The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill facilitates the handling of sensitive credentials by providing templates for ~/.codex/.env and auth.json (found in templates/ directory). It instructs the user to input their OPENAI_API_KEY into these local files while simultaneously configuring the system to transmit these credentials to an untrusted third-party proxy.
[DATA_EXFILTRATION]: The core configuration in SKILL.md and templates/config.toml sets the base_url to https://sub2api.yuepa8.com. This architectural change redirects all authenticated API traffic—including sensitive source code, internal prompts, and API keys—through a third-party server not controlled by the user or the official service provider, enabling silent data interception.
[EXTERNAL_DOWNLOADS]: The skill directs users to an unofficial API relay service (sub2api.yuepa8.com) and references configuration examples from an untrusted personal GitHub repository (github.com/silence2223/codex-switcher). These sources lack the reputation and security guarantees of official vendors.
[COMMAND_EXECUTION]: The skill provides instructions to execute shell commands (e.g., codex exec, codex doctor, and curl) that interact with the modified environment. If the codex binary itself is part of the unofficial ecosystem described, these commands could execute arbitrary logic on the host system.
[PROMPT_INJECTION]: The skill references non-existent models such as gpt-5.5 and gpt-5.4 in templates/config.toml. Using deceptive model naming can mislead the agent or user regarding the capabilities and safety filters of the underlying model being accessed through the proxy.

gomtm-codex-dev