The Agent Skills Directory

[SAFE]: The skill functions as a static analysis tool with well-defined safety boundaries. It explicitly prohibits destructive operations, network exfiltration, and code execution, ensuring that its audit process remains passive and secure.
[COMMAND_EXECUTION]: The skill uses git commands like git diff to identify changes in the project. These are standard, read-only operations necessary for identifying the scope of the code review and do not introduce significant security risks.
[INDIRECT_PROMPT_INJECTION]: The skill handles untrusted data in the form of source code files, which is an inherent risk for indirect prompt injection attacks.
Ingestion points: Files identified through git diffs and specific file paths provided by the user.
Boundary markers: The instructions include a structured report format and specific rules to limit context reading to relevant diff hunks.
Capability inventory: Capabilities are restricted to file system reading and metadata-level shell commands. The Safety Boundary section strictly forbids network requests and code execution.
Sanitization: Audit findings are output as a human-readable report, and the skill does not interpret the code as instructions for the agent environment.

yy-frontend-review