The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a shell script scripts/audit-codex-init.sh that is intended to be run by the agent. This script executes local commands including codex doctor, codex --version, and git status to gather information about the current environment and project state. These operations are scoped to auditing and do not perform unauthorized modifications.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process instructions from project-level files such as AGENTS.md and CLAUDE.md. This represents an indirect prompt injection surface as these files could potentially contain instructions from untrusted external sources (e.g., in a downloaded repository). However, the skill instructions emphasize auditing and reporting findings rather than blind obedience.
Ingestion points: The agent is instructed to read AGENTS.md and CLAUDE.md in the project root.
Boundary markers: The skill suggests summarizing and reporting findings before making changes.
Capability inventory: The skill can execute CLI commands and write to local configuration files after user approval.
Sanitization: The skill relies on user review of the audit output.
[EXTERNAL_DOWNLOADS]: The skill references external documentation from OpenAI's official domain (openai.com) and a well-known technical blog (simonwillison.net). These references are for informational purposes only and do not involve the download or execution of remote code.

laohan-chushicodex