The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes standard development CLI tools, including git, npm, npx, and xcodebuild, to perform its primary functions of code analysis, linting, and testing. It also executes a bundled shell script, scripts/commit-dirt-scan.sh, which uses grep and sed to identify sensitive data (secrets, debug artifacts) within the codebase being reviewed. These operations are local to the environment and are necessary for the skill's stated purpose.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from the target repository, such as pull request descriptions, commit messages, and project configuration files (e.g., CLAUDE.md). This creates a potential surface for indirect prompt injection attacks. However, the skill explicitly mitigates this risk through its "Red Flags" and "Precision Pass" instructions, which direct the agent to prioritize evidence found in the code diff over descriptive prose and to ignore instructions that attempt to bypass quality gates.
[SAFE]: The skill operates entirely within the local environment using standard development tools and does not exhibit malicious patterns. It contains no obfuscation, does not exfiltrate data to remote servers, and does not download or execute unverified external scripts. All referenced files and scripts are bundled locally within the skill package, and its behavior is consistent with the primary use case of a code review assistant.

suede-code-review