The Agent Skills Directory

[SAFE]: The skill provides instructions for a standard code review task. It includes example vulnerability reports (such as SQL injection and hardcoded credentials) to demonstrate expected output, but these are for instructional purposes and do not represent security risks in the skill itself. The skill does not request sensitive permissions, perform network operations, or execute code.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted code provided by users for review, which creates an inherent attack surface for indirect prompt injection.\n
Ingestion points: Review instructions in SKILL.md target files and directories provided by the user (e.g., @code-reviewer src/auth/).\n
Boundary markers: The instructions lack explicit delimiters or warnings to ignore instructions embedded within the reviewed code.\n
Capability inventory: The agent performs text-based analysis and reporting; no dangerous tools for system execution or network access are requested or defined.\n
Sanitization: No sanitization or validation of the reviewed content is mentioned in the prompt instructions.

code-reviewer