The Agent Skills Directory

[SAFE]: The skill defines a structured workflow for design review, involving plan preparation, sub-agent invocation, and iterative feedback application. All operations are confined to the intended design review task.
[DATA_EXPOSURE]: The skill accesses local files (design plans and related source code) to provide context to the Codex sub-agent. This file access is legitimate and necessary for the described functionality.
[PROMPT_INJECTION]: The prompt templates provided for the sub-agent focus on engineering review criteria (architecture, performance, safety, etc.) and do not contain instructions to bypass safety filters or override agent behavior.
[INDIRECT_PROMPT_INJECTION]: The skill processes external plan and source files, which are potential vectors for indirect prompt injection. However, the skill lacks high-risk capabilities such as arbitrary shell execution or network exfiltration, which significantly mitigates the potential impact of such an attack.

plan-loop