The Agent Skills Directory

[SAFE]: The skill provides instructions for performing a security audit of code changes. It defines a structured process for identifying common vulnerability patterns such as SQL injection, command injection, and cross-site scripting (XSS).
[COMMAND_EXECUTION]: The skill utilizes shell commands through the Bash tool, but the scope is restricted to specific, read-only git operations (git diff, git log, git status, git show) required to identify the files changed in a branch.
[DATA_EXFILTRATION]: While the WebFetch tool is included in the allowed-tools metadata, the instructions do not contain any directives to fetch data from or exfiltrate data to external domains. The primary operation remains focused on local file analysis.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves reading and processing untrusted data from a code repository, which presents an inherent surface for indirect prompt injection if a reviewed file contains instructions for the agent.
Ingestion points: Files in the local repository identified via git commands, and the project documentation file CLAUDE.md.
Boundary markers: No explicit delimiters or instructions are provided to the agent to disregard instructions found within the files it is reviewing.
Capability inventory: File system access (Read, Grep, Glob), restricted shell access (Bash for git), and network access (WebFetch).
Sanitization: There is no evidence of sanitization or filtering of the file content before it is processed by the agent.

security-review