The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell scripts and the external codex binary to facilitate its core functions, such as code review, technical questioning, and system diagnostics.
[EXTERNAL_DOWNLOADS]: The skill has a dependency on the external codex command-line tool, which is associated with a trusted technology vendor.
[PROMPT_INJECTION]: The skill utilizes prompt templates (e.g., templates/review-prompt.md) where user-provided task descriptions and repository contexts are interpolated. This creates a surface for indirect prompt injection where malicious content in a repository could influence the subagent's analysis.
Ingestion points: Git diffs and user-provided context files processed by scripts/codex-ask.sh.
Boundary markers: Templates use clear section headers (e.g., Task:, Context:) to delineate instructions from data.
Capability inventory: The skill can read local files and execute the codex binary with configurable sandbox settings.
Sanitization: scripts/codex-ask.sh includes logic to truncate context files that exceed 240 lines or 20,000 bytes to prevent context overflow or massive data ingestion.

codex-cli