard
• Encrypt your disk - Keyring is only secure on encrypted storage
• Keep system updated - Keyring security depends on OS security
• Use strong password - Keyring is encrypted with your login password

• Monitor API usage - Check for unexpected activity
• Enable audit logs - On services that support it (AWS CloudTrail, etc.)
• Set spending alerts - For paid APIs (OpenAI, etc.)

Understanding the security improvements across versions:

Before (v1.0.0):

• Keys injected at session start
• Keys remained in ~/.claude.json after Claude Code exited
• Risk window: Indefinite (until manual cleanup)

After (v1.1.0):

• Keys injected at session start
• Keys automatically removed at session end
• Risk window: Only while Claude Code is actively running

Recommendation: All users should upgrade to v1.1.0 for improved session-scoped security.

• Encryption: AES-256
• Unlocks: When you log in (PAM integration)
• Storage: ~/.local/share/keyrings/
• Persistence: Survives reboots if auto-login disabled

• Encryption: AES-256 (FileVault required for disk encryption)
• Unlocks: Login keychain unlocks with user password
• Storage: ~/Library/Keychains/
• Persistence: Very secure, tied to Apple ecosystem

• Encryption: DPAPI (Data Protection API)
• Unlocks: When logged in
• Storage: Registry + encrypted files
• Persistence: Tied to user account

If you discover a security vulnerability in this plugin:

DO NOT open a public GitHub issue.

Instead:

1. Email: [your-email]@[domain].com
2. Include: Description, reproduction steps, potential impact
3. Response time: 48 hours

Security reviews and contributions welcome from the community.