The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions direct the agent to run npx -y opencode-openai-codex-auth@latest. This command downloads a package from the NPM registry and executes it immediately without verification. The package is provided by an unverified third-party author, which represents a high-risk remote code execution pattern.
[EXTERNAL_DOWNLOADS]: The skill provides instructions to download code from a personal GitHub repository (numman-ali/opencode-openai-codex-auth) for manual installation, which is not a verified or trusted source.
[PROMPT_INJECTION]: The skill metadata and description contain deceptive claims about providing access to "GPT-5.x" and "GPT-5.2 Codex" models. As these models have not been released by OpenAI, these claims constitute metadata poisoning intended to misrepresent the skill's capabilities and encourage the installation of potentially malicious software.
[DATA_EXFILTRATION]: The skill documentation highlights the storage and usage of sensitive authentication files located at ~/.opencode/auth/session.json and references the OPENAI_SESSION_TOKEN. The combination of targeting these credentials and executing unverified remote code creates a significant risk of session theft.

opencode-openai-codex-auth