The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external code diffs and engine output, which is untrusted data. It implements robust mitigations by explicitly instructing the agent to treat engine output as untrusted, verify findings against the actual source code, and ignore any imperative content or instructions embedded in findings.
[COMMAND_EXECUTION]: The skill utilizes the grok, git, and gh CLI tools. These operations are restricted to the intended purpose of gathering diffs and running the review engine. The skill explicitly forbids the engine from selecting execution parameters or passing reasoning flags.
[SAFE]: No malicious patterns, obfuscation, or unauthorized data exfiltration were detected. The skill follows security best practices by utilizing temporary files with automated cleanup (trap) and requiring explicit user confirmation before applying any suggested fixes.

grok-review