claude-glm-agent
Fail
Audited by Gen Agent Trust Hub on Jun 15, 2026
Risk Level: HIGHDATA_EXFILTRATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFE
Full Analysis
- [DATA_EXFILTRATION]: The skill redirects sensitive AI traffic to an external third-party domain.
- It configures "ANTHROPIC_BASE_URL" to "https://api.z.ai/api/anthropic", routing all prompt and response data through this proxy.
- Authentication tokens ("ANTHROPIC_AUTH_TOKEN") are sent to this third-party endpoint for every API request.
- [COMMAND_EXECUTION]: Automated scripts perform sensitive file system operations.
- "configure-glm.sh" and "migrate-env.sh" search for and modify ".env" files to replace official API keys with proxy tokens.
- "init-glm.sh" copies and modifies scripts within the project directory based on auto-detected project types.
- [EXTERNAL_DOWNLOADS]: Outbound network communication with a non-standard domain.
- "test-proxy.sh" uses "curl" to transmit tokens to "api.z.ai" to verify connectivity.
- [CREDENTIALS_UNSAFE]: The skill includes code patterns that may lead to credential exposure.
- "references/client-config.md" provides an example using "dangerouslyAllowBrowser: true", which allows API keys to be used in client-side code, increasing the risk of exposure.
Recommendations
- AI detected serious security threats
Audit Metadata