The Agent Skills Directory

[PROMPT_INJECTION]: The instructions for the codex exec tool include explicit bypass markers such as 'take precedence over all skill directives' and 'Skip ALL skills'. These instructions are designed to override the operational constraints and safety guidelines of secondary agents or processes dispatched by the main agent.
[COMMAND_EXECUTION]: The skill guides the agent to execute shell commands using codex exec while interpolating user-provided test code into the command string. Because the <paste test code here> placeholder lacks boundary markers, escaping, or sanitization, it is vulnerable to command injection if the input contains shell-active characters like backticks or semicolons.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface in Phase 1.5. It ingests untrusted data (test specifications) and interpolates it directly into a prompt for another AI provider without sanitization or protective delimiters, allowing malicious code to potentially control the downstream model.

skill-tdd