claude-code-proxy-patterns
Audited by Socket on Oct 22, 2026
2 alerts found:
SecurityAnomalySUSPICIOUS. The skill’s proxying purpose broadly matches its capabilities, but it requires direct access to Claude Code OAuth secrets, discusses plaintext token fallbacks, enables persistent privileged local interception, and forwards traffic to arbitrary third-party providers. The main concern is not obvious malware but disproportionate credential handling and undocumented trust/provenance for the core proxy implementation.
This fragment describes retrieving and caching Claude Code OAuth credentials from Keychain, with a documented plaintext-file fallback, for a localhost proxy that sends authenticated requests to `api[.]anthropic[.]com`. Its stated rationale is to continue proxy use despite restrictions on third-party OAuth-token use, creating a significant credential-handling and authorization-policy risk; the excerpt does not itself show token exfiltration to an unrelated destination or other clear malware behavior. Severity is moderate to high because access tokens are sensitive and the file fallback is plaintext; the full implementation is not visible.