The Agent Skills Directory

[SAFE]: The skill's behavior is entirely consistent with its stated purpose of code review. It provides detailed markdown templates and checklists for identifying security risks (OWASP Top 10) and performance issues (N+1 queries).
[COMMAND_EXECUTION]: The skill uses Read, Grep, and Glob tools. These are standard file system utilities required for an agent to analyze source code and are used here within the scope of code analysis.
[DATA_EXPOSURE]: While the skill accesses local project files, there are no network-capable tools listed in the allowed-tools configuration, and no instructions directing the agent to send data to external URLs or non-whitelisted domains.
[PROMPT_INJECTION]: The skill defines professional behavioral constraints (e.g., 'Forbidden Phrases' to avoid agreement theater), but these are pedagogical constraints derived from the open-source project 'superpowers' by Jesse Vincent. They do not attempt to override the underlying AI's safety guidelines or core instructions.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted data (pull request descriptions and code). However, the instructions implement a manual summarization 'checkpoint' as the first step of the workflow, which helps anchor the agent's context and mitigates the risk of the agent blindly following instructions embedded in the code it is reviewing.

code-reviewer