The Agent Skills Directory

[SAFE]: No malicious patterns or security violations were detected in the skill instructions or reference materials. The skill's behavior is consistent with its stated purpose of code review.
[PROMPT_INJECTION]: The skill operates on untrusted external data (source code diffs), which presents a surface for indirect prompt injection. This is mitigated by the restricted scope of the skill's capabilities.
Ingestion points: Diff hunks and source code files provided for review in SKILL.md.
Boundary markers: The skill does not define specific delimiters for untrusted content, though it enforces a strict markdown table output format.
Capability inventory: Access is limited to read-only operations via grep and text generation. No file system writing, network operations, or shell execution capabilities are present.
Sanitization: Input data is not sanitized, but the agent is directed to use specific thresholds and tokens for evaluation.
[COMMAND_EXECUTION]: The instructions direct the agent to use grep to inspect project files such as package.json and design token files. This is a legitimate functional requirement for context-aware code review and does not involve unsafe command construction or privilege escalation.

ui-review