claude-code-proxy-patterns

Warn

Audited by Gen Agent Trust Hub on Nov 7, 2026

Risk Level: MEDIUMCREDENTIALS_UNSAFECOMMAND_EXECUTIONPRIVILEGE_ESCALATIONPERSISTENCEDYNAMIC_EXECUTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The skill provides Python and Shell code snippets (WP-01, WP-02, and references/oauth-internals.md) designed to programmatically extract sensitive OAuth access and refresh tokens from the macOS Keychain using the security find-generic-password command. It also identifies and describes a plaintext credential fallback file located at ~/.claude/.credentials.json.
  • [PRIVILEGE_ESCALATION]: Multiple reference implementations (e.g., references/launchd-configuration.md) require the use of sudo to install and manage system-level launchd daemons within protected directories like /Library/LaunchDaemons/.
  • [PERSISTENCE]: The skill establishes persistence on the host machine by configuring a macOS launchd service that starts automatically at boot (RunAtLoad) and includes a KeepAlive configuration to ensure the process restarts if it exits. It also instructs the modification of shell initialization files (~/.zshenv) to maintain required environment variables across sessions.
  • [COMMAND_EXECUTION]: Working patterns and task templates (references/task-templates.md) rely on the execution of shell commands for credential retrieval, system configuration, and process management, creating a significant attack surface if input is not properly sanitized.
  • [DYNAMIC_EXECUTION]: The skill includes instructions and templates for compiling Go source code at runtime (go build) to create the proxy binaries used for request routing, which constitutes dynamic code generation and execution on the local system.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 7, 2026, 02:13 AM