The Agent Skills Directory

[SAFE]: The skill implements a standard security auditing workflow based on the OWASP Top 10 framework. It uses mcp__codex__codex with restricted settings (sandbox: 'read-only', approval-policy: 'never') to ensure the agent does not perform unauthorized actions during the review process.
[COMMAND_EXECUTION]: The skill uses git diff, grep, and npm audit for local code analysis and dependency checking. These are standard development tools used for their intended purpose within the declared allowed-tools scope.
[DATA_EXPOSURE_AND_EXFILTRATION]: While the skill accesses sensitive code patterns (keywords like 'password', 'secret', 'token'), it does so locally to identify vulnerabilities. There is no evidence of these findings being transmitted to unauthorized external domains.
[INDIRECT_PROMPT_INJECTION]: As a code review tool, it processes untrusted local code. The skill mitigates this by using the read-only sandbox for Codex and instructing the agent to focus on specific security patterns rather than executing the code it analyzes.

security-review