The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes git commands (diff, log, rev-parse, ls-files) to collect repository data. It includes a specific security instruction to 'Pass each path as a separate quoted argument, never interpolate a joined shell string', which mitigates command injection vulnerabilities during branch or path resolution.- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from git logs, diffs, and project documentation (Ingestion points: git diff, git log, file content in SKILL.md). While explicit boundary markers for the data are not defined, the skill's focus on structural code smell detection and specific sanitization instructions (quoting arguments and avoiding shell interpolation) addresses the capability surface of git shell execution.- [DATA_EXPOSURE]: The skill accesses project-specific files such as GLOSSARY.md, CODING_STANDARDS.md, and directories like docs/ and .scratch/ to gather context for the review. This behavior is consistent with the intended function of a code review tool and remains restricted to the local repository scope.

code-review