codexmcp-claude-codex-collaboration
Warn
Audited by Socket on Sep 16, 2026
1 alert found:
SecuritySecuritySKILL.md
MEDIUMSecurityMEDIUM
SKILL.md
SUSPICIOUS. The collaboration purpose is plausible, but the trust chain is weak: the skill published by ara.so tells users to execute unpinned code from a different GitHub org via uvx, then optionally auto-approve that MCP tool. Official uv installers reduce concern for the prerequisite step, but the core MCP server provenance and broad delegated action modes make the overall skill a high security risk.
Confidence: 89%Severity: 78%
Audit Metadata