The Agent Skills Directory

[SAFE]: The skill consists entirely of Markdown-based documentation, templates, and examples. It provides best practices for structuring persistent objectives within the OpenAI Codex CLI environment.
[EXTERNAL_DOWNLOADS]: The documentation mentions official package installation commands (e.g., npm install -g @openai/codex@latest) for upgrading the Codex CLI tool itself. These refer to official OpenAI resources and are used in a purely informational context within the documentation.
[INDIRECT_PROMPT_INJECTION]: While the skill provides templates for interacting with an LLM via /goal commands, it does not ingest untrusted external data or provide a vector for indirect prompt injection. Its primary purpose is to help users manually craft more reliable and auditable instructions.
[COMMAND_EXECUTION]: The files contain examples of shell commands (e.g., codex --version, npm run bench:checkout) within Markdown code blocks. These are provided as usage examples for the user to understand the tool's behavior and are not executed by the skill itself.
[METADATA_POISONING]: The metadata in SKILL.md and metadata.json accurately reflects the content and purpose of the skill without containing any deceptive or malicious instructions.

codex-goals