The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md and examples in README.md demonstrate a pattern of piping content into a shell command: echo "[Content]" | codex exec. This interpolation lacks shell escaping for double quotes and other metacharacters, allowing for command injection if the processed content (such as code being reviewed or generated plans) contains malicious shell sequences.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted code for review and planning (Ingestion point: Phase 5 implementation review) without using boundary markers or sanitization. The agent possesses significant capabilities (Capability inventory: codex exec via shell, Edit, Write tools) that could be manipulated if malicious instructions are embedded in the analyzed codebases.

codex-claude-loop