codexisland-macos-notch

Warn

Audited by Gen Agent Trust Hub on Sep 16, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONCREDENTIALS_UNSAFEDATA_EXFILTRATION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the user/agent to download and install software from an untrusted third-party GitHub repository (ericjypark/codex-island) and a personal Homebrew tap.
  • [COMMAND_EXECUTION]: Provides explicit instructions to bypass macOS Gatekeeper security protections using the xattr -dr com.apple.quarantine command on an unsigned binary. This removes the security attribute that triggers malware scanning and developer verification for downloaded files.
  • [CREDENTIALS_UNSAFE]: The skill's primary function involves reading sensitive authentication data from several locations, including the macOS Keychain (Claude Code-credentials), environment variables (CLAUDE_CODE_OAUTH_TOKEN), and local configuration files containing plain-text tokens (~/.codex/auth.json).
  • [DATA_EXFILTRATION]: The skill accesses and processes local session logs (.jsonl files) from ~/.claude/projects/ and ~/.config/claude/. While the skill claims local-only processing, the automated ingestion of chat/code history logs by a third-party utility creates a significant data exposure risk.
  • [DYNAMIC_EXECUTION]: The skill encourages building from source using a provided build.sh script from an external repository, which executes arbitrary build commands and compiles Swift code on the local machine.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 16, 2026, 03:57 AM