The Agent Skills Directory

[SAFE]: The skill defines a standard development workflow for code review that emphasizes isolation. It recommends using a subagent or a fresh session to provide a 'clean boundary' for the review process, which is an effective security practice for limiting context contamination.
[INDIRECT_PROMPT_INJECTION]: The skill identifies three ingestion points for external data: the code diff, the requirement/spec, and test results. While these are untrusted data sources, the skill mitigates risk by requiring a 'fresh context' for analysis and a structured verification process (opening the real code to confirm findings) before any changes are applied. The severity is low as this is the intended functional purpose of a code review tool.

independent-review