The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs users to install code from an external GitHub repository (BenedictKing/codex-review). This repository is not from a trusted organization or well-known service, posing a risk of downloading unverified content.
[REMOTE_CODE_EXECUTION]: The installation process requires the setup of an external 'Codex CLI'. Executing this CLI during the review process involves running unverified external code on the host system.
[COMMAND_EXECUTION]: The installation instructions utilize npx to execute code from the remote repository, which allows for arbitrary code execution during the setup phase.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it is designed to ingest and process untrusted project source code to generate reviews and changelogs.
Ingestion points: Project source files and repository metadata (SKILL.md).
Boundary markers: None identified; the instructions do not suggest using delimiters or specific 'ignore' commands to isolate external code from agent instructions.
Capability inventory: The skill performs file system reads across the project and writes to a CHANGELOG.md file.
Sanitization: There are no documented steps for sanitizing or escaping content within the code files before they are processed by the agent.

codex-review