The Agent Skills Directory

[Automated Network Access]: The skill includes instructions to request network approval and proceed without additional prompting to maintain workflow efficiency. This behavior is intended to streamline issue diagnosis and relies on the platform's tool-based approval mechanisms.
[Indirect Prompt Injection Surface]: By ingesting data from external GitHub issues and comments, the skill introduces a surface for indirect instructions. This is a common aspect of processing third-party reports and is handled by the agent's core safety protocols.
[Vendor-Specific Data Access]: The workflow involves accessing repository data and logs from the openai/codex project. As these are vendor-owned resources, the activity is consistent with the skill's stated purpose of bug diagnosis.

codex-bug