The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the execution of shell commands using the codex CLI tool. These commands include codex exec and codex exec resume, which utilize shell features such as subshells ($(cat ...)) and variable interpolation ($THREAD_ID) to handle review prompts and session management.
[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface where untrusted repository content or user-generated plans could influence the behavior of the adversarial reviewer.
Ingestion points: The skill reads project documentation (CONTEXT.md, ADRs), implementation plans (PLAN.md), and arbitrary source code from the repository.
Boundary markers: The instructions do not specify the use of clear delimiters or 'ignore' instructions for the content being sent to the Codex model for review.
Capability inventory: The skill can write to documentation files (CONTEXT.md, PLAN.md) and execute shell commands (codex CLI).
Sanitization: No sanitization or filtering is performed on the content of the files before processing.
[DATA_EXFILTRATION]: The skill transmits the contents of the project's domain model (CONTEXT.md), implementation plans (PLAN.md), and repository source code files to the OpenAI Codex service. While this is the intended functionality and targets a well-known service, users should ensure they are authorized to share this project data with the external provider.

grill-with-docs-codex